JP7832288B2 - Product recommendation methods and systems - Google Patents

Description

This invention relates to a product recommendation method and system for recommending appropriate products that reflect user needs.

The dictionary definition of artificial intelligence is a technology that uses computer programs to realize human learning abilities, reasoning abilities, perceptual abilities, and natural language comprehension abilities. Such artificial intelligence has made remarkable progress through deep learning.

In particular, with the development of artificial intelligence, various language models have been developed. These language models not only recognize text and understand its meaning, but have also reached a level where they can extract and classify information from vast amounts of text-based data, such as documents, and even directly generate text.

Such language models are actively utilized in various fields, including search engines, document creation (e.g., resume writing, report writing, poster creation, etc.), free-form dialogue on various topics, data parsing in given texts (e.g., data summarization, classification, etc.), expert knowledge provision, programming, and conversion of given texts into appropriately styled texts—in short, they exist in a wide range of text-based applications.

Based on such language models, it is possible to automate various tasks that were previously performed directly by humans. In particular, text-based tasks that were previously generated or modified directly by humans in various service sectors (e.g., shopping services, news services, etc.) can be automated. However, in doing so, research is needed on how to utilize language models while taking into account the characteristics of the service.

This invention provides a product recommendation method and system that can be used in shopping services.

In particular, the present invention provides a product recommendation method and system that can recommend appropriate products to users when they purchase items as gifts for others, according to the recipient and intention of the gift.

Furthermore, the present invention provides a product recommendation method and system that can recommend appropriate products according to the user's situation or the circumstances of the gift, using a language model.

To solve the above problems, the product recommendation method according to the present invention may include the steps of: identifying target story information including at least one of target subject information and target intent information; obtaining a recommended product category corresponding to the target story information based on sample story information including at least one of sample subject information and sample intent information, and sample product categories corresponding to each of the sample story information; and extracting at least one recommended product from among the products belonging to the recommended product category that matches the target story information.

The product recommendation method according to the present invention may include the steps of: receiving target story information from a user terminal, including information about at least one of the recipient of the gift and the intention behind the gift; configuring a first prompt to be input to a pre-specified large-scale language model using at least one dataset (dataset) including sample story information and theme information of sample product categories, and the target story information; obtaining a plurality of theme keywords corresponding to the target story information as the output of the large-scale language model based on the first prompt; providing the plurality of theme keywords to the user terminal, and having at least one of the plurality of theme keywords selected by the user terminal; configuring a second prompt including the specific theme keyword selected by the user terminal; and processing the second prompt including the specific theme keyword as input to the large-scale language model to obtain a recommended gift category corresponding to the target story information from the large-scale language model.

The product recommendation system according to the present invention may include a storage unit that stores sample story information, which includes at least one of sample target information and sample intent information, and sample product categories corresponding to the sample story information; and a control unit that uses a pre-specified large-scale language model to perform a process of obtaining recommended product categories corresponding to target story information based on the sample story information and the sample product categories corresponding to the sample story information, and a process of extracting at least one recommended product from among the products belonging to the recommended product category that matches the target story information.

In the electronic device according to the present invention, a program executed by one or more processes and stored on a computer-readable recording medium may include commands to perform the following steps: identifying target story information including at least one of target target information and target intent information; obtaining a recommended product category corresponding to the target story information based on sample story information including at least one of sample target information and sample intent information, and sample product categories corresponding to each of the sample story information; and extracting at least one recommended product from among the products belonging to the recommended product category that matches the target story information.

As described above, the product recommendation method and system according to the present invention can extract recommended products using a large-scale language model that takes prompts containing story information as input.

Here, the story information is structured to include information about the recipient of the recommended product and the purchaser's intention to buy the recommended product. Therefore, in this invention, it is possible to recommend products that reflect the user's intention to purchase the recommended product.

Thus, the product recommendation method and system according to the present invention can recommend products that match the user's story from a large language model by constructing prompts using story information that reflects the recipient of the recommended product and the intention behind giving the gift. Therefore, according to the present invention, the time required to select a gift can be reduced, and a product optimized according to the recipient and the intention behind the gift can be purchased.

This is a conceptual diagram illustrating an example of how to utilize the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the product recommendation system according to the present invention. This is a flowchart illustrating the product recommendation method according to the present invention. This is a conceptual diagram illustrating the prompts used in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for inputting story information in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for inputting story information in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for inputting story information in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for inputting story information in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for inputting story information in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for inputting story information in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for inputting story information in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating an example of a prompt used in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating an example of a prompt used in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating an example of a prompt used in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating an example of a prompt used in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating an example of a prompt used in the product recommendation method and system according to the present invention. This is a conceptual diagram to explain product categories and product recommendation methods. This is a conceptual diagram to explain product categories and product recommendation methods. This is a conceptual diagram to explain product categories and product recommendation methods. This is a conceptual diagram illustrating the method for extracting recommended products in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the method for extracting recommended products in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the message generation method in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the message generation method in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the message generation method in the product recommendation method and system according to the present invention. This is a conceptual diagram illustrating the message generation method in the product recommendation method and system according to the present invention.

The embodiments disclosed herein will be described in detail below with reference to the accompanying drawings. Regardless of the drawing number, identical or similar components will be denoted by the same or similar reference numerals, and their descriptions will be omitted. The suffixes "module" and "part" used in the following description are added or used interchangeably to facilitate the preparation of the specification and do not inherently possess any significance or usefulness. Furthermore, when describing the embodiments disclosed herein, if it is determined that a specific description of related prior art would obscure the gist of the embodiments disclosed herein, such detailed description will be omitted. The accompanying drawings are merely for the purpose of facilitating the understanding of the embodiments disclosed herein, and the technical ideas disclosed herein are not limited by the accompanying drawings. Any modifications, equivalents, or substitutions included within the concept and scope of the present invention should be understood as being included in the present invention.

Terms including ordinal numbers, such as "first," "second," etc., are used to describe various components, but the components themselves are not limited by these terms. These terms are used solely to distinguish one component from others.

When it is stated that one component is "connected" or "linked" to another component, it should be understood that this connection may be direct, or that other components may exist in between. Conversely, when it is stated that one component is "directly connected" or "directly linked" to another component, it should be understood that there are no other components in between.

Unless otherwise specified, singular expressions can include multiple forms.

In this specification, terms such as "includes" and "have" are intended to indicate the presence of features, figures, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood not to preemptively exclude the existence or possibility of adding one or more other features, figures, steps, actions, components, parts, or combinations thereof.

As mentioned above, with the development of artificial intelligence, various language models have been developed, and based on these, it has become possible to automate various tasks that were previously performed directly by humans. In particular, it has become possible to automate various tasks that were previously selected or generated directly by humans in various service fields (e.g., shopping services, news services, etc.). Specifically, this invention provides a product recommendation method and system that can recommend products suitable for a user, taking into account the various circumstances of the user purchasing the product.

