JP7792940B2 - Information processing device, information processing method, and information processing program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and an information processing program.

Traditionally, creators of content such as web content have attempted to create content that would give users the impression they are aiming for, but there is a possibility that the impression users actually have of the content may be different.

Patent Document 1 discloses a technology in which evaluators are asked to input the rating level of their impressions of content they viewed, and pairs of content and the rating level of the evaluator's impressions of that content are used as training data for training a machine learning model.

Japanese Patent Application Laid-Open No. 2019-133455

However, the above-mentioned conventional technology requires the evaluator to input the impression evaluation level, which poses a problem of high workload.

The present application has been made in consideration of the above, and aims to provide an information processing device, information processing method, and information processing program that can reduce the workload required to estimate a user's impression of content.

The information processing device according to the present application includes a receiving unit, a generating unit, and a providing unit. The receiving unit receives a selection of target content that is the subject of estimation of the impression a user has or has had, and a selection of an impression group, which is a group of impressions. The generating unit uses a generation AI to generate impression content that indicates the impression the user has or is estimated to have had of the target content from among the multiple impressions included in the impression group received by the receiving unit. The providing unit provides the impression content generated by the generating unit.

One aspect of this embodiment has the advantage of reducing the workload required to estimate a user's impression of content.

FIG. 1 is a diagram for explaining information processing according to an embodiment. FIG. 2 is a diagram illustrating an example of the configuration of an information processing system according to the embodiment. FIG. 3 is a diagram illustrating an example of the configuration of the information processing apparatus according to the embodiment. FIG. 4 is a diagram illustrating an example of a user information table stored in the user information storage unit of the information processing apparatus according to the embodiment. FIG. 5 is a diagram illustrating an example of input information input for generation by the estimation processing unit in the processing unit of the information processing device according to the embodiment. FIG. 6 is a diagram showing an example of impression content generated by a second generation processing unit in the processing unit of the information processing device according to the embodiment. FIG. 7 is a flowchart illustrating an example of information processing by the processing unit of the information processing apparatus according to the embodiment. FIG. 8 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the information processing apparatus according to the embodiment.

Below, detailed explanations will be given of the information processing device, information processing method, and information processing program (hereinafter referred to as "embodiments") according to the present application, with reference to the drawings. Note that the information processing device, information processing method, and information processing program according to the present application are not limited to these embodiments. Furthermore, the embodiments can be combined as appropriate to the extent that the processing content is not contradictory. Furthermore, the same components in the following embodiments will be assigned the same reference numerals, and redundant explanations will be omitted.

[1. An example of information processing]
First, an example of information processing according to the embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram for explaining information processing according to the embodiment.

The information processing device 1 shown in FIG. 1 is an information processing device that generates impression content, which is content that indicates the impression that a user U has or is presumed to have had about target content, and is realized, for example, by one or more servers or a cloud system. The user U is a user of the terminal device 2 and a user of the online service for which the target content is provided.

As shown in FIG. 1, the information processing device 1 receives an impression content generation request transmitted from the terminal device 3 of a worker O who creates or provides the target content (step S1). The impression content generation request includes the target content that is the subject of estimation of the impression that the user U has or has had, information indicating an impression group that is a group of impressions, information specifying the estimation method, information indicating the target content type, and information indicating the impression content type.

The target content, impression group, estimation method, target content type, and impression content type specified in the impression content generation request are, for example, the target content, impression group, estimation method, target content type, and impression content type selected by worker O.

By accepting a request to generate impression content, the information processing device 1 accepts selections for target content, impression group, estimation method, target content type, and impression content type. Note that worker O is an employee of the company that creates or provides the target content, but may also be an employee of a company commissioned by such company.

Target content is web content, such as, but not limited to, web pages, catch phrases, creative content, and stamps. Target content is content that is a candidate for or is to be provided on, for example, electronic commerce (EC) sites, news sites, restaurant introduction sites, image posting sites, video viewing sites, etc., and includes, but is not limited to, advertising content. EC sites include, for example, shopping sites, auction sites, etc., but are not limited to, these examples.

The target content is content that can be provided on an online site provided by the information processing device 4, and the user U can use the online site provided by the information processing device 4 by operating the terminal device 2. Online sites provided by the information processing device 4 include, but are not limited to, e-commerce sites, news sites, store introduction sites, image posting sites, and video viewing sites.

On the EC site, the information processing device 4 accepts product content such as product descriptions from workers O and others as posted content, and provides such posted content to each user U. Product descriptions can also be referred to as product descriptions. Furthermore, posted content on the EC site includes not only product content, but also comments and ratings posted about the product.

In addition, the information processing device 4 accepts news articles from workers O and the like as posted content on the news site, and provides such posted content to each user U. On the news site, posted content includes not only news articles but also comments posted on such news articles.

In addition, the information processing device 4 accepts store content, such as a store introduction text from the worker O, as posted content on the store introduction site, and provides such posted content to each user U. In addition to store content, the posted content on the store introduction site also includes reviews and comments posted about the store. Store introduction sites are, for example, sites for each type of store, such as a restaurant introduction site.

In addition, the information processing device 4 accepts image content (e.g., still image content and video content) from workers O and others as posted content on an image posting site, and provides such posted content to each user U. In addition, the information processing device 4 accepts video content and titles and caption text for video content as posted content on a video viewing site, and provides such posted content to each user U.

The impression group specified in the impression content generation request is an impression group specified by user U from among multiple impression groups C1 to Cn (n is an integer equal to or greater than 2). The multiple impression groups C1 to Cn may be, for example, impression groups categorized by user U's values, impression groups categorized by user U's preferences, impression groups categorized by user U's lifestyle, etc., but are not limited to such examples.

In the example shown in Figure 1, the group of impressions based on user U's values is a group of impressions that classify values based on Schwartz's value theory as impressions that user U has or has had, and is classified into 10 classification criteria: power, achievement, hedonism, stimulation, self-determination, universalism, philanthropy, tradition, harmony, and security. In the example shown in Figure 1, the group of impressions based on values based on Schwartz's value theory is further subdivided into multiple classification criteria, and the impressions that user U has or has had are classified into 56 classification criteria.

Specifically, the classification criterion "power" is subdivided into five classification criteria: "social recognition," "wealth," "authority," "social power," and "saving face," the classification criterion "achievement" is subdivided into five classification criteria: "intelligence," "competence," "success," "ambition," and "influence," the classification criterion "hedonism" is subdivided into two classification criteria: "enjoyment of life" and "pleasure," and the classification criterion "stimulation" is subdivided into three classification criteria: "courage," "a varied life," and "a vibrant life."

Furthermore, the classification criterion "self-determination" is further subdivided into six classification criteria: "freedom," "independence," "curiosity," "creativity," "self-selection," and "self-respect," the classification criterion "universalism" is further subdivided into nine classification criteria: "wisdom," "world peace," "world of beauty," "social justice," "inner harmony," "environmental protection," "equality," "generosity," and "harmony with nature," and the classification criterion "philanthropy" is further subdivided into nine classification criteria: "true friendship," "meaning of life," "responsibility," "loyalty," "mature love," "help," "honesty," "tolerance," and "spiritual world."

Furthermore, the classification criterion "harmony" is further subdivided into four classification criteria: "politeness," "self-discipline," "respect for parents and elders," and "obedience." The classification criterion "tradition" is further subdivided into six classification criteria: "humility," "detachment," "respect for tradition," "piety," "moderation," and "acceptance of fate." The classification criterion "security" is further subdivided into seven classification criteria: "health," "family safety," "social order," "cleanliness," "giving back," "sense of belonging," and "national security."

User U's values are not limited to the examples mentioned above, and may be values categorized as traditionalism, success orientation, self-actualization orientation, coexistence orientation, etc., or other values. For example, if the object is a car, user U's values may be values categorized as prioritizing safety, livability, design, driving performance, etc., and if the object is a house, they may be values categorized as prioritizing location, price, layout, design, future prospects, etc.

Impression groups based on user U's lifestyle include, for example, impression groups based on the AIO (Activities, Interest, Opinions) approach and impression groups based on VALS (Values And LifeStyles). Impression groups based on the AIO approach are impression groups classified into one or more impression groups of activities, interests, opinions, etc. Impression groups based on VALS are impression groups classified into, for example, self-actualizers, successful people, group members, those who maintain their livelihoods, those in poverty, those who aspire to success, those with social conscience, intellectuals, and young intellectuals.

The impression group may also be, for example, the impression group indicated by innovator theory, which classifies the process of diffusion of new products. This impression group is classified into innovators, early adopters, early majority, late majority, and laggards.

There are multiple estimation methods for estimating the impression that user U will have or has had of target content, including a first estimation method and a second estimation method. The first estimation method is a method in which a generation AI (Artificial Intelligence) directly estimates the impression that user U will have or has had of target content. The second estimation method is a method in which, for each impression included in an impression group, the generation AI generates reference content that is estimated to have the impression that user U will have, and compares this reference content with the target content to estimate the impression that user U will have or has had of the target content.

