Information processing system, information processing method, and program

Abstract

Easier and more accurate output of representative personalities for groups of users [Solution] The information processing system classifies multiple users into multiple clusters based on the attribute values ​​of multiple types of attributes stored in association with each of the multiple users, calculates the contribution of each of the multiple types of attributes to the classification of each of the multiple users into a target cluster, which is one of the multiple clusters, calculates a representative attribute value that represents the target cluster for that type of attribute based on the attribute values ​​of the multiple types of attributes stored in association with each of the multiple users classified into the target cluster, and outputs information indicating the representative personality of users belonging to the target cluster based on the representative attribute value and the contribution.

Description

[Technical Field]

[0001] The present invention relates to an information processing system, an information processing method, and a program. [Background technology]

[0002] There is a system that uses big data of users to analyze the representative personalities of a group of multiple users who meet given conditions. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] LINE Yahoo Japan Corporation. “Persona - DS.INSIGHT”, [online], [accessed February 28, 2024], Internet<https: / / ds.yahoo.co.jp / service / insight / persona.html> [Non-patent document 2] J.A. Pardo, "AI-powered Personas: Bringing Personas to life through LLMs," [online], [accessed February 28, 2024], Internet <https: / / medium.com / @josetecangas / ai-powered-personas-bringing-personas-to-life-through-llms-d1246da02858> Summary of the Invention [Problem to be solved by the invention]

[0004] Until now, categorizing users into groups and analyzing the personalities of users in those groups has relied on the intuition of data scientists. This has made it difficult to, for example, categorize groups using unknown classification methods or to find attributes that characterize groups from a variety of attributes. Furthermore, there have been limitations to the accuracy of the analysis.

[0005] An object of the present disclosure is to provide a technology that outputs a representative personality of a group of users more easily and with higher accuracy. [Means for solving the problem]

[0006] (1) An information processing system including: a classification means for classifying a plurality of users into a plurality of clusters based on the attribute values ​​of each of a plurality of types of attributes stored in association with each of the plurality of users; a contribution calculation means for calculating the contribution of each of the plurality of types of attributes to classification into a target cluster, which is one of the plurality of clusters, for each of the plurality of users classified into the target cluster; a representative acquisition means for acquiring a representative attribute value representing the target cluster for the type of attribute based on the attribute values ​​of each of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster; and an output means for outputting information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and the contribution.

[0007] (2) In (1), the information processing system further includes a training means for training a machine learning model using training data including attribute values ​​of multiple types of attributes for each of the multiple users and correct answer data indicating the cluster into which each of the multiple users is classified, and the contribution calculation means calculates the contribution of each of the multiple types of attributes to the classification of each of the multiple users into the target cluster based on the learned machine learning model and the attribute values ​​of the multiple types of attributes for each of the multiple users classified into the target cluster.

[0008] (3) In (1) or (2), the information processing system further includes an importance calculation means for calculating the importance of each of the multiple types of attributes with respect to the classification of the multiple users into the target cluster based on the contribution calculated for the multiple users classified into the target cluster, and the output means outputs information indicating the representative personality of the users belonging to the target cluster based on the representative attribute value and the importance.

[0009] (4) In (3), the information processing system further includes an attribute selection means for selecting a portion of the plurality of types of attributes based on the importance, and the output means outputs information indicating a representative personality of users belonging to the target cluster based on a representative attribute value of the selected attribute.

[0010] (5) In (3) or (4), the importance calculation means calculates the average value of the contribution rates calculated for the multiple users classified into the target cluster as the importance for each of the multiple types of attributes.

[0011] (6) In (3) or (4), the information processing system, wherein the importance calculation means calculates, for each of the multiple types of attributes, a value indicating the correlation between the contribution rate and the attribute value for multiple users classified into the target cluster as the importance rate.

[0012] (7) In (3) or (6), the representative acquisition means calculates an average value of the contribution rates calculated for the plurality of users for each of the plurality of types of attributes, selects some of the plurality of users as one or more representative users based on the average value and the contribution rates of the plurality of users classified into the target cluster for at least some of the plurality of types of attributes, and generates representative attribute values ​​representing the target cluster based on the attribute values ​​of the one or more representative users for at least some of the plurality of types of attributes.