Users purchase goods online for themselves or for others, and especially when purchasing goods for others, there may be a recipient and intention to provide the purchased goods.

For example, purchasing a product for someone else may lead to giving it as a gift. In this case, the user can search for and purchase the product online, keeping in mind their intention to give it to someone else and the recipient.

The product recommendation method and system according to the present invention are designed to facilitate the purchase of goods for others. This invention reduces the time users spend selecting goods to purchase for others and provides a product recommendation method and system that can recommend goods according to the user's purchase intent and the recipient of the goods.

In this invention, for the sake of clarity, the terms "purchase of goods" and "purchase of gifts" are used interchangeably, and the invention is not necessarily limited to the purchase of gifts. That is, it goes without saying that the invention can be used when a user purchases goods for themselves or when purchasing goods for others that are not intended as gifts. Therefore, "purchase of gifts" in this invention may be replaced with "purchase of goods," or may be interpreted as "purchase of goods." However, in the case of gift purchases, the purchaser and the user of the goods are different, making it impossible to directly recommend goods that reflect the purchaser's behavioral characteristics and preferences. This invention provides a method to solve this problem. The term "goods" in this invention may be used not only for tangible articles but also for intangible services.

More specifically, as shown in Figure 1, the product recommendation method and system according to the present invention may recommend products to the user using story information. Here, the products recommended to the user may be products that the user can purchase online.

The story information in this invention may include information about at least one of the target and intent. Here, "target" may be understood as the party receiving the product recommended in the product recommendation system according to the present invention. For example, if a user uses the product recommendation system according to the present invention for a gift, "target" may be the person receiving the gift (the party or recipient). Here, the information about the target (or target information) included in the story information may consist of at least one of various pieces of information that can describe a specific person who falls under the target, such as gender, age, occupation, relationship to the user, hobbies, preferences, and place of residence.

Furthermore, "intent" may be interpreted as the reason or purpose for a user purchasing a product recommended in the product recommendation system according to the present invention. For example, if a user purchases a product as a gift, "intent" may be interpreted as at least one piece of information that can explain the intent behind the gift (e.g., a housewarming gift, a moving gift, a wedding anniversary gift, etc.).

On the other hand, in the product recommendation method and system according to the present invention, story information can be input by the user in various ways. For example, in the service page 110 provided by the product recommendation system according to the present invention, story information may be input by the user, or, as shown in Figure 1(a), one of the provided story information lists 112 may be selected.

On the other hand, in the product recommendation system according to the present invention, a language model (LM) or a large language model 10 may be used to obtain recommended product categories corresponding to story information. Furthermore, in the product recommendation system according to the present invention, based on this, recommended products corresponding to the story information may be extracted and provided to the user, as shown in Figure 1(b). Additionally, at least one keyword related to the recommended products or keywords that can classify them may be extracted and provided to the user.

On the other hand, the language model used in this invention is not limited in type; any language model that has been trained on various types of information is acceptable. Meanwhile, large-scale language models, or hyperscale language models, which have been actively researched in recent years, are models trained on large amounts of data using models that have more than 10 times the number of parameters (for example, more than 100 billion parameters) compared to conventional general language models. They can perform inference without fine-tuning of the model, using methods such as few-shot learning. Such large-scale language models can understand patterns within prompts, given a small amount of example data format, and generate and output probability distributions of word sequences or words (or sentences) based on them.

Here, the recommended product category may be understood as the category or type to which the recommended product belongs. Product categories may be classified based on various criteria such as product use, product color, and product price range. For example, depending on the product's use, product categories may include categories such as accessories, stationery, and automotive supplies. Here, the recommended product in this invention may mean a product belonging to a specific category that is actually available for purchase online. That is, the product recommendation system according to the present invention may recommend products to the user and support the user through to the purchase of the recommended products.

Here, the recommended products may be made available for purchase on a shopping service page 120 (for example, a web page or application page) provided by a shopping service linked to the product recommendation system according to the present invention, as shown in Figure 1(b).

The product recommendation method and system according to the present invention will be described in more detail below with reference to the attached drawings. Figure 2 is a conceptual diagram illustrating the product recommendation system according to the present invention, and Figure 3 is a flowchart illustrating the product recommendation method according to the present invention. Figure 4 is a conceptual diagram illustrating prompts used in the product recommendation method and system according to the present invention, and Figures 5, 6, 7A, 7B, 8, 9A, and 9B are conceptual diagrams illustrating the input method of story information in the product recommendation method and system according to the present invention. Furthermore, Figures 10A, 10B, 11, 12A, and 12B are conceptual diagrams illustrating examples of prompts used in the product recommendation method and system according to the present invention, and Figures 13A, 13B, and 13C are conceptual diagrams illustrating product categories and product recommendation methods. Figures 14 and 15 are conceptual diagrams illustrating the extraction method of recommended products in the product recommendation method and system according to the present invention, and Figures 16A, 16B, 17A, and 17B are conceptual diagrams illustrating the message generation method in the product recommendation method and system according to the present invention.

First, as shown in Figure 2, the product recommendation system 200 according to the present invention may include at least one of an input unit 210, a storage unit 220, and a control unit 230. Although not shown, the product recommendation system 200 may further include a communication unit, which may be configured to communicate with at least one server, electronic device (or user terminal).

Furthermore, the product recommendation system 200 may be configured to include at least one of the large-scale language model 10 and the product recommendation engine 20, and may be configured to work in conjunction with at least one of them.

The product recommendation system 200 according to the present invention may acquire recommended product categories from the large language model 10 based on story information input to the large language model 10, and extract recommended products corresponding to the recommended product categories using the product recommendation engine 20. At least a part of the configuration of the product recommendation system 200 may be configured to process information input to at least one of the large language model 10 and the product recommendation engine 20, or to process information output from at least one of the large language model 10 and the product recommendation engine 20.

Regarding the input unit 210, it may be configured to receive information that constitutes a prompt to be input to the large language model 10.

The input unit 210 may include i) an interface connecting a user (or user terminal) to the product recommendation system 200 according to the present invention, or ii) an interface connecting the product recommendation system 200 according to the present invention to at least one other system. Furthermore, the input unit 210 may include interfaces connecting components included in the product recommendation system 200 according to the present invention.

In this invention, by inputting a prompt into the large language model 10, the system may obtain a recommended product category and its corresponding recommended product as a result.

Here, the recommended product category may be obtained from the large language model 10, and the recommended products belonging to the recommended product category may be obtained from the product recommendation engine 20. The control unit 230 may control a series of processes for obtaining recommended products from the product recommendation engine 20 by inputting information about the recommended product category obtained from the large language model 10 to the product recommendation engine 20.

The prompt input to the large language model 10, as shown in Figure 4(a), is data input to the large language model 10 and may include datasets 411, 412, and 413 (hereinafter referred to as "sample datasets") that are referenced in the large language model 10, and dataset 414 (hereinafter referred to as "target dataset") that corresponds to the information to be obtained from the large language model 10.