The target content type is, for example, the type of target content, and is indicated, for example, by a combination of an online site and a content type. For example, the target content type may be, but is not limited to, a product description on a product page on an e-commerce site, a news article on a news site, a comment on a news article on a news site, image content on an image posting site, or the title and caption text of video content on a video viewing site.

The impression content type is a type of impression content, and includes, for example, a first content type and a second content type. The first content type is a type of impression content that includes information indicating the impression that user U is presumed to have of the target content, and is a type of impression content that supports the creation and consideration of the target content. The second content type is a type of impression content that includes information indicating the impression that user U is presumed to have of the target content, and is a type of impression content that supports understanding the impression that user U has of the target content.

Next, the information processing device 1 uses the generation AI to estimate the impression that the user U has or is estimated to have had of the target content selected in step S1 from among the multiple impressions included in the selected impression group, which is the group of impressions selected in step S1 (step S2).

The generation AI used in step S2 may be, for example, a text generation AI, an image generation AI, or a multimodal AI. Such generation AI may be located, for example, in an external information processing device, and the information processing device 1 may have the generation AI generate and acquire reference content for each impression via an API (Application Programming Interface) provided by the external information processing device, but this is not a limitation. For example, the generation AI may be located within the information processing device 1.

Text generation AI is, for example, a language model trained to estimate and output the next token from an input token sequence, such as a transfer-based model or an RNN (Recurrent Neural Network)-based model.

Examples of transfer-based models include, but are not limited to, GPT (Generative Pre-trained Transformer) and BARD (Bidirectional Auto Regressive Dialogues). Examples of RNN-based models include, but are not limited to, RWKV (Receptance Weighted Key Value). It is desirable to keep input information, such as personal information, confidential by training the model in a way that prevents it from being used as a new answer.

Examples of image generation AI include, but are not limited to, StackGAN (Generative Adversarial Networks), AttnGAN, T2I (Text-to-Image) with Transformers, and DALL-E. Examples of multimodal AI include, but are not limited to, models that generate images from text or generate text from images, such as GPT-4V and CM3Leon (Chameleon Multimodal Model).

In step S2, the information processing device 1 uses a generation AI to estimate the impression that the user U has of the target content, or the impression that the user U has had, from among the multiple impressions included in the selected impression group. Specific explanations are provided below. In the following explanations, the multiple impressions included in the selected impression group are described as multiple impressions IM1 to IMm (m is an integer of 2 or greater). Furthermore, when referring to each of the multiple impressions IM1 to IMm without distinguishing them individually, they may be referred to as impressions IM.

First, we will explain the case where the estimation method and impression content type selected and accepted by the information processing device 1 in step S1 are the first estimation method and the first content type. In this case, the information processing device 1 causes the generation AI to estimate the impression IM that the user U has of the target content from among multiple impressions IM1 to IMm included in the selected impression group.

For example, the information processing device 1 inputs to the generation AI as input information information including instruction information instructing the user U to select the impression IM that the user U has of the target content from among the multiple impressions IM1 to IMm indicated in the impression information, impression information indicating the multiple impressions IM1 to IMm included in the selected impression group, and the target content, and causes the generation AI to output information indicating one or more impressions IM that the user U is estimated to have of the target content.

For example, suppose the selected impression group is a group of impressions based on the values of user U, and the target content is a catchphrase, "You can tell the difference because it's something you use every day." In this case, the information processing device 1 includes, for example, the following information in the input information as instruction information: "Please select three or more images from the values below that fit the catchphrase, 'You can tell the difference because it's something you use every day.' The output format should be in CSV format as shown below. \nValue, Reason."

In addition, the information processing device 1 includes information containing the following character string as impression information in the input information: "social power, authority, wealth, saving face, social recognition\n success, competence, ambition, influence, intelligence\n fun, enjoying life\n courage, a varied life, a vibrant life\n creativity, curiosity, freedom, choosing goals, self-respect, independence\n environmental protection, the world of beauty, harmony with nature, generosity, social justice, wisdom, equality, world peace, inner harmony\n assistance, honesty, tolerance, loyalty, responsibility, true friendship, the spiritual world, mature love, the meaning of life\n politeness, respect for parents and elders, self-discipline, obedience\n piety, acceptance of fate, humility, moderation, respect for tradition, detachment\n cleanliness, national security, social order, family security, giving back, health, a sense of belonging."

In addition, instead of the information string "Please select three or more of the values below that fit the catchphrase," the information processing device 1 can also use information such as "Please select three or more of the values below that represent the impression that users are likely to have of the catchphrase," or information such as "Please estimate a score indicating the degree of impression that users are likely to have of the catchphrase for each of the values below. The score should range from 1 to 10, with a higher value representing a higher degree of impression," but is not limited to such examples. Furthermore, the output format is not limited to CSV format.

The instruction information also includes text information corresponding to the target content type selected in step S1. In the example above, the target content type is a catchphrase, so the instruction information includes the string "catchphrase." However, if the target content type is a product description on a product page on an e-commerce site, the instruction information includes the string "product description."

In this way, the information processing device 1 generates instruction information that corresponds to the type of target content by including text information that corresponds to the type of target content in the instruction information, but it is also possible to input instruction information that is independent of the type of target content to the generation AI.

For example, instead of the string "Please select three or more images from the values below that fit the catchphrase 'You'll know the difference because it's something you use every day,'" the information processing device 1 can set instruction information containing the string "Please select three or more images from the values below that fit the information you are entering." as a system message. A system message is a message that instructs the behavior of the generating AI, and is, for example, a message that is set in the input information as the content of "'role':'system'" in OpenAI's GPT model.

In addition, the information processing device 1 can also use the generation AI to estimate the impression that user U has of the target content from multiple impressions IM1-IMm included in the selected impression group by limiting the attributes of user U, etc. In this case, instead of the information string "Please select three or more images from the values below that fit the catchphrase," the information processing device 1 can use information such as "Please select three or more impressions from the values below that are estimated to be formed by a man in his twenties for the catchphrase." or the information string "Please estimate, for each of the values below, a score indicating the degree of impression that a man in his twenties is estimated to have for the catchphrase. The score should range from 1 to 10, with a higher value indicating a higher degree of impression." When having the generation AI estimate one or more impressions IM formed by a user U with an attribute other than a man in his twenties, it is possible to replace the information in the string "man in his twenties" with information indicating another attribute.

Furthermore, when the target content is image content, the information processing device 1 can input information including the target content and instruction information containing information such as the string "Please select three or more images from the values below that fit the input image," as input information to the multimodal AI, and have the multimodal AI output a similarity. Note that, as in the case when the target content is other than image content, the information processing device 1 can limit the attributes of the user U and output a score indicating the degree of impression IM.

Next, we will explain the case where the estimation method and impression content type selected and accepted by the information processing device 1 in step S1 are the first estimation method and the second content type. In this case, the information processing device 1 collects behavioral content, which is content that indicates the behavior of user U related to the target content, from the information processing device 4, etc.

Behavioral content is content that arises from actions by user U that are related to the target content, such as posted content such as reviews, comments, and messages posted by user U regarding the target content. For example, if the target content is a product description or product image, posted content as behavioral content is a review or comment posted about the product introduced in the target content.

In addition, if the target content is a store description or an image of the store interior, the posted content as behavioral content is a review or comment posted about the store introduced by the target content.In addition, if the target content is video content, the posted content as behavioral content is a review or comment posted about the video content.

In addition, posted content as behavioral content may be content such as conversations or posts by user U on the official account of the product or store indicated by the target content in the communication application.In addition, posted content as behavioral content may be content such as posts or conversations by user U on an online bulletin board site, for example.

For example, the information processing device 1 inputs to the generation AI as input information information including instruction information instructing the user U to select the impression IM formed by the behavioral content from among the multiple impressions IM1-IMm indicated in the impression information, impression information indicating the multiple impressions IM1-IMm included in the selected impression group, and the behavioral content, and causes the generation AI to output information indicating the impression IM that the user U is estimated to have formed of the target content.

For example, if the selected impression group is a group of impressions based on user U's values, the behavioral content is represented by the character string "XXX...XXX". In this case, the information processing device 1 generates instruction information including the information on the character string "'XXX...XXX' is a comment posted by the user regarding the target content below. Please select three or more of the values below that represent the image that user U is presumed to have of the target content below. Please output in CSV format as shown below. \nValue, reason", impression information, and target content.

In addition, the information processing device 1 can use, for example, the following information instead of the string "Please select three or more of the values below that represent the impression that user U has of the target content.": "Please estimate a score indicating the degree of impression that user U is estimated to have of the target content for each of the values below. The score should range from 1 to 10, with a higher value representing a higher degree of impression." However, the information processing device is not limited to this example.

The output format is not limited to CSV. Instead of the target content, the instruction information may include a string of characters indicating the product, store, etc. identified by the target content. In this case, this string of characters is used in place of the above-mentioned string "target content below." Part of the above-mentioned instruction information can be set as a system message.

The information processing device 1 can generate behavioral content that includes multiple pieces of user U's content in the instruction information regardless of the attributes of the user U, or can generate behavioral content that includes multiple pieces of user U's content for each attribute of the user U in the instruction information. When the information processing device 1 includes multiple pieces of user U's content for each attribute of the user U in the instruction information, it can cause the generation AI to output information indicating one or more impressions IM that are estimated to have been held by the user U for each attribute of the user U.

