JP7650850B2 - Information processing device, information processing method, and program - Google Patents

Description

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

Conventionally, systems that recommend items to users are known. Patent Document 1 discloses a technology that outputs recommendation information about products offered by a store according to the psychological tendencies of a user obtained from the user's responses to a questionnaire.

JP 2022-026681 A

However, in order to estimate a user's psychological tendencies, a large number of survey responses are required. Therefore, for example, in new stores and stores with a small number of customers, it is difficult to obtain a large number of survey responses, making it difficult to use the above technology.

Therefore, the present invention has been made in consideration of these points, and aims to provide an information processing device, information processing method, and program that can output information that takes into account the psychological characteristics of the user.

The information processing device according to the first aspect of the present invention has a model generation unit that generates a personality inference model that outputs a personality vector indicating the characteristics of a user when a browsing history of multiple items by a user is input by machine learning using teacher data generated based on a first browsing history indicating one or more first items viewed by a first user in a first domain that sells multiple first items, psychological identification information for identifying the psychological characteristics of the first user, and the characteristics of the first items purchased by the first user; a calculation unit that calculates an item recommendation score that becomes higher as the probability of purchase by the second user increases by calculating using a function indicating the correlation between the personality vector output by the personality inference model and a second item vector indicating the characteristics of the multiple second items by inputting a second browsing history indicating one or more second items viewed by a second user in a second domain that sells multiple second items into the personality inference model; and an output unit that outputs information based on the item recommendation score calculated by the calculation unit.

The information processing device may further include an update unit that updates a first weighting factor used when converting the personality vector to a predetermined dimension and a second weighting factor used when converting the second item vector to the predetermined dimension so that the item recommendation score corresponding to the second item purchased by the second user in the second browsing history input to the personality inference model is maximized each time the calculation unit executes the process of calculating the item recommendation score, and the calculation unit may calculate the item recommendation score by performing an operation using a function indicating the correlation between the personality vector converted to the predetermined dimension using the first weighting factor and the second item vector converted to the predetermined dimension using the second weighting factor.

The calculation unit may correct the personality vector by performing a calculation using a function indicating the correlation between a user preference vector indicating the preferences of the second user identified by the second browsing history and the personality vector, and calculate the item recommendation score by performing a calculation using a function indicating the correlation between a corrected personality vector, which is the personality vector after the correction, and the second item vector.

The information processing device further includes an update unit that updates a first weighting factor used when converting the personality vector into a predetermined dimension, a second weighting factor used when converting the second item vector into the predetermined dimension, and a third weighting factor used when converting the user preference vector into the predetermined dimension, so that the item recommendation score corresponding to the second item purchased by the second user in the second browsing history input to the personality inference model is maximized each time the calculation unit executes the process of calculating the item recommendation score, and the calculation unit may correct the personality vector by performing a calculation using a function indicating the relevance between the personality vector converted into the predetermined dimension using the first weighting factor and the user preference vector converted into the predetermined dimension using the third weighting factor, and calculate the item recommendation score by performing a calculation using a function indicating the relevance between the corrected personality vector and the second item vector converted into the predetermined dimension using the second weighting factor.

The output unit may output information indicating the second item corresponding to an item recommendation score that has a relatively high numerical value among the multiple item recommendation scores calculated by the calculation unit.

The information processing method according to the second aspect of the present invention includes the steps of: generating a personality inference model that outputs a personality vector indicating the characteristics of a user when a browsing history of multiple items by a user is input by machine learning using teacher data generated based on a first browsing history indicating one or more first items viewed by a first user in a first domain that sells multiple first items, psychological identification information for identifying the psychological characteristics of the first user, and the characteristics of the first items purchased by the first user, executed by a computer; inputting a second browsing history indicating one or more second items viewed by a second user in a second domain that sells multiple second items into the personality inference model, and calculating an item recommendation score that becomes higher as the probability of purchase by the second user increases by calculating using a function indicating the correlation between the personality vector output by the personality inference model and the second item vector indicating the characteristics of the multiple second items; and outputting information based on the calculated item recommendation score.