Such datasets 411, 412, 413, and 414 may include the story information mentioned above.

The sample dataset may consist of pairs of sample story information and corresponding sample product category information. Here, the sample story information may be generated using information created or produced by the user, or it may be generated by extracting the target and intent from the user's product purchase breakdown. Alternatively, the sample story information may be generated by inputting prompts containing the aforementioned information (e.g., user-created or produced information, user product purchase breakdown, etc.) into a trained language model or large-scale language model.

The target story information included in the target dataset may be data entered by the user in text format, as explained with reference to Figure 1(a), or it may be data selected from a story information list. Furthermore, it may be generated by combining at least one piece of information, such as the user's basic information (e.g., age, gender) and information entered by the user (e.g., target, intention).

The target dataset may not include target product categories corresponding to target story information, i.e., recommended product categories, and may be configured to be obtained from the large language model 10.

The prompt 410 may be configured to include at least one sample dataset 411, 412, 413 and at least one of the target dataset 414. The configuration of the prompt will be described in detail later. The method for retrieving the information and its corresponding product categories will also be described in more detail later.

On the other hand, according to another embodiment of the present invention, the prompt input to the large language model 10 may be an embedding vector generated using a trained embedding model, instead of a text format. Here, the prompt embedding model may be trained using sample story information created or generated by the user, or sample story information generated by extracting the target and intent from the user's product purchase breakdown.

Next, the storage unit 220 may be configured to store various types of information. The storage unit 220 may reside at least one location, either inside or outside, the product recommendation system according to the present invention. It goes without saying that the storage unit 220 may be configured as a server or as cloud-based storage.

The storage unit 220 may be configured to store data relating to at least one of multiple sample datasets and target datasets.

Multiple sample datasets may be configured to include at least one of the following: story information, product category information, theme information, keyword information, and tag information.

At least some of the diverse information included in the sample datasets listed above may consist of information collected from various pages accessible via the internet.

The collected information may consist of information created or produced by any user on the internet. Such data may be referred to as UGC (User Generated Content). Furthermore, at least a portion of the various information included in the sample dataset may consist of information generated by the large-scale language model 10.

Thus, the storage unit 220 may store at least one of the following: story information, product category information, theme information, keyword information, and tag information, collected through various channels. These pieces of information may form pairs, constituting a dataset to be input to the prompt.

Next, the control unit 230 may perform general control necessary for obtaining recommended product categories and recommended products. For example, the control unit 230 may be configured to interact with the large language model 10 so that the prompt 410 is processed as input to the large language model 10. Furthermore, the product recommendation system 200 according to the present invention may include the large language model 10 as a component, in which case the control unit 230 can perform appropriate control over the large language model 10.

As another example, the control unit 230 may be configured to work in conjunction with the product recommendation engine 20 so that information regarding recommended product categories obtained from the large language model 10 is processed as input to the product recommendation engine 20.

On the other hand, the information generated by the large language model 10 may be stored in the storage unit 220, or it may be configured to be output from an output unit (not shown) for user confirmation, and processed appropriately depending on the situation. The control unit 230 may process the information generated by the large language model 10; for example, if information regarding recommendation categories needs to be visually output to the user, it may provide that information to the user terminal.

In the product recommendation method and system according to the present invention, as shown in Figures 1 and 2, recommended product categories may be obtained by inputting a dataset containing at least one of story information and corresponding product category information into the large language model 10.

As explained with reference to Figure 4, the large language model 10 receives a prompt 410 consisting of sample datasets 411, 412, 413 and a target dataset 414. Each dataset may consist of data fields (data field) corresponding to the data type (category, or data type, data attribute) and their corresponding data values.

As shown in Figure 4(a), data fields may be understood as defining the meaning of data values or the type of data values, such as "Story Information" or "Product Category." Such data fields may also be understood as "Category Name (name)," "Data Type," or "Data Attribute."

On the other hand, data fields may follow predefined rules so that the large language model 10 can recognize them. For example, the large language model 10 may recognize that the data entered after "Story Information:" corresponds to the data field named "Story Information".

In the large-scale language model 10, data corresponding to a specific data field (e.g., "Product Category:") in the target dataset may be generated by referencing data pairs included in the dataset.

In this invention, as shown by reference numeral 414b in Figure 4(a), the prompt may be configured such that the end of 410 consists only of a data field, thereby obtaining the corresponding data value as the output of the large language model 10. As a result, as shown by reference numeral 424b in Figure 4(b), data values for the "product category" data field (e.g., picture frames, houseplants, vases, diffusers, candle warmers, lighting) can be obtained.

Needless to say, you may change the names of data fields to more appropriate terminology depending on the context.

Next, the product recommendation engine 20 is an engine that recommends products belonging to recommended product categories obtained from the large language model 10, and may be configured to recommend at least one appropriate product based on at least two similarity levels of story information, product category, and product name. In this invention, the type of product recommendation engine 20 can be anything.

The method for generating product recommendation information will be described in more detail below, with reference to the product recommendation system 200 according to the present invention described above.

As shown in Figure 3, in the present invention, a process for identifying target story information may be performed (S310).

As mentioned above, target story information may also refer to the story information that pertains to the product category and product recommendation.

The target story information may include information about at least one of the target audience and the target intent. Here, the target audience may be understood as the party receiving the product recommended in the product recommendation system according to the present invention. For example, if a user uses the product recommendation system according to the present invention for a gift, the "target audience" is the person (party or recipient) receiving the gift. Here, the information about the target audience (or target information) included in the target story information may consist of at least one of various pieces of information that can describe a specific person who falls under the target audience, such as gender, age, occupation, relationship to the user, hobbies, preferences, and place of residence.

Furthermore, "target intent" may be interpreted as the reasons or purposes for which a user who purchases a product recommended in the product recommendation system according to the present invention purchases the product. For example, if a user purchases a product as a gift, it may be interpreted as at least one of various pieces of information that can explain the intent behind the gift (e.g., a housewarming gift, a moving gift, a wedding anniversary gift, etc.).

On the other hand, the aforementioned "target story information," "target subject," and "target intent" have the same or similar meanings as the "story information," "subject," and "intent" defined above. However, for the sake of clarity, the term "target" has been added to distinguish them from "sample story information," "sample subject information," and "sample intent information."

The control unit 230 may identify (or generate) target story information based on at least one of the following: information input from the user terminal (e.g., user input information) and user information linked to the user account.

The control unit 230 may further include user information about the user in the target story information and identify (or generate) the target story information.

The control unit 230 may extract specific user information to be included in the target story information from user information linked to the user account of the user, or it may request the user to input specific user information. Here, the specific user information may include information regarding at least one of the user's age and gender.

Here, the user's age may be used to infer the age of the recipient of the product (for example, the recipient of the gift), and the user's gender may be used to infer the popularity of the product among users purchasing the product (for example, the gift).