Furthermore, when the target content is an image, the information processing device 1 can input information containing instruction information, such as "The following image is an image posted by a user regarding the following target content," to the multimodal AI as input information, instead of the information containing the string "'XXX...XXX' is a comment posted by a user regarding the following target content," and cause the multimodal AI to output information indicating one or more impressions IM that are estimated to have been held by the user U.

Furthermore, when the behavioral content includes text and images, the information processing device 1 can include the text and images as behavioral content in the instruction information, input information including such instruction information to the multimodal AI as input information, and output it from the multimodal AI.

Next, we will explain the case where the estimation method and impression content type selected and accepted by the information processing device 1 in step S1 are the second estimation method and the first content type. In this case, for each impression IM included in the selected impression group, the information processing device 1 causes the generation AI to generate reference content that is estimated to have the impression IM held by the user U, and compares the reference content with the target content to estimate one or more impressions IM held by the user U regarding the target content.

For each impression IM classified into multiple categories in the selected impression group, the information processing device 1 generates reference content that is estimated to give the impression that the user U will have. If the selected impression group is a group of impressions based on values based on Schwartz's value theory, the multiple impressions IM classified in the selected impression group are the 10 types of values or 56 types of values described above.

For example, the information processing device 1 inputs input information to the generation AI, including instruction information that instructs the generation of reference content that is estimated to give the user U the impression indicated by the impression IM, and causes the generation AI to generate the reference content. The reference content is, for example, content represented by at least one of text and an image.

The information processing device 1 stores standard instruction information in advance, and inputs information including the standard instruction information and impression information to the generation AI as instruction information, causing the generation AI to generate reference content for each impression IM. For example, if the target content is a product description, the standard instruction information is the string "I will post certain values. Please write 10 product descriptions that are likely to share those values with Japanese users. Please do not include too many of the values you have posted directly in the sentences."

Furthermore, the impression information is, for example, information on the character string "Values: Social Power." As a result, each of the 10 product descriptions that are predicted to give the user U the impression IM of social power is generated as reference content.

When the generating AI is a GPT, the standard instruction information is included in the input information (prompt) as an instruction, and the impression information is included in the input information (prompt) as a user message, but this example is not limited to this. For example, instead of the string "Post a certain value. That value" in the standard instruction information, the input information may include information containing the string "Social power as a value" as a user message.

After generating reference content for each impression IM, the information processing device 1 compares the target content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM. In order to compare the target content with the reference content for each impression IM, the information processing device 1 vectorizes the target content and each reference content.

Content vectorization is performed, for example, by embedding using a language model (e.g., a transformer-based model). The vectorized content is represented, for example, as a vector with several hundred dimensions, but is not limited to such an example.

Examples of embedding using a language model include text-embedding-ada and BERT (Bidirectional Encoder Representations from Transformers) provided by OpenAI (registered trademark), but are not limited to these examples.

Note that content vectorization is not limited to embedding using a language model; content vectorization may also be performed using, for example, Doc2Vec or the average of word embeddings. Word2Vec, fastText, etc. are used for word embedding.

The information processing device 1 classifies the target content into an impression IM corresponding to a reference content whose vector similarity with the target content is equal to or greater than a threshold. The information processing device 1 can classify the target content into only one impression IM, or into two or more impression IMs, for example.

For example, the information processing device 1 can classify the target content into only one impression IM by classifying the target content into an impression IM that corresponds to the reference content with the highest similarity, where the vector similarity is equal to or greater than a threshold value.

Furthermore, when there are two or more reference contents whose vector similarity with the target content is equal to or greater than a threshold, the information processing device 1 can classify the target content into two or more impressions IM corresponding to these two or more reference contents, respectively.

The vector similarity is cosine similarity, but it may also be Jaccard similarity, or the inverse of Euclidean distance or Manhattan distance. Furthermore, when Euclidean distance or Manhattan distance is used, the information processing device 1 classifies, for example, target content whose Euclidean distance or Manhattan distance from the reference content is less than a threshold value into an impression IM corresponding to the reference content.

In addition, the information processing device 1 can use any method to determine the similarity between vectors or classify vectors, and ultimately the similarity between content items or classify content items, such as by learning a classification model using vectors and the corresponding hand-labeled data and then using that model for classification.

Furthermore, when the reference content and target content are images, the information processing device 1 can input information including the reference content, the target content, and instruction information indicating an instruction to output the similarity between the reference content and the target content to the multimodal AI as input information, and cause the multimodal AI to output the similarity.

In this case, the information processing device 1 classifies the impression IM into, for example, an impression IM generated by multimodal AI that corresponds to the reference content with the highest similarity, which is equal to or greater than a threshold, or an impression IM generated by multimodal AI that corresponds to the reference content with a similarity equal to or greater than a threshold.

Furthermore, when the reference content and target content are images, the information processing device 1 can also cause the multimodal AI to generate explanatory text that explains what images are contained in each of the reference content and target content.

In this case, the information processing device 1 vectorizes each of the explanatory text of the reference content and the explanatory text of the target content, and calculates the similarity between the vectorized explanatory text of the reference content and the vectorized explanatory text of the target content. The method of calculating the similarity and the group of impressions of the target content's impression IM are the same as when the reference content and the target content are text.

Note that when the reference content and the target content are images, the information processing device 1 can also directly vectorize each of the reference content and the target content to calculate the similarity.

Furthermore, when the reference content and the target content each contain text and images, the information processing device 1 classifies the impressions into impression IMs corresponding to reference content whose integrated score, which is a score obtained by weighting the similarity between the text and the similarity between the images, is equal to or greater than a threshold, and impression IMs corresponding to reference content whose integrated score is equal to or greater than a threshold and is the largest.

Next, we will explain the case where the estimation method and impression content type selected and accepted by the information processing device 1 in step S1 are the second estimation method and the second content type. In this case, the information processing device 1 collects behavioral content, which is content that indicates the behavior of user U related to the target content, from the information processing device 4, etc.

The behavioral content is content that arises from actions by user U that are related to the target content, and, as in the case where the estimation method selected and accepted in step S1 is the first estimation method, is posted content such as reviews, comments, and messages posted by user U regarding the target content.

For each impression IM included in the selected impression group, the information processing device 1 causes the generation AI to generate reference content that is estimated to have been the impression IM held by the user U, and compares the reference content with the behavioral content to estimate one or more impressions IM held by the user U regarding the target content.

For each impression IM classified into multiple categories in the selected impression group, the information processing device 1 generates reference content that is presumed to have been the impression that the user U had. For example, the information processing device 1 inputs input information to the generation AI that includes instruction information that instructs the generation of reference content that is content that is presumed to have been the impression that the user U had indicated in the impression IM and corresponds to the behavioral content, and causes the generation AI to generate the reference content. The reference content is, for example, content that is displayed as at least one of text and an image.

The information processing device 1 stores first standard instruction information in advance, and inputs information including the standard instruction information and impression information to the generation AI as instruction information, causing the generation AI to generate reference content for each impression IM. For example, if the target content is a product description, the standard instruction information is the following string: "I will post certain values. Please write 10 comments by users who are presumed to share those values with Japanese users. Please do not include too many of the values you posted directly in the sentences."

The impression information is, for example, the string "Values: Social Power." As a result, each of the 10 comments from user U that are estimated to have given user U the impression IM of social power is generated as reference content.

After generating reference content for each impression IM, the information processing device 1 compares the impression IM held by the user U or the behavioral content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM using the same comparison method as when the estimation method selected and accepted in step S1 is the first estimation method.

For example, the information processing device 1 compares, for each piece of dynamic content, the behavioral content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM, and estimates one or more impressions IM held by the user U regarding the target content based on the comparison results.

The information processing device 1 compares, for each behavioral content, the behavioral content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM. In order to compare the behavioral content with the reference content for each impression IM, the information processing device 1 vectorizes each behavioral content and each reference content.

For each piece of behavioral content, the information processing device 1 performs a process of vectorizing the behavioral content and calculating the similarity between the vector and each reference content. Then, the information processing device 1 estimates, for example, the impressions IM corresponding to the reference content whose average vector similarity with each behavioral content is equal to or greater than a threshold, as one or more impressions IM that the user U has of the target content.

Furthermore, the information processing device 1 estimates, for example, impressions IM corresponding to reference content whose median value of vector similarity with each behavioral content is equal to or greater than a threshold value as one or more impressions IM that the user U has of the target content.

Next, based on the estimation results described above, the information processing device 1 generates, as impression content, content including information indicating one or more impressions IM into which the target content has been classified (step S3). For example, the information processing device 1 generates, as impression content, content that maps multiple impressions IM1-IMm classified into specific impression groups and that includes information indicating impressions IM that the user U has or is estimated to have had of the target content.

Next, the information processing device 1 provides the impression content generated in step S3 to the worker O (step S4). In step S4, the information processing device 1 provides the impression content generated in step S3 to the worker O, for example, by transmitting the impression content generated in step S3 to the terminal device 3 of the worker O.