The program according to the third aspect of the present invention causes a computer to function as: a model generation unit that generates a personality inference model that outputs a personality vector that indicates the characteristics of a user when a browsing history of multiple items by a user is input, by performing machine learning using teacher data generated based on a first browsing history indicating one or more first items viewed by a first user in a first domain that sells multiple first items, psychological identification information for identifying the psychological characteristics of the first user, and the characteristics of the first items purchased by the first user; a calculation unit that calculates an item recommendation score that becomes higher as the probability of purchase by the second user increases by calculating using a function that indicates the relationship between the personality vector output by the personality inference model by inputting a second browsing history indicating one or more second items viewed by a second user in a second domain that sells multiple second items, and a second item vector that indicates the characteristics of the multiple second items; and an output unit that outputs information based on the item recommendation score calculated by the calculation unit.

The present invention has the effect of being able to output information that takes into account the psychological characteristics of the user.

FIG. 1 is a diagram for explaining an overview of an information processing system. FIG. 13 is a diagram illustrating an example of a questionnaire. FIG. 2 is a diagram showing a configuration of an information processing device 3. 13 is a flowchart showing the flow of processing executed by the information processing device 3.

[Overview of Information Processing System S]
1 is a diagram for explaining an overview of an information processing system S. The information processing system S is a system used to support sales of items in a domain. A domain is, for example, a shopping site that sells items on the Internet. The information processing system S has a first domain server 1, a second domain server 2, and an information processing device 3.

The first domain server 1 is a server used to provide a first sales service that sells multiple first items in the first domain. The first domain server 1 manages the first item data, the first browsing history of a first user who is a customer of the first sales service, the first purchasing history of the first user, and psychological identification information of the first user.

The first item data is, for example, data in which the identification information (ID) of the first item is associated with an explanation of the first item posted on a web page displaying details of the first item. The first browsing history is information indicating one or more first items viewed by a first user in a first domain, and the ID of the first user is associated with the ID of the first item viewed by the first user. The first purchasing history is information indicating a first item purchased by a first user in a first domain (hereinafter referred to as a "first purchased item"), and the ID of the first user is associated with the ID of the first purchased item by the first user.

The psychological identification information is information for identifying the psychological characteristics of the first user. The psychological characteristics are, for example, indicated by the Big Five, which classifies a person's personality into five factors (openness, conscientiousness, extraversion, agreeableness, and neuroticism). The psychological identification information is, for example, the results of the first user's responses to a questionnaire, or the numerical values of each of the five factors in the Big Five calculated based on the results of the first user's responses to a questionnaire.

Figure 2 is a schematic diagram of an example of a questionnaire. As shown in Figure 2, the questionnaire includes questions and options indicated by numerical values, and the first user answers the questions in the questionnaire by selecting an option that applies to the content of the question from among multiple options for each question. The numerical values of each of the five factors in the Big Five are calculated, for example, by tallying up the options selected by the first user for each question using a predetermined method.

The psychological identification information is not limited to the results of questionnaire responses, etc., but may be any information that allows the calculation of the numerical values of each of the five factors in the Big Five. For example, it may be information for identifying the physical activity of the first user obtained from an acceleration sensor, or biometric information of the first user, such as heart rate obtained from a heart rate sensor.

Returning to FIG. 1, the second domain server 2 is a server used to provide a second sales service that sells multiple second items in the second domain. The second domain is, for example, a domain that sells the same type of second items as the first domain. The second domain does not have psychological specific information and is, for example, a new domain. The second domain may be a domain that is smaller in scale or has fewer customers than the first domain. The second domain server 2 manages the second item data and the second browsing history of the second user who is a customer of the second sales service.

The second item data is, for example, data in which the ID of the second item is associated with an explanation of the second item that is posted on a web page that displays details of the second item. The second browsing history is information indicating one or more second items viewed by a second user in a second domain, and the ID of the second user is associated with the ID of the second item viewed by the second user.

The information processing device 3 is a device that provides information to the domain to support the sale of items, and is, for example, a server. The information processing device 3 is capable of communicating with the first domain server 1 and the second domain server 2 via the network.