[0013] (8) In any of (1) to (7), the representative acquisition means acquires a relative value indicating the relative relationship between the representative attribute value calculated for the multiple types of attributes and an overall representative attribute value calculated for a group consisting of the multiple clusters, and the output means outputs information indicating a representative personality of a user belonging to the target cluster based on the contribution degree, the acquired relative value, and the representative attribute value.

[0014] (9) In (8), the representative acquisition means acquires a relative value indicating the probability that multiple users classified into the multiple clusters belong to a category indicated by the representative attribute value of the target cluster for an attribute indicating a category among the multiple types of attributes.

[0015] (10) In the information processing system of (8) or (9), the output means outputs information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value of an attribute selected from the plurality of types of attributes based on the contribution rate and the acquired relative value.

[0016] (11) In any of (1) to (10), the output means inputs instructions to a language model, the instructions including at least some representative attribute values ​​of the multiple types of attributes for the representative cluster, to generate a sentence indicating a personality, and outputs information indicating a representative personality of a user belonging to the target cluster based on the output of the language model in response to the input.

[0017] (12) In (11), the information processing system further includes an attribute selection means for selecting a portion of the plurality of types of attributes based on the contribution degree, and the output means inputs instructions including representative attribute values ​​of the selected attributes to a language model to generate a sentence indicating personality, and outputs information indicating the representative personality of users belonging to the target cluster based on the output of the language model in response to the input.

[0018] (13) In any of (8) to (10), the output means inputs instructions to a language model, the instructions including relative values ​​of at least some of the multiple types of attributes for the representative cluster, for generating a sentence indicating personality, and outputs information indicating the representative personality of a user belonging to the target cluster based on the output of the language model in response to the input.

[0019] (14) In any of (1) to (13), the output means inputs instructions based on the representative attribute value and the contribution degree, to a language model, instructions for generating a sentence to generate an image indicating personality, inputs instructions for generating an image based on the output of the language model in response to the input, to an image generation model, and outputs information including the image output from the image generation model as information indicating the representative personality of users belonging to the target cluster.

[0020] (15) An information processing method including the steps of: classifying a plurality of users into a plurality of clusters based on the attribute values ​​of each of a plurality of types of attributes stored in association with each of the plurality of users; calculating the contribution of each of the plurality of types of attributes to classification into a target cluster, which is one of the plurality of clusters, for each of the plurality of users classified into the target cluster; obtaining a representative attribute value for the type of attribute that represents the target cluster based on the attribute values ​​of each of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster; and outputting information indicating a representative personality of users belonging to the target cluster based on the representative attribute value and the contribution.

[0021] (16) A program for causing a computer to function as a classification means for classifying a plurality of users into a plurality of clusters based on the attribute values ​​of each of a plurality of types of attributes stored in association with each of the plurality of users; a contribution calculation means for calculating the contribution of each of the plurality of types of attributes to the classification of each of the plurality of users into a target cluster, which is one of the plurality of clusters; a representative acquisition means for acquiring a representative attribute value representing the target cluster for the type of attribute based on the attribute values ​​of each of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster; and an output means for outputting information indicating the representative personality of users belonging to the target cluster based on the representative attribute value and the contribution. [Effects of the Invention]

[0022] The present invention allows for the representative personalities of a group of users to be output more easily and accurately. [Brief explanation of the drawings]

[0023] [Figure 1] FIG. 1 is a diagram illustrating an example of elements related to an information processing system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram showing functions realized by the information processing system. [Figure 3] FIG. 10 is a diagram illustrating an example of data stored in an attribute database. [Figure 4] FIG. 2 is a flow chart schematically illustrating processing of the information processing system. [Figure 5] FIG. 10 is a diagram illustrating an example of a clustering result. [Figure 6] FIG. 10 is a diagram illustrating an example of a calculated contribution degree. [Figure 7] FIG. 10 is a flowchart illustrating an example of processing by a cluster attribute determination unit. [Figure 8] FIG. 10 is a flowchart illustrating an example of a process for calculating a representative attribute value and an index. [Figure 9]FIG. 10 is a diagram illustrating an example of a calculated representative attribute value and an index. [Figure 10] FIG. 10 is a flowchart illustrating another example of the processing of the cluster attribute determination unit. [Figure 11] FIG. 10 is a flowchart showing an example of processing by a personality output unit. [Figure 12] FIG. 10 is a diagram showing an example of a first instruction text. [Figure 13] FIG. 10 is a diagram showing an example of an explanatory sentence generated based on a first command text. [Figure 14] FIG. 10 is a diagram showing an example of a second instruction text. [Figure 15] FIG. 10 is a diagram showing an example of an instruction sentence generated based on a second instruction text. DETAILED DESCRIPTION OF THE INVENTION

[0024] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Duplicate descriptions of components with the same reference numerals will be omitted.