The impression content shown in Figure 1 is content in which multiple values based on Schwartz's value theory are mapped as multiple impressions IM1 to IMm (m = 56 in Figure 1), and impressions IM classified as impressions IM that user U has or has had based on values based on Schwartz's value theory are highlighted. In the example shown in Figure 1, content showing introductory text or introductory images of four different models of a certain car model is shown as the target content.

In the impression content shown in Figure 1, model A automobiles are classified into the impressions "freedom," "independence," and "self-respect," which are subdivisions of the impression "self-determination," and into the impression "inner harmony," which is a subdivision of the impression "universalism."

Furthermore, in the impression content shown in Figure 1, model B automobiles are classified into the impressions "Variety of life" and "Vibrant life" which are subdivisions of the impression "Stimulation," the impressions "Enjoyment of life" and "Fun" which are subdivisions of the impression "Hedonism," and the impression "Belonging" which is subdivisions of the impression "Safety."

Furthermore, in the impression content shown in Figure 1, a model C car is classified into the impression "freedom" and impression "goal selection" which are subdivided from the impression "self-determination," the impression "enjoyment of life" and impression "fun" which are subdivided from the impression "hedonism," the impression "competence" which is subdivided from the impression "achievement," and the impression "health" and classification "family safety" which are subdivided from the impression "safety."

Furthermore, in the impression content shown in Figure 1, model D automobiles are classified into the impressions "creativity," "freedom," and "goal choice," which are subdivisions of the impression "self-determination," the impression "enjoyment of life," which is a subdivision of the impression "hedonism," the impression "influence," which is a subdivision of the impression "achievement," and the impression "wealth," which is a subdivision of the impression "power."

In this way, the information processing device 1 accepts a selection of target content that will be used to estimate the impression IM that the user U has or has had, and a selection of an impression group that is a group of impressions IM, and uses the generation AI to generate impression content that indicates the impression IM that the user U has or is estimated to have had of the target content from among the multiple impressions IM included in the accepted impression group. This allows the information processing device 1 to reduce the workload required to estimate the user U's impression of the target content.

The following provides a detailed description of the configuration of an information processing system that includes an information processing device 1 that performs such processing, multiple terminal devices 2, terminal device 3, and information processing device 4.

2. Configuration of the information processing system
2 is a diagram illustrating an example of a configuration of an information processing system according to an embodiment. As illustrated in FIG. 2, the information processing system 100 according to the embodiment includes an information processing device 1, a plurality of terminal devices 2, a terminal device 3, and an information processing device 4.

The multiple terminal devices 2 are used by different users U. The terminal device 3 is, for example, the terminal device of the worker O. The terminal devices 2 and 3 are, for example, notebook PCs (Personal Computers), desktop PCs, smartphones, tablet PCs, and wearable devices. Examples of wearable devices include, but are not limited to, smart glasses or smart watches.

The information processing device 4 provides various online services to the user U. For example, the information processing device 4 provides the user U with content from various sites such as e-commerce sites, news sites, store introduction sites, image posting sites, video viewing sites, and online bulletin board sites, but is not limited to these examples.

For example, on an EC site, the information processing device 4 accepts product content such as product descriptions from workers O and others as posted content, and provides such posted content to each user U. Note that product descriptions can also be referred to as product descriptions. Furthermore, posted content on an EC site includes not only product content, but also comments and ratings posted about the product.

In addition, the information processing device 4 accepts news articles from workers O and the like as posted content on the news site, and provides such posted content to each user U. On the news site, posted content includes not only news articles but also comments posted on such news articles.

In addition, the information processing device 4 accepts store content, such as a store introduction text from the worker O, as posted content on the store introduction site, and provides such posted content to each user U. In addition to store content, the posted content on the store introduction site also includes reviews and comments posted about the store. Store introduction sites are, for example, sites for each type of store, such as a restaurant introduction site.

In addition, the information processing device 4 accepts image content (e.g., still image content and video content) from workers O and others as posted content on an image posting site, and provides such posted content to each user U. In addition, the information processing device 4 accepts video content and titles and caption text for video content as posted content on a video viewing site, and provides such posted content to each user U.

Information processing device 1, terminal device 2, terminal device 3, and information processing device 4 are connected to each other via network N so that they can communicate with each other via wired or wireless connections. Note that the information processing system 100 shown in FIG. 2 may include multiple information processing devices 1, etc.

Network N includes, for example, a WAN (Wide Area Network) such as the Internet, and mobile communication networks such as LTE (Long Term Evolution), 4G (4th Generation), and 5G (5th Generation: 5th generation mobile communication system).

Terminal devices 2 and 3 can connect to network N via short-range wireless communication such as a mobile communication network, Bluetooth (registered trademark), or wireless LAN (Local Area Network), and communicate with information processing device 1, information processing device 4, etc.

3. Configuration of information processing device 1
3 is a diagram showing an example of the configuration of the information processing device 1 according to the embodiment. As shown in FIG. 3, the information processing device 1 includes a communication unit 10, a storage unit 11, and a processing unit 12.

3.1. Communication Unit 10
The communication unit 10 is realized by, for example, a communication module or a network interface card (NIC). The communication unit 10 is connected to a network N via a wired or wireless connection and transmits and receives information to and from various other devices. For example, the communication unit 10 transmits and receives information to and from each of the terminal device 2, the terminal device 3, and the information processing device 4 via the network N.

[3.2. Storage unit 11]
The storage unit 11 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 11 has a user information storage unit 20.

[3.2.1. User information storage unit 20]
The user information storage unit 20 stores user information including information about the user U. Fig. 4 is a diagram showing an example of a user information table stored in the user information storage unit 20 of the information processing device 1 according to the embodiment. As shown in Fig. 4, the user information table stored in the user information storage unit 20 includes items such as "user ID,""attributeinformation," and "behavioral history."

"User ID" is identification information that identifies user U. "Attribute information" is attribute information of user U that corresponds to the "user ID," and includes, for example, psychographic attribute information and demographic attribute information. Demographic attributes include, for example, gender, age, place of residence, and occupation, while psychographic attributes include interests such as travel, clothing, cars, and religion, lifestyle, thoughts, and ideological tendencies.

"Behavioral history" refers to the user U's behavioral history in online services, and includes information such as search history, browsing history, posting history, and purchase history. Search history includes information on search queries used by user U in past searches and the content viewed by user U from among the search results. Search query information includes, for example, search keywords and search phrases.

The browsing history includes, for example, information indicating the content that the user U viewed on the online service, and the posting history includes, for example, information indicating the content that the user U posted in the past on the online service (e.g., reviews, comments, etc.). The purchase history includes information on the items of transactions that the user U has made in the past.

3.3. Processing Unit 12
The processing unit 12 is a controller, and is realized by a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit) executing various programs (corresponding to examples of information processing programs) stored in a storage device inside the information processing device 1 using RAM or the like as a working area.

The processing unit 12 is also a controller and may be realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), or GPGPU (General Purpose Graphics Processing Unit).

As shown in FIG. 3, the processing unit 12 has a reception unit 30, an acquisition unit 31, a generation unit 32, and a provision unit 33, and realizes or executes the information processing functions and actions described below. Note that the internal configuration of the processing unit 12 is not limited to the configuration shown in FIG. 3, and may be any other configuration that performs the information processing described below.

3.3.1. Reception unit 30
The reception unit 30 receives various information and requests from the terminal device 2 or 3 via the network N and the communication unit 10. For example, the reception unit 30 receives a selection of target content that is to be used to estimate an impression IM held by the user U or an impression IM that the user U has held, and a selection of an impression group that is a group of impressions IM.

For example, an impression content generation request sent from the terminal device 3 of a worker O who creates or provides the target content is accepted. The impression content generation request includes the target content that is the subject of estimation of the impression IM that the user U has or has had, information indicating an impression group that is a group of impressions IM, information specifying the estimation method, information indicating the target content type, and information indicating the impression content type.

The target content, impression group, estimation method, target content type, and impression content type specified in the impression content generation request are, for example, the target content, impression group, estimation method, target content type, and impression content type selected by worker O.

By receiving a request for impression content generation, the reception unit 30 receives selections for target content, impression group, estimation method, target content type, and impression content type. Note that worker O is an employee of the company that creates or provides the target content, but may also be an employee of a company commissioned by such company.

Target content is web content, such as, but not limited to, web pages, catch phrases, creative content, and stamps. Target content is content that is a candidate for or is to be provided on, for example, e-commerce sites, news sites, restaurant introduction sites, image posting sites, video viewing sites, etc., and includes, but is not limited to, advertising content. E-commerce sites include, for example, shopping sites, auction sites, etc., but are not limited to, these examples.

The target content is content that can be provided on an online site provided by the information processing device 4, and the user U can use the online site provided by the information processing device 4 by operating the terminal device 2. Online sites provided by the information processing device 4 include, but are not limited to, e-commerce sites, news sites, store introduction sites, image posting sites, and video viewing sites.