The processing executed by the information processing system S will be described below. First, the processing corresponding to the first domain will be described. The information processing device 3 acquires first domain data from the first domain server ((1) in FIG. 1). The first domain data includes first item data, a first browsing history, psychological identification information, and a first purchasing history.

The information processing device 3 generates training data based on the first domain data ((2) in FIG. 1). The training data includes the first browsing history, psychological identification information, and characteristics of the first purchased item based on the first purchasing history.

The information processing device 3 generates a personality inference model M by machine learning using the teacher data ((3) in FIG. 1). The personality inference model M is a model, for example a neural network, that outputs a personality vector when a browsing history of multiple items by a user is input. The personality vector is information that indicates the characteristics of the user. The user characteristics include, for example, at least one of the user's preferences identified based on the input browsing history, the user's psychological characteristics identified based on psychological identification information, or the characteristics of items purchased by the user.

Next, the process corresponding to the second domain will be described. In the process corresponding to the second domain, the personality inference model M generated in the process corresponding to the first domain is transferred to output information for supporting the sale of the second item. First, the information processing device 3 acquires second domain data from the second domain server ((4) in FIG. 1). The second domain data includes second item data and second browsing history.

The information processing device 3 applies the transferred personality inference model M and inputs the second browsing history to the personality inference model M, causing the personality inference model M to output a personality vector ((5) in FIG. 1). The information processing device 3 calculates a second item vector indicating the characteristics of the second item based on the second item data included in the second domain data ((6) in FIG. 1). The second item vector is, for example, a numerical representation of the number of occurrences (score) of each of multiple words appearing in the description of the second item, or the number of occurrences of words corresponding to each of the attributes (elements) of multiple words. The attributes of a word are properties that indicate the characteristics of the item indicated by the word.

The information processing device 3 calculates an item recommendation score corresponding to the second item by performing an operation using a function indicating the correlation between the personality vector and the second item vector ((7) in FIG. 1). The item recommendation score is a numerical value indicating the degree to which the second item is recommended, and for example, the higher the numerical value, the higher the likelihood that the second user will purchase the second item. Details of the function will be described later. The information processing device 3 then outputs information based on the calculated item recommendation score ((8) in FIG. 1).

In this way, even if the second domain cannot obtain the psychological specific information of the second user, the information processing system S can output information corresponding to the purchasing preferences of the second user, taking into account the psychological characteristics of the second user, by reusing the psychological specific information of the first user.
The configuration of the information processing device 3 will be described below.

[Configuration of information processing device 3]
3 is a diagram showing the configuration of the information processing device 3. The information processing device 3 has a communication unit 31, a storage unit 32, and a control unit 33. The control unit 33 has a first acquisition unit 331, a data generation unit 332, a model generation unit 333, a second acquisition unit 334, a calculation unit 335, an output unit 336, and an update unit 337.

The communication unit 31 is an interface for connecting to a network, and includes, for example, a communication controller. The storage unit 32 is a storage medium such as a ROM (Read Only Memory), a RAM (Random Access Memory), or a hard disk. The storage unit 32 stores a program executed by the control unit 33.

The control unit 33 is, for example, a CPU (Central Processing Unit). The control unit 33 executes a program stored in the memory unit 32, thereby functioning as a first acquisition unit 331, a data generation unit 332, a model generation unit 333, a second acquisition unit 334, a calculation unit 335, an output unit 336, and an update unit 337. The processes performed by each of the first acquisition unit 331, the data generation unit 332, the model generation unit 333, the second acquisition unit 334, the calculation unit 335, the output unit 336, and the update unit 337 will be described later.

[Description of model generation process]
First, a process in which the information processing system S generates the personality inference model M will be described. The first acquisition unit 331 acquires, from the first domain server 1, first domain data including first item data, a first browsing history, psychology identification information, and a first purchase history.

The data generating unit 332 generates teacher data based on the first domain data. The teacher data includes the first browsing history of each of the multiple first users, the psychological identification information of each of the multiple first users, and the characteristics of the first purchased items of each of the multiple first users. Specifically, the teacher data is data in which, for each first user who answered the questionnaire and purchased a first item, a first user preference vector based on the first browsing history of the first user, a first user psychological vector based on the psychological identification information of the first user, and a first purchased item vector indicating the characteristics of the first purchased item purchased by the first user included in the first purchase history are associated with each first user.