[0025] FIG. 1 is a diagram illustrating an example of elements related to an information processing system 1 according to an embodiment of the present invention. Based on instructions from an administrator, the information processing system 1 acquires attribute information of multiple users and classifies the multiple users into multiple clusters. The information processing system 1 acquires information describing the personalities of representative users of each cluster using a large-scale language model system 2 and an image generation system 3, and outputs the information to the administrator. The administrator may receive instructions and output by operating an input / output device included in the information processing system 1, or may receive instructions and output via a computer (not shown) that can communicate with the information processing system 1.

[0026] The large-scale language model system 2 includes a general-purpose large-scale language model implemented by one or more computers. The large-scale language model system 2 receives instructions from the information processing system 1, inputs the instructions into the large-scale language model, and passes the output obtained by inputting the instructions to the information processing system 1. The instructions are in text format and are also called prompts. Hereinafter, instructions in text format will also be referred to as instruction text. This general-purpose large-scale language model is trained using data from a wide range of fields. The large-scale language model system 2 may provide a service such as ChatGPT (registered trademark), for example.

[0027] The image generation system 3 includes an image generation model implemented by one or more computers. The image generation system 3 receives instructions from the information processing system 1 and passes the output obtained by inputting the instructions into the image generation model to the information processing system 1. The instructions are in text format and will hereinafter also be referred to as instruction text. The image generation model may be, for example, a machine learning model based on a diffusion model. The image generation system 3 may provide services such as DALL-E (registered trademark) or Stable Diffusion. The image generation system 3 may be launched using the same API as the large-scale language model system 2.

[0028] Hereinafter, simply referring to a "large-scale language model" refers to a large-scale language model included in the large-scale language model system 2, and referring to an "image generation model" refers to an image generation model included in the image generation system 3. The large-scale language model system 2 may be provided within the information processing system 1. In this embodiment, the information processing system 1 inputs an instruction to the large-scale language model to generate some information, and obtains the output of the large-scale language model as that information. Hereinafter, inputting an instruction to generate that information into the large-scale language model will also be referred to as instructing the large-scale language model to generate information.

[0029] The information processing system 1 includes one or more computers (e.g., server computers). The information processing system 1 includes one or more processors 11, one or more storages 12, and one or more communication units 13. The information processing system 1 may include multiple computers each including one or more processors 11, storages 12, and communication units 13, or may include a single computer having one or more processors 11 and storages 12. The information processing system 1 may be implemented on one or more virtual server or container platforms.

[0030] The processor 11 operates according to a program (also referred to as an instruction code) stored in the storage 12. The processor 11 also controls the communication unit 13. The processor 11 includes, for example, a CPU (Central Processing Unit), and may further include a GPU (Graphic Processing Unit) and an NPU (Neural Processing Unit). The program may be provided via the Internet or the like, or may be provided by being stored in a computer-readable storage medium such as a flash memory or a DVD-ROM.

[0031] The storage 12 is composed of memory elements such as RAM and flash memory, and external storage devices such as a hard disk drive (HDD) and a solid state drive (SSD). The storage 12 stores the above programs. The storage 12 also stores information input from the processor 11 and the communication unit 13 and calculation results.

[0032] The communication unit 13 is a communication interface, such as a network interface card, that communicates with other devices. The communication unit 13 is configured with an integrated circuit, an antenna, a communication terminal, etc. that realize a wireless LAN or a wired LAN. Under the control of the processor 11, the communication unit 13 inputs information received from other devices via a network to the processor 11 or the storage 12, and transmits the information to other devices.

[0033] The hardware configuration of the information processing system 1 is not limited to the above example. For example, the information processing system 1 may include a device for reading a computer-readable information storage medium (e.g., an optical disk drive or a memory card slot) or a device for inputting and outputting data to and from an external device (e.g., a USB port). The external device may be an input device or an output device.