The impression group specified in the impression content generation request is an impression group specified by user U from among multiple impression groups C1 to Cn (n is an integer equal to or greater than 2). The multiple impression groups C1 to Cn may be, for example, impression groups categorized by user U's values, impression groups categorized by user U's preferences, impression groups categorized by user U's lifestyle, etc., but are not limited to such examples.

The impression group based on user U's values is, for example, an impression group that classifies values based on Schwartz's value theory as impressions IM that user U holds or has held, and each of the 10 classification conditions mentioned above is further subdivided into multiple classification conditions, resulting in 56 classification conditions, but is not limited to such an example.

User U's values are not limited to the examples mentioned above, and may be values categorized as traditionalism, success orientation, self-actualization orientation, coexistence orientation, etc., or other values. For example, if the object is a car, user U's values may be values categorized as prioritizing safety, livability, design, driving performance, etc., and if the object is a house, they may be values categorized as prioritizing location, price, layout, design, future prospects, etc.

The impression group based on user U's lifestyle may be, for example, an impression group based on the AIO approach or an impression group based on VALS. The impression group based on the AIO approach is an impression group categorized by one or more of activities, interests, opinions, etc. The impression group based on VALS is an impression group categorized by, for example, self-actualizers, successful people, group members, those who maintain their livelihood, those in poverty, those who aspire to success, those with social conscience, intellectuals, and young intellectuals.

The impression group may also be, for example, the impression group indicated by innovator theory, which classifies the process of diffusion of new products. This impression group is classified into innovators, early adopters, early majority, late majority, and laggards.

There are multiple estimation methods for estimating the impression IM that a user U has or has had on target content, including a first estimation method and a second estimation method. The first estimation method is a method in which the generation AI directly estimates the impression IM that a user U has or has had on target content. The second estimation method is a method in which, for each impression IM included in a group of impressions IM, the generation AI generates reference content that is estimated to represent the impression IM that the user U has, and compares the reference content with the target content to estimate the impression IM that the user U has or has had on target content.

The target content type is, for example, the type of target content, and is indicated, for example, by a combination of an online site and a content type. For example, the target content type may be, but is not limited to, a product description on a product page on an e-commerce site, a news article on a news site, a comment on a news article on a news site, image content on an image posting site, or the title and caption text of video content on a video viewing site.

The impression content type is a type of impression content, and includes, for example, a first content type and a second content type. The first content type is a type of impression content that includes information indicating the impression IM that the user U is presumed to have had of the target content, and is a type of impression content that supports the creation and consideration of the target content. The second content type is a type of impression content that includes information indicating the impression IM that the user U is presumed to have had of the target content, and is a type of impression content that supports understanding the impression IM that the user U is presumed to have had of the target content.

[3.3.2. Acquisition unit 31]
The acquisition unit 31 acquires various pieces of information from the information processing device 4 and the storage unit 11. For example, the acquisition unit 31 acquires various pieces of content and information about each user U from the information processing device 4 via the network N and the communication unit 10. The acquisition unit 31 also acquires information about each user U from the user information storage unit 20 of the storage unit 11.

For example, when the reception unit 30 receives a request for impression content generation, the acquisition unit 31 acquires from the impression content generation request the target content that is the subject of estimation of the impression IM that the user U has or has had, information indicating an impression group that is a group of impressions IM, information specifying the estimation method, information indicating the target content type, and information indicating the impression content type.

Furthermore, when the target content type indicated in the impression content generation request received by the receiving unit 30 is the second content type, the acquisition unit 31 collects behavioral content, which is content indicating the user U's behavior regarding the target content, from the information processing device 4, etc.

Behavioral content is content that arises from actions by user U that are related to the target content, such as posted content such as reviews, comments, and messages posted by user U regarding the target content. For example, if the target content is a product description or product image, posted content as behavioral content is a review or comment posted about the product introduced in the target content.

In addition, if the target content is a store description or an image of the store interior, the posted content as behavioral content is a review or comment posted about the store introduced by the target content.In addition, if the target content is video content, the posted content as behavioral content is a review or comment posted about the video content.

In addition, posted content as behavioral content may be content such as conversations or posts by user U on the official account of the product or store indicated by the target content in the communication application.In addition, posted content as behavioral content may be content such as posts or conversations by user U on an online bulletin board site, for example.

[3.3.3. Generation unit 32]
The generation unit 32 uses the generation AI to generate impression content indicating the impression IM that the user U has or is estimated to have had about the target content from among the multiple impressions IM included in the impression group received by the reception unit 30.

The generation AI used in step S2 may be, for example, a text generation AI, an image generation AI, or a multimodal AI. Such generation AI may be located, for example, in an external information processing device, and the generation unit 32 may have the generation AI generate and obtain reference content for each impression via an API provided by the external information processing device, but this is not a limitation. For example, the generation AI may be located within the generation unit 32.

Text generation AI is, for example, a language model trained to estimate and output the next token from an input token sequence, such as a transfer-based model or an RNN-based model.

Examples of transfer-based models include, but are not limited to, GPT and BARD. Examples of RNN-based models include, but are not limited to, RWKV. It is desirable to keep input information, such as personal information, confidential by training the model in a way that prevents it from being used as a new answer.

Examples of image generation AI include, but are not limited to, StackGAN, AttnGAN, T2I (Text-to-Image) with Transformers, and DALL-E. Examples of multimodal AI include, but are not limited to, models that generate images from text and generate text from images, such as GPT-4V and CM3Leon.

The generation unit 32 can use multiple estimation methods, including a first estimation method and a second estimation method, as estimation methods for estimating the impression IM that the user U has or has had, regarding the target content. The first estimation method is a method in which the generation AI directly estimates the impression IM that the user U has or has had regarding the target content.

The second estimation method is a method in which, for each impression included in the impression group, the generation AI generates reference content that is estimated to represent the impression that user U has or has had, and compares this reference content with the target content to estimate the impression IM that user U has or has had of the target content.

Here, it is assumed that the estimation method and impression content type indicated in the impression content request received by the reception unit 30 are the first estimation method and first content type. In this case, the generation unit 32 inputs, as input information to the generation AI, information including, for example, instruction information instructing the user U to select the impression IM that the user U has of the target content from among the multiple impressions IM1-IMm indicated in the impression information, impression information indicating the multiple impressions IM1-IMm included in the selected impression group, and the target content, and causes the generation AI to output information indicating one or more impressions IM that the user U is estimated to have of the target content.

Also, assume that the estimation method and impression content type indicated in the impression content request received by the reception unit 30 are the first estimation method and the second content type. In this case, the generation unit 32 inputs, as input information to the generation AI, information including, for example, instruction information instructing the user to select the impression IM formed by the user U from the behavioral content from among the multiple impressions IM1-IMm indicated in the impression information, impression information indicating the multiple impressions IM1-IMm included in the selected impression group, and the behavioral content, and causes the generation AI to output information indicating the impression IM that is estimated to have been formed by the user U with respect to the target content.

Also, assume that the estimation method and impression content type indicated in the impression content request received by the reception unit 30 are the second estimation method and the first content type. In this case, the generation unit 32 causes the generation AI to generate reference content that is estimated to have the impression IM of the user U for each impression IM included in the selected impression group, and compares the reference content with the target content to estimate one or more impressions IM that the user U has of the target content.

Also, assume that the estimation method and impression content type indicated in the impression content request received by the reception unit 30 are the second estimation method and second content type. In this case, the generation unit 32 causes the generation AI to generate reference content that is estimated to have been the impression IM held by the user U for each impression IM included in the selected impression group, and compares the reference content with the behavioral content to estimate one or more impressions IM held by the user U regarding the target content.

The generation unit 32 includes a first generation processing unit 40, an estimation processing unit 41, and a second generation processing unit 42. The first generation processing unit 40 executes processing when the estimation method indicated in the impression content request received by the reception unit 30 is the second estimation method.

3.3.3.1. First Generation Processing Unit 40
When the estimation method indicated in the impression content request received by the receiving unit 30 is the second estimation method, the first generation processing unit 40 generates reference content that is estimated to be the impression that the user U has or has had for each impression IM classified into multiple categories in the selected impression group.

3.3.3.2. Processing for the first content type
First, a case will be described in which the impression content type indicated in the impression content request received by the receiving unit 30 is the first content type. In this case, the first generation processing unit 40 generates, for each impression IM classified into multiple categories in the selected impression group, reference content that is estimated to give the user U the impression.

For example, the first generation processing unit 40 causes the generation AI to generate reference content that is estimated to be the impression IM that the user U has for each impression IM included in the selected impression group. If the selected impression group is an impression group based on values based on Schwartz's value theory, the multiple impressions IM classified in the selected impression group are the 10 types of values or 56 types of values described above.

For example, the first generation processing unit 40 inputs input information to the generation AI, including instruction information that instructs the generation of reference content that is estimated to give the user U the impression indicated by the impression IM, and causes the generation AI to generate the reference content. The reference content is, for example, content represented by at least one of text and an image.