The data generating unit 332 calculates, for each first user, a first user preference vector indicating the preferences of the first user identified by the first browsing history of the first user. Specifically, the data generating unit 332 first calculates, for each first item included in the first browsing history, a first item vector indicating the characteristics of the first item based on the number of occurrences of words included in the description associated with the first item in the first item data. For example, a word group indicating the characteristics of the item is determined in advance, and the data generating unit 332 first calculates the first item vector by tallying up the number of occurrences of words included in the description associated with the first item in the first item data for each word in the word group for each attribute of the word. The data generating unit 332 may calculate the first item vector by integrating multiple attributes or deleting some attributes and converting the tallying result into a predetermined dimension. The predetermined dimension is, for example, the same dimension as the dimension of the first user psychological vector and is set in advance in the information processing device 3.

Then, the data generating unit 332 calculates a first user preference vector based on each calculated first item vector. The data generating unit 332 calculates the first user preference vector, for example, by averaging or summing values corresponding to each element (attribute) included in multiple first item vectors corresponding to the browsing history of the first user. When the data generating unit 332 calculates the first user preference vector in this manner, a first user preference vector indicating the characteristics of the first items viewed by the first user is obtained.

The data generation unit 332 calculates the first user psychological vector based on the psychological identification information. For example, the data generation unit 332 calculates the first user psychological vector by converting the numerical values of each of the five factors identified by the psychological identification information into a predetermined dimension.

The data generation unit 332 refers to the first item data and calculates the first purchased item vector. Specifically, the data generation unit 332 calculates the first purchased item vector by tallying up the number of times each word in the word group appears in the description associated with the first purchased item in the first item data, and converting the tallying result into a predetermined dimension. When the first purchase history includes multiple first purchased items, the data generation unit 332 may calculate the average of the first purchased item vectors of each first purchased item as the first purchased item vector.

Although an example in which the data generation unit 332 generates teacher data associated with each first user has been described above, this is not limiting. For example, the data generation unit 332 may generate teacher data associated with each predetermined group. The predetermined group may be, for example, gender, age group, or area of residence. In this case, the data generation unit 332 generates teacher data for each predetermined group in which, for example, the average of the first user preference vector, the average of the first user psychological vector, and the average of the first purchased item vector are associated.

The model generation unit 333 generates a personality inference model M by machine learning using the teacher data. In this way, the information processing device 3 can generate a personality inference model M that has learned the association between the psychological characteristics of the user, the characteristics of the first user, and the characteristics of the first item purchased. The model generation unit 333 stores the generated personality inference model M in the storage unit 32.

[Description of Item Recommendation Score Calculation Process]
Next, a process of calculating an item recommendation score by the information processing system S will be described. The second acquisition unit 334 acquires second domain data from the second domain server 2. The second domain data includes a second browsing history and second item data.

The calculation unit 335 calculates the item recommendation score by performing an operation using a first function that indicates the correlation between a second personality vector output by the personality inference model M by inputting the second browsing history into the personality inference model M and a second item vector that indicates the characteristics of a plurality of second items. The second personality vector is a personality vector that corresponds to the second user. The calculation unit 335 calculates the item recommendation score by, for example, executing the following five steps.

First, in the first step, the calculation unit 335 calculates a second user preference vector indicating the preferences of the second user identified by the second browsing history. Specifically, the calculation unit 335 first calculates, for each second item included in the second browsing history, a second item vector indicating the characteristics of the second item based on the number of occurrences of words included in the description associated with the second item in the second item data. Then, the calculation unit 335 calculates the second user preference vector by averaging or summing values corresponding to each element (attribute) included in multiple second item vectors corresponding to the browsing history of the second user.

In the second step, the calculation unit 335 acquires the personality inference model M stored in the memory unit 32, and acquires the second personality vector output by the personality inference model M by inputting the calculated second user preference vector into the personality inference model M.