[0034] Next, the functions provided by the information processing system 1 will be described. FIG. 2 is a block diagram showing the functions realized by the information processing system 1. The information processing system 1 functionally includes a classification unit 51, a classification training unit 52, a classification model 53, a contribution calculation unit 54, a cluster attribute determination unit 55, a personality output unit 56, and an attribute database 70. The cluster attribute determination unit 55 functionally includes an importance calculation unit 57, a representative acquisition unit 58, and an attribute selection unit 59. The classification unit 51, the classification training unit 52, the classification model 53, the contribution calculation unit 54, the cluster attribute determination unit 55, the personality output unit 56, and the attribute database 70 are realized by the processor 11 executing programs corresponding to each function stored in the storage 12 and controlling the communication unit 13, etc.

[0035] The attribute database 70 is a database that stores information (attribute data) about the attributes (hereinafter also referred to as "characteristics") of multiple users who are the targets of processing by the information processing system 1. The attribute database 70 may be a database that centrally manages attribute information. The attribute database 70 may collect attribute information managed by multiple other systems from those systems and store it in the storage 12.

[0036] FIG. 3 is a diagram showing an example of data stored in the attribute database 70. The attribute database 70 stores attribute data for each of a plurality of users in association with the user. FIG. 3 shows the attribute data stored in the attribute database 70 in a table format. The "UID," "age," "income," and "purchase_skincare" columns store attribute values ​​(attribute values) for the user's identification information, age, income, and skin care product purchase amount, respectively. In FIG. 3, the attribute data is associated with the user based on the user's identification information. Although FIG. 3 shows only three attributes, the actual number of attributes may be greater.

[0037] The classification unit 51 classifies the multiple users into multiple clusters based on the attribute values ​​of multiple types of attributes stored in the attribute database 70 in association with each of the multiple users.

[0038] The classification training unit 52 trains the classification model 53 using training data including attribute values ​​of multiple types of attributes for each of multiple users stored in the attribute database 70 and correct answer data indicating the cluster into which each of the multiple users is classified.

[0039] The contribution calculation unit 54 calculates the contribution of multiple types of attributes to classification into each of the target clusters for each of the multiple users classified into a target cluster, which is one of the multiple clusters. The contribution calculation unit 54 may calculate the contribution of each of the multiple users to the value of the probability (output of the classification model 53) that the user belongs to each of the multiple clusters. In the latter process, the contribution may also be calculated for users not classified into the target cluster.

[0040] The contribution calculation unit 54 calculates the contribution of each of the multiple types of attributes to classification into the target cluster for each of the multiple users classified into the target cluster, based on the trained classification model 53 and the attribute values ​​of the multiple types of attributes for each of the multiple users classified into the target cluster.

[0041] The cluster attribute determination unit 55 calculates a representative attribute value representing the target cluster for each type of attribute based on the attribute values ​​of multiple types of attributes stored in association with each of the multiple users classified into the target cluster. The cluster attribute determination unit 55 also calculates an index value based on the representative attribute value and the attribute values ​​in a group consisting of multiple clusters. The index value indicates whether the representative attribute value is common in multiple clusters.

[0042] Here, the target cluster is a cluster that is the target of processing, and by sequentially specifying a plurality of clusters as target clusters, the above processing may be performed on the plurality of clusters. The same applies to the processing described below.

[0043] The importance calculation unit 57 calculates the importance of each of the multiple attributes with respect to the classification of the multiple users into the target cluster based on the contribution degrees calculated for at least the multiple users classified into the target cluster. The importance calculation unit 57 may calculate, as the importance, an average value of the contribution degrees calculated for at least the multiple users classified into the target cluster for each of the multiple attributes. Furthermore, the importance calculation unit 57 may calculate, as the importance, a value indicating a correlation based on the contribution degrees and attribute values ​​of the multiple users classified into the target cluster for each of the multiple attributes.

[0044] The representative acquisition unit 58 calculates a representative attribute value representing the target cluster for each type of attribute based on the attribute values ​​of multiple types of attributes stored in association with each of the multiple users classified into the target cluster. The cluster attribute determination unit 55 also acquires, as index values, relative values ​​indicating the relative relationship between the representative attribute value and an overall representative attribute value calculated for a group consisting of multiple clusters for multiple types of attributes expressed numerically. The cluster attribute determination unit 55 acquires, as index values, relative values ​​indicating the likelihood that multiple users classified into multiple clusters belong to the category indicated by the representative attribute value of the target cluster for multiple types of attributes expressed as categories.