The first generation processing unit 40 stores standard instruction information in advance, and inputs information including the standard instruction information and impression information to the generation AI as instruction information, causing the generation AI to generate reference content for each impression IM. For example, if the target content is a product description, the standard instruction information is the string "I will post certain values. Please write 10 product descriptions that are likely to share those values with Japanese users. Please do not include too many of the values you have posted directly in the sentences."

Furthermore, the impression information is, for example, information on the character string "Values: Social Power." As a result, each of the 10 product descriptions that are predicted to give the user U the impression IM of social power is generated as reference content.

When the generating AI is a GPT, the standard instruction information is included in the input information (prompt) as an instruction, and the impression information is included in the input information (prompt) as a user message, but this example is not limited to this. For example, instead of the string "Post a certain value. That value" in the standard instruction information, the input information may include information containing the string "Social power as a value" as a user message.

3.3.3.3. Processing for the second content type
Next, a case will be described in which the impression content type indicated in the impression content request received by the receiving unit 30 is the second content type. In this case, the first generation processing unit 40 generates, for each impression IM classified into multiple categories in the selected impression group, reference content that is estimated to have given the user U the impression.

For each impression IM classified into multiple categories in the selected impression group, the first generation processing unit 40 generates reference content that is presumed to have been the impression that the user U had. For example, the first generation processing unit 40 inputs input information to the generation AI that includes instruction information that instructs the generation of reference content that is content that is presumed to have been the impression that the user U had indicated in the impression IM and corresponds to the behavioral content, and causes the generation AI to generate reference content. The reference content is, for example, content that is expressed as at least one of text and an image.

The first generation processing unit 40 stores first standard instruction information in advance, and inputs information including the standard instruction information and impression information to the generation AI as instruction information, causing the generation AI to generate reference content for each impression IM. For example, if the target content is a product description, the standard instruction information is the string "I will post a certain value. Please write 10 comments by users who are presumed to share those values with Japanese users. Please do not include too many of the posted values directly in the sentences."

The impression information is, for example, the string "Values: Social Power." As a result, each of the 10 comments from user U that are estimated to have given user U the impression IM of social power is generated as reference content.

3.3.3.4. Estimation Processing Unit 41
The estimation processing unit 41 estimates the impression IM that the user U has or has had with respect to the target content. The estimation processing unit 41 estimates the impression IM that the user U has or has had with respect to the target content, based on the estimation method and impression content type indicated in the impression content request accepted by the accepting unit 30.

3.3.3.4.1. Processing for the First Estimation Method and the First Content Type
A case will be described in which the estimation method and impression content type indicated in the impression content request received by the receiving unit 30 are the first estimation method and the first content type. In this case, the estimation processing unit 41 causes the generation AI to estimate the impression IM that the user U has of the target content from among multiple impressions IM1 to IMm included in the selected impression group.

For example, the estimation processing unit 41 inputs to the generation AI as input information instruction information instructing the user U to select the impression IM that the user U has of the target content from among the multiple impressions IM1 to IMm indicated in the impression information, impression information indicating the multiple impressions IM1 to IMm included in the selected impression group, and information including the target content, and causes the generation AI to output information indicating one or more impressions IM that the user U is estimated to have of the target content.

Figure 5 is a diagram showing an example of input information input to the generation AI by the estimation processing unit 41 in the processing unit 12 of the information processing device 1 according to the embodiment. This is an example in which the selected impression group is a group of impressions based on the values of the user U, and the target content is a catchphrase that reads, "You can tell the difference because it's something you use every day."

In the input information 50 shown in Figure 5, the estimation processing unit 41 includes the information "Please select three or more images from the values below that fit the catchphrase 'You can tell the difference because it's something you use every day.' The output format should be in CSV format as shown below. \nValues, reasons" as instruction information in the input information.

Furthermore, in the input information 50, the estimation processing unit 41 includes information containing the character strings "Social power, Authority, Wealth, ..., Family security, Reciprocation of favors, Health, Sense of belonging" as impression information.

For example, instead of the information in the string "Please select three or more of the values below that fit the catchphrase," the estimation processing unit 41 can use information such as "Please select three or more of the values below that represent the impression that users are likely to have of the catchphrase," or information in the string "Please estimate a score indicating the degree of impression that users are likely to have of the catchphrase for each of the values below. The score should range from 1 to 10, with a higher value representing a higher degree of impression." However, the output format is not limited to such examples.

The instruction information also includes text information according to the type of target content. In the example above, the type of target content is a catchphrase, so the instruction information includes the string "catchphrase." However, if the type of target content is a product description on a product page on an e-commerce site, the instruction information includes the string "product description."

In this way, the estimation processing unit 41 generates instruction information that corresponds to the type of target content by including text information that corresponds to the type of target content in the instruction information, but it is also possible to input instruction information that is independent of the type of target content to the generation AI.

For example, instead of the string "Please select three or more images from the values below that fit the catchphrase 'You can tell the difference because it's something you use every day,'" the estimation processing unit 41 can set instruction information containing the string "Please select three or more images from the values below that fit the information you input." as the system message. A system message is a message that instructs the behavior of the generating AI, and is, for example, a message set in the input information as the content of "'role':'system'" in OpenAI's GPT model.

The estimation processing unit 41 can also use the generation AI to estimate the impression IM that user U has of the target content from multiple impressions IM1 to IMm included in the selected impression group by limiting the attributes of user U, etc. In this case, instead of the information in the string "Please select three or more images from the values below that fit the catchphrase," the estimation processing unit 41 can use information such as "Please select three or more impressions from the values below that are estimated to be formed by a man in his twenties for the catchphrase." or the information in the string "Please estimate, for each of the values below, a score indicating the degree of impression that a man in his twenties is estimated to have for the catchphrase. The score should range from 1 to 10, with a higher value indicating a higher degree of impression." When the generation AI is to estimate one or more impressions IM formed by users U with attributes other than men in their twenties, the information in the string "men in their twenties" can be replaced with information indicating other attributes.

Furthermore, when the target content is image content, the estimation processing unit 41 can input information including the target content and instruction information containing information such as the string "Please select three or more images from the values below that fit the input image," as input information to the multimodal AI, and cause the multimodal AI to output a similarity. Note that, as in the case when the target content is other than image content, the estimation processing unit 41 can limit the attributes of the user U and output a score indicating the degree of impression IM.

3.3.3.4.2. Processing in the case of the first estimation method and the second content type
A case will be described where the estimation method and impression content type indicated in the impression content request received by the receiving unit 30 are the first estimation method and the second content type.

In this case, the estimation processing unit 41 inputs, as input information to the generation AI, information including, for example, instruction information instructing the user U to select the impression IM formed by the behavioral content from among the multiple impressions IM1 to IMm indicated in the impression information, impression information indicating the multiple impressions IM1 to IMm included in the selected impression group, and the behavioral content, and causes the generation AI to output information indicating the impression IM that is estimated to have been formed by the user U regarding the target content.

For example, if the selected impression group is a group of impressions based on user U's values, the behavioral content is represented by the character string "XXX...XXX". In this case, the estimation processing unit 41 generates instruction information including the information on the character string "'XXX...XXX' is a comment posted by the user regarding the target content below. Please select three or more of the values below that represent the image that user U is estimated to have of the target content below. Please output in CSV format as shown below. \nValues, reasons", impression information, and target content.

In addition, the estimation processing unit 41 can use, for example, information such as "Please estimate a score indicating the degree of impression that user U is estimated to have had of the target content for each of the values below. The score should range from 1 to 10, with a higher value indicating a higher degree of impression," instead of the information in the string "Please select three or more of the values below that represent the impression that user U has of the target content.", but is not limited to this example.

The output format is not limited to CSV. Instead of the target content, the instruction information may include a string of characters indicating the product, store, etc. identified by the target content. In this case, this string of characters is used in place of the above-mentioned string "target content below." Part of the above-mentioned instruction information can be set as a system message.

The estimation processing unit 41 can generate behavioral content that includes multiple user U's content regardless of the user U's attributes in the instruction information, or can generate behavioral content that includes multiple user U's content for each user U's attribute in the instruction information. When the estimation processing unit 41 includes multiple user U's content for each user U's attribute in the instruction information, it can cause the generation AI to output information indicating one or more impressions IM that are estimated to have been held by the user U for each user U's attribute.

Furthermore, when the target content is an image, the estimation processing unit 41 can input information containing instruction information such as "The following image is an image posted by a user regarding the following target content" to the multimodal AI as input information, instead of the information in the string "'XXX...XXX' is a comment posted by a user regarding the following target content," and cause the multimodal AI to output information indicating one or more impressions IM that are estimated to have been held by the user U.

In addition, when the behavioral content includes text and images, the estimation processing unit 41 can include the text and images as behavioral content in the instruction information, input information including such instruction information to the multimodal AI as input information, and output it from the multimodal AI.

3.3.3.4.3. Processing in the case of the second estimation method and the first content type
A case where the estimation method and impression content type indicated in the impression content request received by the receiving unit 30 are the second estimation method and the first content type will be described.

In this case, the estimation processing unit 41 compares the target content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM, and estimates one or more impressions IM held by the user U regarding the target content based on the comparison results.

The estimation processing unit 41 compares the target content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM. In order to compare the target content with the reference content for each impression IM, the estimation processing unit 41 vectorizes the target content and each reference content.