For example, if the target story information includes information about the user's age (e.g., a teenage user), it may be inferred that the teenage user's friends are also teenagers, and that the teenage user's parents are in their 40s or 50s. Similarly, if the target story information includes age as user information, the age of the recipient of the gift may be estimated.

Furthermore, user gender may be used to consider the popularity of gifts among users of that gender. For example, women in their 20s tend to primarily give jewelry as birthday presents for their sisters, while men in their 20s tend to primarily give digital devices as birthday presents for their brothers.

In this invention, by including user information together with target story information, the age of the recipient of the gift can be inferred, and product categories and products that take into account the popularity of gifts among the user giving the gift can be recommended.

On the other hand, target story information may be included in the prompt 410 so that it is processed as input to the large language model 10 described above. The control unit 230 may generate target story information using at least one of user input information and user information based on natural language processing (NLP). The control unit 230 may also generate target story information in sentence format so that it includes at least one of user input information and user information. Therefore, user input information and user information are not necessarily limited to a sentence format, and may be transformed into a data structure that can be processed by the large language model 10 through processing performed by the control unit 230.

On the other hand, in this invention, information may be input by the user in various ways, and this information may be used to identify the target story information.

As an example, as shown in Figure 5, in this invention, target story information may be identified based on information entered into an information input area 510 contained in a specific page 500 provided by a product recommendation service or shopping service.

The information input area 510 may consist of various information input areas, such as a product search area or a story information input area.

The control unit 230 may use user input information (for example, "My best friend moved last week. What should I buy her? Something expensive is fine.") entered from the user terminal (or user account) in the information input area 510 to identify target story information.

Thus, the control unit 230 may identify target story information based on the product recommendation request received from the user account in the information input area 510.

The control unit 230 may extract information regarding at least one of the target object and target intent from the user input information. The control unit 230 may analyze the user input information based on natural language processing and extract information regarding at least one of the target object (e.g., a close friend) and target intent (e.g., a housewarming gift) necessary to identify the target story information. Then, it may generate the target story information using the extracted information.

As another example, the control unit 230 may use a chatbot to identify the target story information, as shown in Figures 6(a) and 6(b). The control unit 230 may provide messages 611, 612, and 616 to the user terminal to identify the target story information. Alternatively, the control unit 230 may identify the target story information based on message 616 received from the user terminal.

The dialogue messages 611, 612, and 616 for identifying the target story information may be provided to the user terminal multiple times. The content of the next message may be appropriately selected and provided to the user terminal based on the content of message 616 received from the user terminal. Here, the control unit 230 may identify any missing information in the target story information and configure the messages provided to the user in a manner that further guides them to obtain that information.

On the other hand, in the product recommendation system 200 according to the present invention, a chatbot may be used to provide information (e.g., a question message) to the user terminal for identifying target story information. Based on the target story information, at least one of the extracted recommended product category information and recommended product information 614, 616, and 623 may be provided to the user terminal via the chatbot. This will be described in detail later.

As another example, the control unit 230 may, as shown in Figures 7A and 7B, receive user input information for identifying target story information on a specific page provided to the user terminal in the product recommendation system 200 according to the present invention.

As shown in Figure 7A, the specific page 710 may be configured to include at least one of the following: a first input area 711 for receiving target information corresponding to a recipient (or person, for example, a person receiving a gift) who will receive (get) the product recommended in the product recommendation system 200 according to the present invention; and a second input area 712 for receiving target intention information corresponding to the intention to give (present) the product recommended to the recipient.

In other words, the specific page 710 may include multiple input areas 711 and 712 for receiving information from the user regarding the recipient (target) of the recommended product and the intention to give (present) the product. The control unit 230 may identify story information based on the user input information received in the multiple input areas 711 and 712.

On the other hand, as shown in Figures 7A and 7B, the information 713 and 715 constituting the specific page 710 may change according to the user input information entered on the specific page 710.

For example, as shown in Figure 7A, on a specific page 710, the target and intent may be entered sequentially by the user. Therefore, a single area 713 on the specific page 710 may provide information about multiple different targets (e.g., friends, parents, baby, mother, sister). Here, the information about the targets provided to the user may be appropriately determined considering user information (e.g., gender, age, and other family relationship information). For example, if the user is a woman in her 40s, the targets may include her husband, in-laws, and children; if the user is a teenage boy, the targets may include friends, teachers, and parents.

Information regarding such targets may be represented by graphic objects (or icons), as shown in Figure 7A. The control unit 230 may, when one of the graphic objects (for example, graphic object 713a corresponding to "friend") is selected, identify the target by matching the information (for example, "friend") to the selected graphic object.

Information regarding a specific object may be input into the first input area 711, as shown in Figure 7B. Furthermore, it goes without saying that the user can also directly input information regarding the object into the first input area 711. That is, the first input area 711 may be configured to receive information directly from the user terminal.

Next, as shown in Figure 7B, a specific area 715 of page 710 may be configured to provide information related to multiple different intentions (such as satisfaction, gratitude, birthday, friendship, housewarming, etc.). Here, the intentions provided to the user may be determined by considering pre-selected information about the target and user information. For example, if the target is different, such as "friend" versus "teacher," the information provided regarding the intentions may also differ.

Such information regarding intent may be represented by graphic objects (or icons), as shown in Figure 7A. The control unit 230 may, when one of the graphic objects (for example, graphic object 715a corresponding to "housewarming gift") is selected, identify the information matched to the selected graphic object (for example, "housewarming gift") as the intent.

Furthermore, information regarding a specific intention may be entered into the second input area 712. The second input area may display intentions with similar characteristics grouped together, and one of the grouped intentions may be selected by the user. Needless to say, the user may also directly input information regarding intentions into the second input area 712.

Thus, in this invention, information regarding the target and intent may be received based on user input information. When the control unit 230 receives information regarding the target and intent, it may use the received information to identify story information. That is, the target target and target intent may be identified by the method described above, and target story information may be identified based on these.

The control unit 230 may use a natural language processing algorithm to generate a text containing information about the object (or target object) and intention (or target intention) received in the first input area 711 and the second input area 712, as shown in Figures 7A and 7B. Such a text may be processed into a form included in the prompts of the large language model 10, as described above.

As another example, as shown in Figure 8, the control unit 230 may use a specific page 800 to provide the user terminal with a list 810 containing various story information, and the user may select one of the provided story information items to identify the story information.

As shown in the figure, a specific page 800 may contain a list 810 composed of multiple different story information. The control unit 230 may, at the user terminal, select one of the items 820 in the list 810 and identify the story information corresponding to the selected item (for example, "Please recommend a gift for meeting my future mother-in-law.") as the target story information.

As another example, the control unit 230 may, as shown in Figure 9A, use a specific page 910 to allow the user to input story information. As shown in the same figure, the specific page 910 may also provide guide information 911 to guide the user in inputting story information.

Furthermore, specific pages 910 may be provided with guide information 912 to guide the user in entering information that must be included in the story information (e.g., information regarding the target and intent). By referring to this information, the user can enter the story information, including the target and intent.