In the third step, the calculation unit 335 converts the second personality vector into a predetermined dimension using the first weighting coefficient. The first weighting coefficient is a coefficient used when converting the second personality vector into a predetermined dimension, and is a coefficient tuned so as to maximize the item recommendation score corresponding to the purchased second item. For example, the first weighting coefficient is provided for each element of the second personality vector, and the calculation unit 335 converts the second personality vector into a predetermined dimension by multiplying, for each element of the second personality vector, the numerical value of the element by the first weighting coefficient corresponding to the element, and averaging or summing the multiplication results corresponding to the multiple elements. Details of the process of tuning the first weighting coefficient will be described later.

In the fourth step, the calculation unit 335 calculates a second item vector for each second item. The calculation unit 335 converts each calculated second item vector into a predetermined dimension using a second weighting factor. The second weighting factor is a factor used when converting the second item vector into a predetermined dimension, and is a factor tuned so that the item recommendation score corresponding to the purchased second item is maximized. For example, the second weighting factor is provided for each element of the second item vector, and the calculation unit 335 multiplies, for each element of the second item vector, the numerical value of the element by the second weighting factor corresponding to the element, and converts the second item vector into a predetermined dimension by averaging or summing the multiplication results corresponding to the multiple elements. Details of the process of tuning the second weighting factor will be described later.

Then, in the fifth step, the calculation unit 335 calculates the item recommendation score by performing an operation using a first function that indicates the correlation between the second personality vector converted into a predetermined dimension using the first weighting coefficient and the second item vector converted into a predetermined dimension using the second weighting coefficient. The first function is, for example, the DOT function.

For example, suppose the second personality vector converted to a predetermined dimension is [a, b, c, d], and the second item vector corresponding to a second item converted to a predetermined dimension is [e, f, g, h]. In this case, the calculation unit 335 calculates the item recommendation score using the DOT function [a*e+b*f+c*g+d*h].

The first function may be a function other than the DOT function, and may be, for example, a CONCAT function or an ADD function. For example, when the first function is a CONCAT function, the calculation unit 335 calculates a one-dimensional vector using a predetermined weighting factor for the CONCAT function [a, b, c, d, e, f, g, h] that connects the second personality vector [a, b, c, d] and the second item vector [e, f, g, h], thereby calculating the item recommendation score. For example, when the first function is an ADD function, the calculation unit 335 calculates a one-dimensional vector using a predetermined weighting factor for the ADD function [a+e, b+f, c+g, d+h] that adds the second personality vector [a, b, c, d] and the second item vector [e, f, g, h] for each dimension, thereby calculating the item recommendation score.

The calculation unit 335 may further use the second user preference vector to calculate the item recommendation score. Specifically, the calculation unit 335 first corrects the second personality vector by performing a calculation using a second function indicating the association between the second user preference vector indicating the preference of the second user identified by the second browsing history and the second personality vector. The second function is, for example, an ADD function or a MEAN function. For example, when the second function is a MEAN function, the calculation unit 335 calculates the item recommendation score by calculating a one-dimensional vector using a predetermined weighting coefficient for the MEAN function [(a+e)/2, (b+f)/2, (c+g)/2, (d+h)/2] obtained by averaging each dimension in the second personality vector [a, b, c, d] and the second item vector [e, f, g, h].

Then, the calculation unit 335 calculates the item recommendation score by performing an operation using a first function that indicates the correlation between the corrected personality vector, which is the second personality vector after correction, and the second item vector.

More specifically, the calculation unit 335 first corrects the second personality vector by performing a calculation using a second function indicating the relationship between the second personality vector converted to a predetermined dimension using the first weighting coefficient and the second user preference vector converted to a predetermined dimension using the third weighting coefficient. The third weighting coefficient is a third weighting coefficient used when converting the second user preference vector to a predetermined dimension, and is a coefficient tuned so that the item recommendation score corresponding to the purchased second item is maximized. For example, the third weighting coefficient is provided for each element of the second user preference vector, and the calculation unit 335 multiplies, for each element of the second user preference vector, the numerical value of the element by the third weighting coefficient corresponding to the element, and converts the second user preference vector to a predetermined dimension by averaging or summing the multiplication results corresponding to the multiple elements. Details of the process of tuning the third weighting coefficient will be described later.