Content vectorization is performed, for example, by embedding using a language model (e.g., a transformer-based model). The vectorized content is represented, for example, by a vector with several hundred dimensions, but this example is not limiting. Embedding using a language model is, for example, embedding using text-embedding-ada or BERT provided by OpenAI (registered trademark), but this example is not limiting.

Note that content vectorization is not limited to embedding using a language model; content vectorization may also be performed using, for example, Doc2Vec or the average of word embeddings. Word2Vec, fastText, etc. are used for word embedding.

The estimation processing unit 41 classifies the target content into an impression IM corresponding to a reference content whose vector similarity with the target content is equal to or greater than a threshold. The estimation processing unit 41 can, for example, classify the target content into only one impression IM, or into two or more impression IMs.

For example, the estimation processing unit 41 can classify the target content into only one impression IM by classifying the target content into an impression IM that corresponds to the reference content with the highest similarity, where the vector similarity is equal to or greater than a threshold value.

Furthermore, when there are two or more reference contents whose vector similarity with the target content is equal to or greater than a threshold, the estimation processing unit 41 can classify the target content into two or more impressions IM corresponding to these two or more reference contents, respectively.

The vector similarity is cosine similarity, but it may also be Jaccard similarity, or the inverse of the Euclidean distance or the inverse of the Manhattan distance. Furthermore, when using the Euclidean distance or the Manhattan distance, the estimation processing unit 41 classifies, for example, target content whose Euclidean distance or the Manhattan distance from the reference content is less than a threshold value into an impression IM corresponding to the reference content.

The estimation processing unit 41 can also use any method to determine the similarity between vectors or classify vectors, and ultimately the similarity between content items or classify content items, such as by learning a classification model using vectors and the corresponding hand-labeled data and then using that model for classification.

In addition, when the reference content and the target content are images, the estimation processing unit 41 can input information including the reference content, the target content, and instruction information indicating an instruction to output the similarity between the reference content and the target content to the multimodal AI as input information, and cause the multimodal AI to output the similarity.

In this case, the estimation processing unit 41 classifies the impression IM into, for example, an impression IM generated by multimodal AI that corresponds to the reference content with the highest similarity, which is equal to or greater than a threshold, or an impression IM generated by multimodal AI that corresponds to the reference content with a similarity equal to or greater than a threshold.

In addition, when the reference content and the target content are images, the estimation processing unit 41 can also cause the multimodal AI to generate explanatory text that explains what images are contained in each of the reference content and the target content.

In this case, the estimation processing unit 41 vectorizes each of the explanatory text of the reference content and the explanatory text of the target content, and calculates the similarity between the vectorized explanatory text of the reference content and the vectorized explanatory text of the target content. The method of calculating the similarity and the group of impressions of the target content to the impression IM are the same as when the reference content and the target content are text.

In addition, when the reference content and the target content are images, the estimation processing unit 41 can also directly vectorize each of the reference content and the target content to calculate the similarity.

Furthermore, when the reference content and the target content each contain text and images, the estimation processing unit 41 classifies the impressions into impressions IM corresponding to reference content whose integrated score, which is a score obtained by weighting the similarity between the text and the similarity between the images, is equal to or greater than a threshold, and impressions IM corresponding to reference content whose integrated score is equal to or greater than a threshold and is the largest.

3.3.3.4.4. Processing for the second estimation method and the second content type
Next, a case where the estimation method and the impression content type indicated in the impression content request received by the receiving unit 30 are the second estimation method and the second content type will be described.

In this case, the estimation processing unit 41 compares, for each piece of moving content, the behavioral content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM, and estimates one or more impressions IM held by the user U regarding the target content based on the comparison results.

The estimation processing unit 41 compares, for each behavioral content, the behavioral content that is the subject of estimation of the impression IM held by the user U with the reference content for each impression IM. In order to compare the behavioral content with the reference content for each impression IM, the estimation processing unit 41 vectorizes each behavioral content and each reference content. The method for vectorizing content is as described above.

The estimation processing unit 41 performs a process of vectorizing the behavioral content and calculating the similarity between the vector of each reference content for each behavioral content. Then, the estimation processing unit 41 estimates, for example, impressions IM corresponding to reference content whose average vector similarity with each behavioral content is equal to or greater than a threshold, as one or more impressions IM that the user U has had of the target content.

Furthermore, the estimation processing unit 41 estimates, for example, impressions IM corresponding to reference content whose median value of the vector similarity with each behavioral content is equal to or greater than a threshold value as one or more impressions IM that the user U has of the target content.

3.3.3.5. Second generation processing unit 42
The second generation processing unit 42 generates, based on the estimation result by the estimation processing unit 41, content including information indicating one or more impressions IM into which the target content has been classified, as impression content.

For example, the second generation processing unit 42 generates, as impression content, content that maps multiple impressions IM1 to IMm classified into specific impression groups and includes information indicating the impressions IM that the user U has or is presumed to have had regarding the target content.

[3.3.4. Providing Department 33]
The providing unit 33 provides the impression content generated by the generating unit 32. For example, the providing unit 33 provides the impression content generated by the generating unit 32 to the worker O by transmitting the impression content generated by the generating unit 32 to the terminal device 3 of the worker O via the communication unit 10 and the network N.

The terminal device 3 receives impression content transmitted from the providing unit 33 of the information processing device 1 via the communication unit 10 and the network N, and displays the received impression content. Figure 6 is a diagram showing an example of impression content generated by the second generation processing unit 42 in the processing unit 12 of the information processing device 1 according to the embodiment.

The impression content 60 shown in Figure 6 is content in which impressions IM classified as impressions held by user U based on values based on Schwartz's value theory are highlighted in content in which multiple values based on Schwartz's value theory are mapped as multiple impressions IM1 to IMm (m = 56 in Figure 1). In the example shown in Figure 1, the target content is content that shows introductory text or introductory images of four different models of a certain car model.

In impression content 60, the model A automobile is classified into the impressions "freedom," "independence," and "self-respect," which are subdivisions of the impression "self-determination," and into the impression "inner harmony," which is a subdivision of the impression "universalism."

Furthermore, in impression content 60, the model B car is classified into the impressions "Variety of life" and "Vibrant life" which are subdivisions of the impression "Stimulation," the impressions "Enjoyment of life" and "Fun" which are subdivisions of the impression "Hedonism," and the impression "Belonging" which is subdivisions of the impression "Safety."

Furthermore, in impression content 60, a model C automobile is classified into the impression "freedom" and impression "goal selection" which are subdivisions of the impression "self-determination," the impression "enjoyment of life" and impression "fun" which are subdivisions of the impression "hedonism," the impression "competence" which is subdivisions of the impression "achievement," and the impression "health" and classification "family safety" which are subdivisions of the impression "safety."

Furthermore, in impression content 60, the model D automobile is classified into the impressions "creativity," "freedom," and "goal selection" which are subdivisions of the impression "self-determination," the impression "enjoyment of life" which is subdivisions of the impression "hedonism," the impression "influence" which is subdivisions of the impression "achievement," and the impression "wealth" which is subdivisions of the impression "power."

4. Processing Procedure
Next, a procedure of information processing by the processing unit 12 of the information processing device 1 according to the embodiment will be described. Fig. 7 is a flowchart showing an example of information processing by the processing unit 12 of the information processing device 1 according to the embodiment.

As shown in FIG. 7, the processing unit 12 of the information processing device 1 determines whether an impression content request has been received (step S10). If the processing unit 12 determines that an impression content request has been received (step S10: Yes), it uses the generation AI to estimate the impression that the user U has or is estimated to have had, based on the information included in the impression content request (step S11).

Next, the processing unit 12 generates impression content based on the impression estimated in step S11 (step S12). Then, the processing unit 12 provides the impression content generated in step S12 to the worker O (step S13).

When the processing of step S13 is completed, or when it is determined that a request for video content has not been received (step S10: No), the processing unit 12 determines whether the operation end time has arrived (step S14). The processing unit 12 determines that the operation end time has arrived, for example, when the power to the information processing device 1 is turned off.

If the processing unit 12 determines that the operation end timing has not yet arrived (step S14: No), it proceeds to step S10. If the processing unit 12 determines that the operation end timing has arrived (step S14: Yes), it terminates the processing shown in FIG. 7.

5. Modifications
In the above example, the generation unit 32 estimated the impression IM that the user U had based on the behavioral content of the user U, but it can also estimate the impression IM that the user U had of the target content based on attribute information (e.g., psychographic attributes) and other behavioral information of the user U who performed an action related to the target content.

The generation unit 32 can also estimate the impression IM that the user U has or has had regarding the target content by performing a weighted addition of the impression IM estimated using the first estimation method and the impression IM estimated using the second estimation method.

The generation unit 32 can also estimate an impression IM that is equal to or greater than a threshold value as the impression IM that the user U has or has had regarding the target content, by performing a weighted addition of the score of each impression IM estimated using the first estimation method and the score of each impression IM estimated using the second estimation method.