Furthermore, a specific page 910 may be provided with sample story information 913 containing information about the target and intent. This sample story information allows the user to intuitively understand what kind of story information they need to input.

On the other hand, sample story information may be configured to be selected by the user's terminal. By selecting sample story information, the user can input story information.

Furthermore, as shown in Figure 9B, a specific page 920 may be provided with an additional information input area 922 for more specifically identifying the story information.

The control unit 230 may provide an information input (or selection) area 922a on a specific page 920 for more specifically identifying at least one of the subjects and intentions that must be included in the story information.

Furthermore, the specific page 920 may be provided with an area 922b where various information necessary for product recommendation, such as the price range of the product, is entered.

The control unit 230 may identify story information based on user input information received from the user terminal on such a page.

Thus, in the product recommendation system 200 according to the present invention, story information (target story information) may be identified based on user input information received through various methods.

As mentioned above, user input information does not necessarily have to be in sentence format. In that case, the control unit 230 may use the user input information to generate a sentence 1014a in a form that can be included in the prompt 1010, as shown in Figure 10A(a).

For example, if user input consists of multiple words, the control unit 230 may combine the words to generate a sentence and include it in the prompt that is processed as input to the large language model 10.

On the other hand, the control unit 230 may identify story information using user information stored on a specific server linked to the user account, in order to recommend products more suitable for the user. In this case, as shown in Figure 10B(a), the control unit 230 may further utilize user information 1032 (e.g., teens, male) in addition to story information 1031 based on information entered by the user (user input information) to identify story information 1045, as shown in Figure 10B(b). The story information may include first information 1045a based on user information (e.g., "I'm a male student in my teens.") and second information 1045b based on user input information (e.g., "I'm a woman in my 30s who likes interior design; please recommend a housewarming gift."). The control unit 230 may generate at least one sentence that includes user input information and user information, and include it in the prompt 1040 which is processed as input to the large language model 10.

On the other hand, user information may include at least one of the gender, age, residential area, and occupation of a specific user corresponding to a specific user account, and may also include various other information stored in the storage unit 220 regarding that specific user.

In this invention, when recommending products, user information may be taken into consideration to recommend product categories and products that are appropriate to each user's situation. For example, if age is entered as user information, the price range of products that a teenage user and a user in their 30s can purchase may differ.

Furthermore, in this invention, when recommending products, user information may be considered to infer information about the recipient of the product (for example, the recipient of a gift corresponding to the product), and at least one product category and product suitable for the recipient may be recommended using the inferred information about the recipient. For example, user information may include age (e.g., a teenage user), and user input information (e.g., "I'm thinking of giving a gift to a friend") may be received. In this case, it is inferred that the teenage user's friend is also a teenager, and as a result, it is inferred (or estimated) that the recipient of the gift is a teenage user. In this way, in this invention, by inferring information about the recipient based on user information, it is possible to recommend a product category or product suitable for the recipient.

As another example, user information such as the user's age (e.g., a teenage user) may be entered, and user input information (e.g., "I'm thinking of giving my parents a present") may be received. In this case, the parents of a teenage user are statistically inferred to be in their 40s or 50s, and as a result, the recipient of the present is inferred (or estimated) to be a user in their 40s or 50s. Thus, in this invention, by inferring the recipient's information based on user information, it is possible to recommend a product category or product suitable for the recipient (e.g., a user in their 40s or 50s).

In this invention, story information can be identified by reflecting user information so that product categories and products suitable for each user's situation are recommended.

As a result, in this invention, the types of recommended product categories obtained as output of the large-scale language model 10 may vary depending on the user information. Similarly, the types of recommended products obtained as output of the product recommendation engine 20 may vary depending on the user information.

As described above, in this invention, information including at least one of the subject and intent can be input by the user in various ways, and story information can be identified based on that information.

Once the story information is identified in this way, that is, once the target story information is identified, the present invention may perform a process to obtain a recommended product category corresponding to the target story information based on the sample story information and the sample product category (S320).

The control unit 230 may configure the prompt to include target story information in order to obtain recommended product categories. The control unit 230 may also configure the prompt to include sample story information, the sample product categories corresponding to the sample story information, and the target story information. Furthermore, the control unit 230 may process the prompt as input to the large language model 10.

Furthermore, the large-scale language model 10 may generate product recommendation categories as output (or response) to the input prompt, based on the input prompt.

As explained with reference to Figure 4, the prompt may include the sample datasets 411, 412, and 413 referenced in the large language model 10, and the target dataset 414 corresponding to the information to be obtained from the large language model 10.

The types of data fields included in sample datasets 411, 412, and 413 can be varied.

As an example, sample datasets 411, 412, and 413 may include data fields corresponding to story information and data fields corresponding to product categories, as explained with reference to Figure 4.

As another example, the sample dataset may include data fields corresponding to story information, data fields corresponding to tags, and data fields corresponding to product categories. Here, tags may also be referred to as images, topics, keywords, etc.

In the large-scale language model 10, the recommended product categories may be output by referring to the data constituting the sample dataset included in the prompt and the data constituting the target dataset.

On the other hand, the control unit 230 may select the data to be input at the prompt, such as data constituting the sample dataset, for example, sample story information, sample tag information, sample product category, etc. This information may be stored in the storage unit 220.

Furthermore, this information may be based on user-generated content (UGC) collected from the internet. Additionally, this information may be generated from a large-scale language model 10.

The control unit 230 may configure a prompt 1110 to obtain sample story information, as shown in Figure 11(a), and prompt the large language model to input it. Here, as shown in Figure 11(a), the data field corresponding to the story information of the target dataset 1111 included in the prompt 1110 may exist as an empty area.

In the large-scale language model 10, based on a prompt inputting a dataset (or sample dataset) consisting of gift categories and story information data fields, at least one story information 1120 may be acquired, as shown in Figure 11(b). The various story information acquired in this way may be utilized as information included in the sample dataset, as explained with reference to Figure 4. On the other hand, the data fields corresponding to gift categories and story information included in the sample dataset shown in Figure 11(a) may be information collected from the internet. This information may also be user-generated data (UGC).

Thus, the control unit 230 may acquire recommended product categories from the large language model 10 as output (response) to the prompt, based on the prompt input to the large language model 10.

On the other hand, when a recommended product category is obtained from the large language model 10, the control unit 230 may provide information about the recommended product category to the user terminal. For example, as shown in Figure 13A, information about the recommended product category obtained from the large language model 10 (e.g., kitchenware, cleaning supplies, interior goods, bedding) may be provided in a specific area 1330 of the page 1310 where story information is entered. The user can then view the recommended product category and select a specific recommended product category (e.g., "interior goods") to select the product category they wish to purchase.

Thus, the product categories obtained from the large-scale language model 10 may be provided to the user terminal as a response to the user input information on the page where story information (user input information) is entered from the user terminal. Therefore, the user can predict what products will be recommended based on the information they have entered, and by selecting a specific product category, they can specify the range of recommended products.