The first weighting factor and the third weighting factor vary according to the degree of difference between the first domain and the second domain, for example. The difference between the first domain and the second domain is, for example, the difference between the features of the first item sold by the first domain and the features of the second item sold by the second domain, the difference between the behavior of a first user browsing a web page in the first domain and the behavior of a second user browsing a web page in the second domain, etc. For example, the greater the degree of difference between the first domain and the second domain, the smaller the value of the second personality vector converted by the first weighting factor, and the larger the value of the user preference vector converted by the third weighting factor. On the other hand, the smaller the degree of difference between the first domain and the second domain, the larger the value of the second personality vector converted by the first weighting factor, and the smaller the value of the user preference vector converted by the third weighting factor.

The calculation unit 335 then calculates the item recommendation score by performing an operation using a first function that indicates the association between the corrected personality vector and the second item vector that has been converted to a predetermined dimension using the second weighting coefficient. The information processing device 3 can calculate the item recommendation score that takes into account the degree of difference between the first domain and the second domain by further calculating the item recommendation score using the second user preference vector.

The output unit 336 outputs information based on the item recommendation scores calculated by the calculation unit 335. The output unit 336 transmits, for example, information based on the item recommendation scores to the second domain server 2. The output unit 336 outputs, for example, each item recommendation score calculated by the calculation unit 335 in association with the ID of the second item.

The output unit 336 may output information indicating a second item corresponding to an item recommendation score with a relatively high numerical value among the multiple item recommendation scores calculated by the calculation unit 335. The output unit 336 may, for example, output information indicating a second item corresponding to an item recommendation score with the highest numerical value among the multiple item recommendation scores calculated by the calculation unit 335. The output unit 336 may, for example, output information indicating a plurality of second items corresponding to a predetermined number of item recommendation scores (e.g., the top three item recommendation scores) in ascending numerical value. In this way, the operator of the second domain can ascertain the second items that are highly likely to be purchased by the second user.

Although an example in which the information processing device 3 outputs information based on the item recommendation score corresponding to the second domain has been described above, this is not limiting. For example, the information processing device 3 may output information based on the item recommendation score corresponding to the first domain.

In this case, first, the second acquisition unit 334 acquires the first browsing history and the first item data from the first domain server 1. The calculation unit 335 inputs the first browsing history into the personality inference model M, and calculates an item recommendation score by performing an operation using a first function that indicates the correlation between the first personality vector output by the personality inference model M and the first item vector that indicates the characteristics of multiple first items. Then, the output unit 336 outputs information based on the item recommendation score calculated by the calculation unit 335. In this way, the business operator of the first domain can identify the first items that are highly likely to be purchased by the first user.

[Description of Tuning Process]
Next, a process of tuning the weighting coefficient by the information processing system S will be described. The second acquisition unit 334 acquires the second item data, the second browsing history, and information indicating the second purchased item from the second domain server 2. Note that the second acquisition unit 334 is not limited to the information indicating the second purchased item, and may acquire information in which information indicating whether or not the second user has purchased the second item is labeled for each of a plurality of second items sold in the second domain.

The update unit 337 updates each weighting factor based on the second item data, the second browsing history, and the characteristics of the second purchased item acquired by the second acquisition unit 334. Specifically, each time the calculation unit 335 executes the process of calculating the item recommendation score, the update unit 337 updates the first weighting factor and the second weighting factor so that the item recommendation score corresponding to the second purchased item purchased by the second user in the second browsing history input to the personality inference model M is maximized.

For example, first, the update unit 337 changes the first weighting coefficient and the second weighting coefficient each time the calculation unit 335 executes the process of calculating the item recommendation score. Then, the update unit 337 updates the combination of the changed first weighting coefficient and the second weighting coefficient to a combination of the first weighting coefficient and the second weighting coefficient that maximizes the item recommendation score corresponding to the second purchased item. Also, for example, the update unit 337 may update the first weighting coefficient and the second weighting coefficient so that, among the item recommendation scores corresponding to each of the multiple second items, the item recommendation score corresponding to the second purchased item is higher than the item recommendation scores corresponding to the second items other than the second purchased item.