The generation unit 32 can also estimate an impression IM for which the estimation results from the first estimation method and the second estimation method match as the impression IM that the user U has or has had with the target content.

The generation unit 32 can also estimate multiple impressions IM, including an impression IM estimated using the first estimation method and an impression IM estimated using the second estimation method, as the impression IM that the user U has or has had regarding the target content.

6. Hardware Configuration
The information processing device 1 according to the embodiment described above is realized by, for example, a computer 80 configured as shown in Fig. 8. Fig. 8 is a hardware configuration diagram showing an example of the computer 80 that realizes the functions of the information processing device 1 according to the embodiment. The computer 80 has a CPU 81, a RAM 82, a ROM (Read Only Memory) 83, an HDD (Hard Disk Drive) 84, a communication interface (I/F) 85, an input/output interface (I/F) 86, and a media interface (I/F) 87.

The CPU 81 operates based on programs stored in the ROM 83 or HDD 84 and controls each component. The ROM 83 stores a boot program executed by the CPU 81 when the computer 80 starts up, as well as programs that depend on the computer 80's hardware.

The HDD 84 stores programs executed by the CPU 81 and data used by such programs. The communication interface 85 receives data from other devices via the network N (see Figure 2) and sends it to the CPU 81, and transmits data generated by the CPU 81 to other devices via the network N.

The CPU 81 controls output devices such as a display and printer, and input devices such as a keyboard or mouse, via the input/output interface 86. The CPU 81 acquires data from the input devices via the input/output interface 86. The CPU 81 also outputs data generated via the input/output interface 86 to the output devices.

The media interface 87 reads programs or data stored on the recording medium 88 and provides them to the CPU 81 via the RAM 82. The CPU 81 loads the programs from the recording medium 88 onto the RAM 82 via the media interface 87 and executes the loaded programs. The recording medium 88 may be, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase Change Rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory.

For example, when the computer 80 functions as the information processing device 1 according to the embodiment, the CPU 81 of the computer 80 executes programs loaded onto the RAM 82 to realize the functions of the processing unit 12. In addition, the HDD 84 stores data in the storage unit 11. The CPU 81 of the computer 80 reads and executes these programs from the recording medium 88, but as another example, these programs may be obtained from another device via the network N.

[7. Other]
Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using known methods. In addition, the information including the processing procedures, specific names, various data, and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown in the drawings.

Furthermore, the components of each device shown in the figure are functional concepts and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

For example, the information processing device 1 described above may be realized by a terminal device and a server computer, or by multiple server computers. Furthermore, depending on the function, the configuration can be flexibly changed, such as by calling an external platform using an API or network computing.

Furthermore, the above-described embodiments and variations can be combined as appropriate to the extent that the processing content is not contradictory.

8. Effects
As described above, the information processing device 1 according to the embodiment includes a receiving unit 30, a generating unit 32, and a providing unit 33. The receiving unit 30 receives a selection of target content that is a target for estimating an impression IM that the user U has or has had, and a selection of an impression group that is a group of impressions IM. The generating unit 32 uses a generation AI to generate impression content that indicates an impression IM that the user U has or is estimated to have had of the target content, from among the multiple impressions IM included in the impression group received by the receiving unit 30. The providing unit 33 provides the impression content generated by the generating unit 32. This enables the information processing device 1 to reduce the workload required to estimate the user U's impression of the target content.

The generation unit 32 also causes the generation AI to estimate the impression IM that the user U has or is estimated to have had of the target content from among multiple impression IMs. This allows the information processing device 1 to reduce the workload required to estimate the user U's impression of the target content.

The generation unit 32 also inputs, as input information to the generation AI, information including impression information indicating the multiple impressions IM included in the impression group and instruction information instructing the generation AI to select the impression IM that the user U has or has had regarding the target content from the multiple impressions IM indicated by the impression information, and causes the generation AI to output information indicating the impression IM that the user U has or is estimated to have had regarding the target content. This allows the information processing device 1 to accurately estimate the user U's impression of the target content.

The generation unit 32 also inputs information further including the target content as input information to the generation AI, and causes the generation AI to output information indicating the impression IM that the user U is estimated to have of the target content. This allows the information processing device 1 to accurately estimate the impression the user U has of the target content.

The generation unit 32 also inputs information to the generation AI as input information that further includes the target content or behavioral content, which is content generated by user U's behavior related to the target content, and causes the generation AI to output information indicating the impression that user U is estimated to have had of the target content. This allows the information processing device 1 to accurately estimate user U's impression of the target content.

The generation unit 32 also includes a first generation processing unit 40, an estimation processing unit 41, and a second generation processing unit 42. The first generation processing unit 40 generates, for each impression IM included in the impression group, reference content that is estimated to represent the impression IM that the user U has or has had. The estimation processing unit 41 estimates the impression IM that the user U has or has had based on the results of a comparison between each reference content and either action content, which is content resulting from an action by the user U related to the target content, or the target content. The second generation processing unit 42 generates impression content based on the estimation results by the estimation processing unit 41. This allows the information processing device 1 to accurately estimate the user U's impression of the target content.

The receiving unit 30 also receives the selection of one impression group from a plurality of predetermined impression groups. This allows the information processing device 1 to reduce the workload required to estimate the user U's impression of the target content.

The generation unit 32 also generates, as impression content, content that maps multiple impressions included in the impression group received by the reception unit 30 and that includes information indicating the impression IM that the user U has or is presumed to have had of the target content. This allows the information processing device 1 to enable the worker O to easily understand the impression that the user U has of the target content.

The above describes in detail the embodiments of the present application based on the drawings, but this is merely an example, and the present invention can be implemented in other forms that incorporate various modifications and improvements based on the knowledge of those skilled in the art, including the aspects described in the Disclosure of the Invention section.

Furthermore, the "section, module, unit" mentioned above can be read as "means" or "circuit." For example, an acquisition unit can be read as an acquisition means or an acquisition circuit.

REFERENCE SIGNS LIST 1, 4 Information processing device 2, 3 Terminal device 10 Communication unit 11 Storage unit 12 Processing unit 20 User information storage unit 30 Reception unit 31 Acquisition unit 32 Generation unit 33 Provision unit 40 First generation processing unit 41 Estimation processing unit 42 Second generation processing unit 100 Information processing system N Network

Claims (10)

a reception unit that receives a selection of target content that is a target of an impression held by a user or an impression that has been held by a user, and a selection of an impression group that is a group of impressions;
a generation unit that generates, using a generation AI, impression content indicating an impression that the user has or is presumed to have had regarding the target content from among a plurality of impressions included in the impression group received by the reception unit;
a providing unit that provides the impression content generated by the generating unit. The generation unit
The information processing device according to claim 1 , wherein the generation AI estimates the impression that the user has or is estimated to have had about the target content from the plurality of impressions. The generation unit
The information processing device according to claim 2, characterized in that information including impression information indicating the plurality of impressions included in the impression group and instruction information instructing the user to select an impression that the user has or has had regarding the target content from the plurality of impressions indicated by the impression information is input to the generation AI as input information, and the generation AI is caused to output information indicating the impression that the user has or is estimated to have had regarding the target content. The generation unit
The information processing device according to claim 3, characterized in that information further including the target content is input to the generation AI as the input information, and the generation AI outputs information indicating the impression that the user is estimated to have of the target content. The generation unit
The information processing device of claim 3, characterized in that information further including the target content or behavioral content, which is content generated by an action by the user related to the target content, is input to the generation AI as the input information, and the generation AI is caused to output information indicating the impression that the user is estimated to have had of the target content. The generation unit
a first generation processing unit that generates, for each impression included in the group of impressions, reference content that is estimated to be the impression that the user has or has had;
an estimation processing unit that estimates an impression that the user has or is estimated to have had based on a comparison result of comparing the target content or behavioral content, which is content generated by an action by the user related to the target content, with each of the reference contents;
The information processing device according to any one of claims 1 to 5, further comprising: a second generation processing unit that generates the impression content based on a result of estimation by the estimation processing unit. The reception unit
6. The information processing apparatus according to claim 1, wherein the information processing apparatus accepts the selection of one impression group from a plurality of predetermined impression groups. The generation unit
The information processing device according to any one of claims 1 to 5, characterized in that the impression content is generated by mapping the plurality of impressions included in the impression group received by the receiving unit, and the content includes information indicating the impression that the user has or is estimated to have had regarding the target content. 1. A computer-implemented information processing method, comprising:
a receiving step of receiving a selection of target content for which the impression held by the user or the impression held by the user is to be estimated, and a selection of an impression group which is a group of impressions;
a generating step of generating impression content indicating an impression that the user has or is presumed to have had regarding the target content from among a plurality of impressions included in the impression group received by the receiving step, using a generating AI;
a providing step of providing the impression content generated in the generating step. a receiving step for receiving a selection of target content for which the user has an impression or an impression that has been received by the user is to be estimated, and a selection of an impression group that is a group of impressions;
a generation procedure for generating, using a generation AI, impression content indicating an impression that the user has or is presumed to have had regarding the target content from among a plurality of impressions included in the impression group accepted by the acceptance procedure;
and a providing procedure for providing the impression content generated by the generating procedure.