As an example, as shown in Figure 5, information regarding recommended product categories 530 may be provided as a response to the information entered in the input area 510.

As another example, as shown in Figure 6(a), the chatbot may provide information 615 regarding recommended product categories as a response to a message 613 entered by the user.

Up to this point, we have described how to immediately obtain information about recommended product categories from the large language model 10 as a response to story information based on user input.

Furthermore, the control unit 230 may, when recommending products, first retrieve tags corresponding to story information from the large language model 10 in order to reflect the user's intentions. Then, when at least one of the retrieved tags is selected by the user, the control unit 230 may secondarily retrieve a recommended product category from the large language model 10 that reflects information related to the selected tag. In this way, the control unit 230 can recommend products that reflect the story information based on user input and the user's intentions.

More specifically, the control unit 230 may receive target story information from the user terminal, including information about at least one of the target object (e.g., the recipient of the gift) and the target intention (e.g., the intention behind the gift). It may also configure a first prompt to be input to a pre-specified large-scale language model using at least one dataset (or sample dataset) containing sample story information and theme keywords (or tags) for sample product categories, as well as the target story information. As shown in Figure 12A, prompt 1210, consisting of a sample dataset containing sample story information and sample tags, and a target dataset 1220 containing target story information and tag data fields, may be input to the large-scale language model 10. Based on prompt 1210, the control unit 230 may obtain multiple tags (or theme keywords) corresponding to the target story information as the output 1210 of the large-scale language model.

As shown in Figures 6 and 13B, the acquired tags (e.g., high-end, health essentials, stylish, great atmosphere, Instagrammable, home cafe, etc.) may be provided to areas 615 and 1360 on specific pages 610 and 1350 as a response to user input information.

Furthermore, in this invention, the user terminal may be provided with multiple tags (or theme keywords), and at least one of the multiple tags may be selected on the user terminal. Additionally, the control unit 230 may configure a prompt (or a second prompt) to include the selected tag, and may retrieve a recommended product category from the second prompt.

The control unit 230 may process a second prompt containing a specific tag selected by the user as input to the large language model 10, thereby obtaining a recommended gift category corresponding to the target story information from the large language model 10.

Thus, the control unit 230 may obtain a recommended product category that reflects the user's intent by sequentially inputting multiple prompts to the large language model 10. On the other hand, at least one data field constituting the first prompt and the data field constituting the second prompt may be different.

For example, the first prompt may include a data field corresponding to story information and a data field corresponding to tags, and the second prompt may include a data field corresponding to story information, a data field corresponding to tags, and a data field corresponding to product categories.

Here, the second prompt may consist of at least one first dataset containing sample story information, tag information (e.g., theme information for a sample product category), and a sample product category, and a second dataset containing target story information and specific tags. Here, the specific tags constituting the second dataset may consist of tags selected on the user's terminal.

On the other hand, once the recommended product categories are obtained by the various methods described above, the product recommendation system and method according to the present invention may then proceed to obtain recommended products corresponding to the recommended product categories (S330) and provide recommended product information (S340).

The control unit 230 may, upon obtaining a recommended product category from the large language model 10, extract at least one recommended product corresponding to that category from the product recommendation engine 20.

The extracted recommended products may consist of products available for purchase online. As shown in Figures 5, 6, 7A, and 7B, information regarding the recommended products (540, 550, 616, 623, 714, 716) may be provided on the page that receives the user input.

More specifically, the recommended products may be configured to be available for purchase on at least one product sales website accessible via the internet, and when a recommended product is selected, product sales information with a link to the recommended product's sales page may be provided to the user's terminal.

On the other hand, as shown in Figure 13A, the control unit 230 may provide information about products belonging to a recommended product category, obtained from the user based on user input information and at least one of user-selected tags, to a specific area 1340 of the specific page 1310.

As shown in Figure 13A, when a user selects one of several recommended product categories (for example, "interior goods"), the control unit 230 may provide the user terminal with product information 1340 belonging to the selected recommended product category. The control unit 230 may also control the product recommendation engine 20 so that it extracts products corresponding to the recommended product category.

Furthermore, as shown in Figure 13C, when one of the multiple tags obtained from the user (for example, "great atmosphere" or "Instagrammable") is selected, the control unit 230 may provide the user terminal with product information 1395 belonging to the selected tag. The control unit 230 may also control the product recommendation engine 20 so that products corresponding to the selected tag are extracted from the product recommendation engine 20.

On the other hand, the control unit 230 may extract at least one of multiple products as a recommended product, in descending order of similarity, based on the similarity between the target story information and multiple products belonging to the recommended product category.

As shown in Figure 14, the control unit 230 may determine the similarity 1410 between the user input information (or story information 1421) and the product name 1431 extracted from the product recommendation engine, based on the embedded model structure. As shown in Figures 15(a), (b), and (c), it may then extract the top n products with the highest similarity as recommended products and provide the user terminal with product information (e.g., thumbnails) linked to the sales pages of these recommended products.

As shown in Figure 14, the embedding model may use the values of the hidden layer of a large language model (e.g., Hyperclova) 1422, 1432 to embed user input information and product names.

Furthermore, the present invention may also provide a service for creating messages (e.g., gift messages) corresponding to story information based on user input.

For example, users may be offered recommended products, and these products could be purchased directly from their device. Therefore, users can purchase recommended products as gifts corresponding to the story information they entered, as described above. Furthermore, there may be a need for users to create gift messages that correspond to their story information and purchased recommended products.

The control unit 230 may generate a message related to at least one of the story information and the recommended product when a recommended product is purchased, for example, when a product is purchased on the product sales page included in the product information for the recommended product, using the story information and information related to the recommended product.

Here, the control unit 230 may generate a message such that the target information included in the story information is information about the recipient of the recommended product, and the intention information included in the story information is information about the intention behind giving the recommended product.

Furthermore, the control unit 230 may generate the gift message 1600 using the large language model 10, as shown in Figures 16A and 16B.

As an example, as shown in Figure 16A, the large language model 10 may be configured to receive a prompt containing a story (or story information) 1610 and ordered items (or purchased items) 1620, and in response, receive a gift message (for example, an example message for a card) 1630.

The control unit 230 may identify a sample dataset containing data fields corresponding to story information, ordered items, and gift messages, along with their corresponding data values, in order to retrieve the gift message from the large language model 10. Alternatively, the prompt may be configured to include the sample dataset.

The prompt may further include a target dataset containing data fields corresponding to the target gift message to be generated. The target dataset may include target story information, data fields and corresponding data values for target ordered items, and data fields corresponding to the target gift message. Here, the data values for the data fields corresponding to the target gift message may be left as empty spaces.

Thus, the control unit 230 may configure prompts to generate a gift message based on user input information and the user's product purchase history, and input this message to the large language model 10. Alternatively, the control unit 230 may retrieve the gift message from the large language model 10.