[0045] The representative acquisition unit 58 may generate the representative attribute value by the following process. The representative acquisition unit 58 calculates the average value of the contribution degrees calculated for multiple users for each of multiple types of attributes. The representative acquisition unit 58 selects some of the multiple users as one or more representative users based on the average value for at least some of the multiple types of attributes and the contribution degrees of multiple users classified into the target cluster. The representative acquisition unit 58 generates a representative attribute value representing the target cluster based on the attribute values ​​of one or more representative users for at least some of the multiple types of attributes. Here, the users targeted for calculating the average contribution degrees may be all users or may be users classified into the target cluster.

[0046] The attribute selection unit 59 selects some of the multiple types of attributes based on the importance. The attribute selection unit 59 may select some of the multiple types of attributes based on at least one of the importance and the index value. Here, the multiple types of attributes may be classified into multiple groups, and the attribute selection unit 59 may select no more than a predetermined number of attributes for each group based on at least one of the importance and the index value.

[0047] The personality output unit 56 outputs information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and importance. The importance is calculated by the importance calculation unit 57 based on the contribution degree. More specifically, the personality output unit 56 outputs information indicating a representative personality of a user belonging to the target cluster based on at least the representative attribute value of the attribute selected based on the importance. The information indicating the personality includes at least one of a description of the representative user and a generated image of the representative user.

[0048] The personality output unit 56 inputs to the large-scale language model an instruction to generate a sentence indicating a personality, the instruction including a representative attribute value of at least a portion (e.g., selected attributes) of multiple types of attributes for the representative cluster, and outputs information indicating a representative personality of a user belonging to the target cluster based on the output of the large-scale language model in response to the input. The personality output unit 56 may input to the large-scale language model an instruction including a relative value of at least a portion (e.g., selected attributes) of multiple types of attributes, or may input to the large-scale language model an instruction including the representative attribute value and the relative value.

[0049] The personality output unit 56 inputs, to the large-scale language model, instructions based on the representative attribute value and the contribution degree, for generating a sentence for generating an image showing the personality. The personality output unit 56 inputs, to the image generation model, instructions for generating an image based on the output of the large-scale language model in response to the input, and outputs information including the image output from the image generation model as information showing the representative personality of users belonging to the target cluster.

[0050] Next, a more detailed description will be given of the processing of the information processing system 1. FIG.

[0051] First, the classification unit 51 clusters a plurality of users based on attribute values ​​of a plurality of types of attributes for each of the plurality of users stored in the attribute database 70 (S101). The plurality of users is classified into a plurality of clusters by clustering. The clustering method may be, for example, a Gaussian Mixture Model. The number of clusters may be set in advance by the user, or may be determined by the classification unit 51 based on the number of users, etc. The plurality of users may also be classified by other known clustering methods.

[0052] Fig. 5 is a diagram showing an example of the clustering result. In the example of Fig. 5, the "UID" and "cluster" columns are the identification information of the user and the identification information of the classified cluster, respectively.

[0053] The classification training unit 52 trains the classification model 53 based on the clustering results (S102). The training data used in the training includes attribute values ​​for multiple types of attributes for multiple users and ground truth data indicating the clusters into which each of the multiple users is classified. The classification model 53 is, for example, a machine learning model that classifies data using a decision tree such as LightGBM. The classification model 53 may also be a machine learning model that classifies data based on other techniques. When attribute values ​​for multiple attributes for a user are input, the trained classification model 53 outputs a value indicating the probability that the user belongs to each of multiple clusters.

[0054] The contribution calculation unit 54 calculates the contribution of each of the multiple attributes of the multiple users for each cluster using the trained classification model 53 (S103). Here, the contribution calculation unit 54 calculates the contribution using a technique that explains the prediction result of the trained classification model 53.

[0055] More specifically, the contribution calculation unit 54 calculates the Shapley Value (also called the SHAP value) as the contribution using a method called SHAP (SHapley Additive exPlanations). In SHAP, the marginal contribution of a feature is calculated by applying a method for calculating the Shapley Value as the marginal contribution of a player in game theory to the variation from the average predicted value. The library for calculating SHAP is publicly available as an open source library, and the details of the calculation will not be explained here. The Shapley Value has the property that the sum of the Shapley Values ​​for all features for a given user corresponds to the difference between the probability of belonging to a cluster output from the classification model 53 and a reference value. The Shapley Value is calculated for each combination of user, attribute, and cluster.