When the calculation unit 335 further calculates the item recommendation score using the second user preference vector, the update unit 337 updates the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient each time the calculation unit 335 executes the process of calculating the item recommendation score so that the item recommendation score corresponding to the second purchased item purchased by the second user of the second browsing history input to the personality inference model M is maximized.

When the calculation unit 335 calculates the item recommendation score using a predetermined weighting factor, the update unit 337 further updates the predetermined weighting factor each time the calculation unit 335 executes the process of calculating the item recommendation score, so that the item recommendation score corresponding to the second purchased item purchased by the second user in the second browsing history input to the personality inference model M is maximized.

In the above, an example has been described in which the calculation unit 335 calculates the item recommendation score using the personality inference model M stored in the storage unit 32, i.e., the personality inference model M generated by the model generation unit 333, but this is not limiting. For example, the calculation unit 335 may calculate the item recommendation score using the personality inference model M tuned for the second domain by the update unit 337. Specifically, the update unit 337 fine-tunes the personality inference model M stored in the storage unit 32 using the second item data.

[Processing of information processing device 3]
Next, a description will be given of the flow of processing executed by the information processing device 3. Fig. 4 is a flowchart showing the flow of processing executed by the information processing device 3. This flowchart starts when the first acquisition unit 331 acquires first domain data including first item data, first browsing history, psychology specification information, and first purchase history from the first domain server 1 (S1).

The data generation unit 332 generates training data in which the first user preference vector, the first user psychological vector, and the first purchased item vector are associated based on the first domain data (S2). The model generation unit 333 generates a personality inference model M by machine learning using the training data (S3).

The second acquisition unit 334 acquires second domain data including second item data and second browsing history from the second domain server 2 (S4). The calculation unit 335 calculates a second user preference vector based on the second browsing history (S5). The calculation unit 335 calculates the second item vector based on the second item data (S6).

The calculation unit 335 inputs the second user preference vector into the personality inference model M, and performs an operation using a first function that indicates the correlation between the second personality vector output by the personality inference model M and the second item vector to calculate an item recommendation score (S7). Then, the output unit 336 outputs information based on the item recommendation score calculated by the calculation unit 335 (S8).

[Effects of this embodiment]
As described above, the information processing system S generates a personality inference model M by machine learning using teacher data generated based on the first browsing history, psychological identification information, and characteristics of the first purchased item, and outputs information based on an item recommendation score calculated by calculating a personality vector output by the personality inference model M by inputting the second browsing history into the generated personality inference model M and a function indicating the relevance between the second item vector based on the second item data. In this way, even if the second domain cannot obtain psychological identification information of the second user, the information processing system S can output information in which the psychological characteristics of the second user are taken into consideration by reusing the psychological identification information of the first user.

Furthermore, this invention will make it possible to contribute to Goal 9 of the United Nations' Sustainable Development Goals (SDGs), which is "Build resilient infrastructure, promote inclusive and sustainable industrialization, and promote innovation and infrastructure."

Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments, and various modifications and changes are possible within the scope of the gist of the invention. For example, all or part of the device can be configured by distributing or integrating functionally or physically in any unit. In addition, new embodiments resulting from any combination of multiple embodiments are also included in the embodiments of the present invention. The effect of the new embodiment resulting from the combination also has the effect of the original embodiment.