As another example, as shown in Figure 16B, the prompt 1640 input to the large language model may include data fields corresponding to the intention (e.g., intention to give a gift or purchase a product), the giver (e.g., the person purchasing the product or giving the gift), the recipient (e.g., the person receiving the product or gift), the date (e.g., the date of purchase or the date the gift is given), the gift (e.g., the type of product purchased, the product name, etc.), and the card (e.g., the gift message).

As shown in the figure, prompt 1640 may include a sample dataset containing the aforementioned data fields and their corresponding data values. The control unit 230 may use the large language model 10 to retrieve the gift message (for example, corresponding to the "card" data field) contained in the target dataset 1650.

Thus, the control unit 230 may identify a sample dataset containing data fields corresponding to intention, giver, recipient, date, gift, and card, along with their corresponding data values. Alternatively, the prompt may be configured to include the sample dataset.

As shown in Figure 16B, the prompt 1640 may further include a target dataset 1650 containing data fields corresponding to the target gift message to be generated (e.g., a card). The target dataset may include data fields corresponding to intention, giver, recipient, date, and gift, along with their corresponding data values, and a data field corresponding to the target gift message (e.g., a card). Here, the data value of the data field corresponding to the target gift message may exist as an empty area.

Thus, the control unit 230 may configure a prompt to generate a gift message and input it to the large language model 10. Alternatively, the control unit 230 may retrieve the gift message from the large language model 10.

On the other hand, gift messages 1731 and 1732 obtained based on a large language model, or by other means, may be provided on the message input area 1730 of the gift message creation page 1700, as shown in Figures 17A and 17B.

The control unit 230 may create a gift message based on at least one of the story information, purchased items, and user information.

The content and tone of the gift message may be determined based on at least one of the following: story information, purchased items, and user information.

The control unit 230 may generate a gift message considering the recipient (the person receiving the gift), user information, the user's tone of voice, and the frequency of stamp usage. For example, the control unit 230 may consider the relationship between the user and the recipient, generating a polite message if the recipient is a superior (e.g., parents, teacher, boss), and a more casual message if the recipient is a friend. Furthermore, the control unit 230 may consider the user's age and gender, generating a gift message using commonly used language and slang within that group. In addition, the control unit 230 may analyze user-generated content (UGC) to identify the user's writing habits (tone of voice, stamp usage frequency, etc.) and generate a gift message that reflects these habits. By doing so, the control unit 230 can generate a gift message that appears as if it were directly created by the user.

Furthermore, the gift message creation page 1700 may include an input area 1720 for selecting the recipient of the message and the tone of the message. As shown in Figure 17B, the control unit 230 may generate and provide a message 1732 based on the information selected by the user in the input area, after the recipient information (e.g., a colleague) and tone information (e.g., a humorous message) have been selected.

Furthermore, on the gift message creation page 1700, the image 1710 to be included in the gift message may be selected.

On the other hand, although not shown in the diagram, the control unit 230 may provide the user terminal with a list of multiple sample gift messages corresponding to various targets and situations. The control unit 230 may also identify a gift message when at least one of the message lists is selected by the user terminal.

As described above, the product recommendation method and system according to the present invention can improve the convenience of gift purchases for users by providing services ranging from product recommendations and purchases to the generation of gift messages, based on the user's story information.

Furthermore, the product recommendation method and system according to the present invention can extract recommended products using a large-scale language model that takes prompts containing story information as input.

Here, the story information is structured to include information about the recipient of the recommended product and the purchaser's intention to buy the recommended product. Therefore, in this invention, it is possible to recommend products that reflect the user's intention to purchase the recommended product.

Thus, the product recommendation method and system according to the present invention can recommend products that match the user's story from a large language model by constructing prompts using story information that reflects the recipient of the recommended product and the intention behind giving the gift. Therefore, according to the present invention, the time required to select a gift can be reduced, and a product optimized according to the recipient and the intention behind the gift can be purchased.

On the other hand, the present invention described above may be implemented as a program that is executed by one or more processes on a computer and can be stored on a computer-readable medium (or recording medium).

Furthermore, the present invention described above may be implemented as computer-readable code or commands on a medium on which a program is recorded. That is, the present invention may be provided in the form of a program.

Furthermore, computer-readable media include all types of recording devices on which data read by computer systems is stored. Examples of computer-readable media include HDDs (Hard Disk Drives), SSDs (Solid State Disks), SSDs (Silicon Disk Drives), ROMs, RAMs, CD-ROMs, magnetic tapes, floppy disks, and optical data storage devices.

Furthermore, the computer-readable medium includes storage, including a server or cloud storage accessible by electronic devices via communication. In this case, the computer can download the program according to the present invention from the server or cloud storage via wired or wireless communication.

Furthermore, in the present invention, the aforementioned computer is a processor, i.e., a CPU (Central
An electronic device equipped with a Processing Unit (Central Processing Unit), of any type.

Furthermore, the above detailed description should not be interpreted restrictively in any way and should be considered illustrative. The scope of the present invention should be determined by a reasonable interpretation of the claims, and any modifications within the equivalent scope of the present invention are included.

(Note 1)
In product recommendation methods,
A step of identifying target story information that includes at least one of target information and target intent information,
A step of obtaining a recommended product category corresponding to the target story information based on sample story information which includes at least one of sample target information and sample intent information, and sample product categories which correspond to each of the sample story information,
A product recommendation method characterized by including the step of extracting at least one recommended product that matches the target story information from among the products belonging to the recommended product category.
(Note 2)
The step of obtaining the aforementioned recommended product categories is:
A step of constructing a prompt using the sample story information, the sample product category corresponding to the sample story information, and the target story information;
The step includes processing the aforementioned prompt as input to a pre-specified large language model,
The aforementioned recommended product categories are:
The product recommendation method according to Appendix 1, characterized in that it is generated as the output of the large language model.
(Note 3)
The product recommendation method according to Appendix 1, characterized in that at least one of the sample story information and the sample product category is obtained as output of the large language model by inputting prompts configured based on User Generated Content (UGC) created by multiple users on a page accessible via the Internet into the large language model.
(Note 4)
The aforementioned target story information is identified based on product recommendation requests received from specific user accounts.
In the step of identifying the target story information,
The product recommendation method according to Appendix 1, characterized in that it identifies the target story information using user input information received from the specified user account and user information stored on a specific server linked to the specified user account.
(Note 5)
The user input information is,
The aforementioned specific user account is received on a specific page provided to the user terminal to which it is logged in.
The aforementioned specific page is,
The product recommendation method according to Appendix 4, characterized by including a first input area for receiving target information corresponding to a recipient who will be given the recommended product as a gift, and a second input area for receiving target intention information corresponding to the intention to give the recommended product to the recipient.
(Note 6)
The step of identifying the target story information is:
The step of generating a text as target story information, which includes the target object information and target intent information received in the first input area and the second input area, using a natural language processing algorithm,
The product recommendation method according to Appendix 5, characterized in that the aforementioned text is included in the prompt.
(Note 7)
The user information includes at least one of the gender, age, residential area, and occupation of the specific user corresponding to the specific user account.
In the step of generating the aforementioned text,
The product recommendation method according to Appendix 6, characterized in that the text is generated such that the user information is included in the target story information.
(Note 8)
In the step of extracting the recommended products,
The product recommendation method according to Appendix 1, characterized in that, based on the similarity between the target story information and the multiple products belonging to the recommended product category, at least one of the multiple products is selected as the recommended product in order of decreasing similarity.
(Note 9)
The aforementioned multiple products are configured to be available for purchase on at least one product sales website accessible via the internet.
The product recommendation method according to Appendix 8, further comprising the step of providing the user terminal with product sales information linked to the sales page of the recommended product once the recommended product has been extracted.
(Note 10)
The product recommendation method according to Appendix 9, further comprising the step of generating a message about the recommended product using the target story information and information about the recommended product when the recommended product is purchased on the sales page.
(Note 11)
In the step of generating the aforementioned message,
The product recommendation method according to Appendix 10, characterized in that the message is generated such that the target information is included as information about the recipient to whom the recommended product will be given as a gift, and the target intention information is included as information about the intention to give the recommended product as a gift.
(Note 12)
In product recommendation methods,
The steps include receiving target story information from a user terminal, which includes information about the recipient of the gift and at least one of the intentions behind the gift,
A step of configuring a first prompt to be input into a pre-identified large language model using at least one dataset (dataset) containing sample story information and theme information of sample product categories, and the target story information,
Based on the first prompt, the step of obtaining multiple theme keywords corresponding to the target story information as the output of the large language model,
The steps include providing the user terminal with the plurality of theme keywords, and selecting at least one of the plurality of theme keywords on the user terminal,
The steps include configuring a second prompt that includes a specific theme keyword selected on the user terminal,
A product recommendation method characterized by including the step of obtaining a recommended gift category corresponding to the target story information from the large language model by processing the second prompt containing the specified theme keyword as input to the large language model.
(Note 13)
The second prompt is,
A first dataset comprising the aforementioned sample story information, the theme information of the sample product category, and the sample product category,
The product recommendation method according to Appendix 12, characterized in that it comprises the target story information and a second dataset including the specific theme keywords.
(Note 14)
The product recommendation method according to Appendix 13, further comprising the step of extracting at least one recommended product from among the products belonging to the recommended product category that corresponds to the target story information.
(Note 15)
In the product recommendation system,
A storage unit that stores sample story information including at least one of sample target information and sample intent information, and sample product categories corresponding to each of the sample story information,
A product recommendation system characterized by including a control unit that performs a process of obtaining a recommended product category corresponding to a target story information based on the sample story information and the sample product category corresponding to the sample story information, using a pre-identified large language model, and a process of extracting at least one recommended product from among the products belonging to the recommended product category that matches the target story information.
(Note 16)
A program executed by one or more processes in an electronic device,
A step of identifying target story information that includes at least one of target information and target intent information,
A step of obtaining a recommended product category corresponding to the target story information based on sample story information which includes at least one of sample target information and sample intent information, and sample product categories which correspond to each of the sample story information,
A program characterized by including a command to cause a computer to perform the steps of: extracting at least one recommended product from among the products belonging to the recommended product category that matches the target story information.

10 Large language model 200 Product recommendation system 220 Storage unit 230 Control unit 411, 412, 413 Sample dataset 414, 1111, 1220, 1650 Target dataset 410, 1010, 1040, 1110, 1210, 1640 Prompt 500, 610, 710, 800, 910, 920, 1310, 1350 Specific page 611, 612, 613, 616, 1732 Message 711 First input area 712 Second input area 913 Sample story information 1014a Text 1032 User information

Claims (10)

In a product recommendation method using a chatbot,
The chatbot provides the user terminal with a question message to identify the target story information.
The steps include identifying the target story information, which includes at least one of the target target information and the target intent information, based on the user message received from the user terminal,
The steps include obtaining a recommended product category that matches the target story information as a response to the user message from a large language model,
The product recommendation engine performs the steps of: extracting recommended products corresponding to the recommended product categories obtained from the large language model;
A method for recommending products using a chatbot, comprising the step of providing the recommended products to the user terminal via the chatbot. In the step of providing the aforementioned question message to the user terminal,
A product recommendation method using a chatbot according to claim 1, characterized in that a message is configured to guide the user terminal to obtain additional information necessary for constructing the target story information. In the step of providing the aforementioned question message to the user terminal,
A product recommendation method using a chatbot according to claim 2, characterized in that the question message is provided to the user terminal multiple times in order to induce the aforementioned additional information. The question message provided to the user terminal multiple times in order to induce the aforementioned additional information is:
A product recommendation method using a chatbot according to claim 3, characterized in that it includes questions related to at least one of the target recipients who will receive the recommended product as a gift and target intent information corresponding to the purpose of giving the recommended product as a gift. The step of obtaining the aforementioned recommended product categories is:
The steps include obtaining multiple theme keywords from the aforementioned large-scale language model that correspond to the target story information and are related to the theme of the product category,
The steps include providing the user terminal with the multiple theme keywords via the chatbot,
The user terminal selects one of the multiple theme keywords,
A product recommendation method using a chatbot according to claim 1, characterized by comprising the step of obtaining the recommended product category from the large language model that reflects information about any of the selected theme keywords. The first prompt input to the aforementioned large language model is:
Including the aforementioned target story information,
The product recommendation method using a chatbot according to claim 5, characterized in that the plurality of theme keywords are obtained from the large language model as output for the first prompt. The second prompt input to the aforementioned large language model is:
The target story information and any of the selected theme keywords from the plurality of theme keywords include,
The product recommendation method using a chatbot according to claim 6, characterized in that the recommended product category is obtained from the large language model as output for the second prompt. In the step of providing the recommended products to the user terminal via the chatbot,
The product recommendation method using a chatbot according to claim 7, characterized in that the sales page of the recommended product and the product sales information linked to it are both provided to the user terminal. In a product recommendation system using a chatbot,
The aforementioned chatbot provides the user's device with a question message to identify the target story information.
Based on the user message received from the user terminal, the target story information, which includes at least one of the target information and the target intent information, is identified.
From the large language model, as a response to the user message, a recommended product category that matches the target story information is obtained.
The product recommendation engine extracts recommended products corresponding to the recommended product categories obtained from the large language model,
A product recommendation system using a chatbot, which provides the recommended products to the user terminal via the chatbot. The chatbot provides the user's device with a question message to identify the target story information.
The steps include identifying the target story information, which includes at least one of the target target information and the target intent information, based on the user message received from the user terminal,
The steps include obtaining a recommended product category that matches the target story information as a response to the user message from a large language model,
The product recommendation engine performs the steps of: extracting recommended products corresponding to the recommended product categories obtained from the large language model;
A program characterized by including a command to cause a computer to perform the steps of providing the recommended products to the user terminal via the chatbot.