Reference Signs List 1 First domain server 2 Second domain server 3 Information processing device 31 Communication unit 32 Storage unit 33 Control unit 331 First acquisition unit 332 Data generation unit 333 Model generation unit 334 Second acquisition unit 335 Calculation unit 336 Output unit 337 Update unit S Information processing system

Claims (7)

a model generation unit that performs machine learning using training data generated based on a first browsing history indicating one or more first items browsed by a first user in a first domain that sells a plurality of first items, psychological identification information for identifying a psychological characteristic of the first user, and characteristics of the first items purchased by the first user, to generate a personality inference model that outputs a personality vector indicating characteristics of the user when a browsing history of a plurality of items by the user is input;
a calculation unit that calculates an item recommendation score, the higher the value, the higher the probability that the second user will purchase the item, by inputting a second browsing history indicating one or more second items browsed by a second user in a second domain that sells a plurality of second items into the personality inference model and performing a calculation using a function indicating the correlation between the personality vector output by the personality inference model and a second item vector indicating characteristics of the plurality of second items;
an output unit that outputs information based on the item recommendation score calculated by the calculation unit;
An information processing device having the above configuration. an update unit that updates a first weighting factor used in converting the personality vector into a predetermined dimension and a second weighting factor used in converting the second item vector into the predetermined dimension so that the item recommendation score corresponding to the second item purchased by the second user of the second browsing history input to the personality inference model is maximized every time the calculation unit executes a process of calculating the item recommendation score,
the calculation unit calculates the item recommendation score by performing an operation using a function indicating a correlation between the personality vector converted into the predetermined dimension using the first weighting coefficient and the second item vector converted into the predetermined dimension using the second weighting coefficient.
The information processing device according to claim 1 . the calculation unit corrects the personality vector by performing a calculation using a function indicating a correlation between a user preference vector indicating a preference of the second user specified by the second browsing history and the personality vector, and calculates the item recommendation score by performing a calculation using a function indicating a correlation between a corrected personality vector, which is the personality vector after the correction, and the second item vector.
The information processing device according to claim 1 . an update unit that updates a first weighting factor used when converting the personality vector into a predetermined dimension, a second weighting factor used when converting the second item vector into the predetermined dimension, and a third weighting factor used when converting the user preference vector into the predetermined dimension, so that the item recommendation score corresponding to the second item purchased by the second user of the second browsing history input to the personality inference model is maximized every time the calculation unit executes the process of calculating the item recommendation score,
the calculation unit corrects the personality vector by performing a calculation using a function indicating a relationship between the personality vector converted into the predetermined dimension using the first weighting coefficient and the user preference vector converted into the predetermined dimension using the third weighting coefficient, and calculates the item recommendation score by performing a calculation using a function indicating a relationship between the corrected personality vector and the second item vector converted into the predetermined dimension using the second weighting coefficient.
The information processing device according to claim 3 . the output unit outputs information indicating the second item corresponding to the item recommendation score having a relatively high value among the plurality of item recommendation scores calculated by the calculation unit.
The information processing device according to claim 1 . The computer executes
a step of generating a personality inference model that outputs a personality vector indicating the characteristics of a user when a browsing history of a plurality of items by a user is input, by performing machine learning using training data generated based on a first browsing history indicating one or more first items viewed by a first user in a first domain that sells a plurality of first items, psychological identification information for identifying a psychological characteristic of the first user, and characteristics of the first items purchased by the first user;
A step of inputting a second browsing history indicating one or more second items viewed by a second user in a second domain that sells a plurality of second items into the personality inference model, and calculating an item recommendation score, the higher the value, the higher the probability that the second user will purchase the item, by performing a calculation using a function indicating the correlation between the personality vector output by the personality inference model and a second item vector indicating characteristics of the plurality of second items;
outputting information based on the calculated item recommendation scores;
An information processing method comprising the steps of: Computer,
a model generation unit that generates a personality inference model by performing machine learning using training data generated based on a first browsing history indicating one or more first items browsed by a first user in a first domain that sells a plurality of first items, psychology identification information for identifying a psychological characteristic of the first user, and characteristics of the first items purchased by the first user, when a browsing history of a plurality of items by the user is input, and outputs a personality vector indicating characteristics of the user;
a calculation unit that calculates an item recommendation score, the numerical value of which increases as the probability of purchase by the second user increases, by inputting a second browsing history indicating one or more second items browsed by a second user in a second domain that sells a plurality of second items into the personality inference model and performing a calculation using a function indicating a correlation between the personality vector output by the personality inference model and a second item vector indicating characteristics of the plurality of second items; and an output unit that outputs information based on the item recommendation score calculated by the calculation unit.
A program to function as a

Images