[0056] Fig. 6 is a diagram showing an example of calculated contribution rates. In the example of Fig. 5, the columns "Model output," "Reference value," "SHAP_age," "SHAP_income," and "SHAP_purchase_skincare" respectively show a value indicating the probability of being classified into the cluster shown in Cluster, a reference value indicating that probability, the contribution rate of the age attribute, the contribution rate of the income attribute, and the contribution rate of the purchase amount of skin care products. As can be seen from Fig. 6, the contribution rate is calculated for each combination of user, cluster, and attribute.

[0057] 6, the contribution may be calculated by other methods that explain the prediction results of the trained classification model 53. Also, the contribution may be calculated for each combination of a user and an attribute, only for the cluster to which the user belongs.

[0058] Once the contribution degree is calculated, the cluster attribute determining unit 55 calculates a representative attribute value and an index value for each of the multiple attributes that describe each cluster (S104).

[0059] The calculation of the representative attribute value and the index value will be further explained. Fig. 7 is a flow diagram showing an example of the processing of the cluster attribute determination unit 55. The processing shown in Fig. 7 is executed for each cluster. The cluster that is the target of the processing is denoted as the target cluster. The method shown in Fig. 7 is called the "Top Shapley Approach."

[0060] First, the importance calculation unit 57 calculates the average value Av(ft) of the contributions of all users for each of a plurality of attributes (S201). ft indicates the attribute to be processed (here, the average value is calculated). The average value Av(ft) corresponds to the importance of the attribute ft for classifying a plurality of users into a target cluster.

[0061] For each attribute group, the attribute selection unit 59 extracts attributes with the mth average value Av(Ft) from the multiple attributes belonging to that group (S202). Here, it is assumed that multiple types of attributes are classified into multiple groups. The multiple groups may include, for example, at least some of lifestyle, life stage, work, tendency, finance, service availability time, purchasing tendency, and way of thinking. In S202, the attribute selection unit 59 selects up to m attributes for each group. Here, it is assumed that the number m of attributes selected for each group is predetermined. The value of m may be, for example, 3 for the lifestyle group and 1 for the work group.

[0062] Then, the representative obtaining unit 58 calculates a representative attribute value and an index value in the target cluster for the extracted attribute (S203).

[0063] The calculation of the representative attribute value and the index value will be further described below. Fig. 8 is a flow diagram showing an example of the process of calculating the representative attribute value and the index, and is a diagram explaining the process of S203 in more detail.

[0064] First, the representative obtaining unit 58 selects one of the extracted attributes (S251), and determines whether the type of the attribute value of the selected attribute is a category (S252).

[0065] If the type of attribute value is category (Y in S252), the representative acquisition unit 58 sets the most common attribute value in the target cluster as the representative attribute value of that attribute (S253). Then, the representative acquisition unit 58 sets an index value depending on whether the proportion of users who have the set attribute value exceeds a threshold value overall (S254). Specifically, the representative acquisition unit 58 sets the index value to 1 if the proportion of users who have the set attribute value exceeds the threshold value, and sets the index value to 0 if not. The proportion of users who have the set attribute value is calculated for all users, including clusters different from the target cluster. The threshold value may be a value greater than or equal to 0.5 and less than 1.

[0066] If the type of the attribute value is a numeric value (N in S252), the representative acquisition unit 58 sets the average of the attribute values ​​in the target cluster as the representative attribute value of that attribute (S255).The representative acquisition unit 58 then sets the value obtained by dividing the representative attribute value by the average of the attribute values ​​of all users as the index value (S256).

[0067] Once the index value is set, the representative acquisition unit 58 determines whether any unprocessed attributes exist (S257). If any unprocessed attributes exist (Y in S257), the representative acquisition unit 58 selects one of the unprocessed attributes (S258) and repeats the processing from S252 onwards. If no unprocessed attributes exist (N in S257), the processing in Fig. 8 ends. This loop calculates a representative attribute value and an index value for each extracted attribute.

[0068] Once the representative attribute value and index value have been calculated, the attribute selection unit 59 determines, from the extracted multiple attributes, an attribute to be input to the large-scale language model based on the index value (S204). The attribute selection unit 59 may determine, as an attribute to be input, an attribute whose index value is within a predetermined range. The predetermined range may be a range excluding around 1. This is because when the index value for a certain attribute is around 1, the index value indicates that the attribute is not sufficiently different from other clusters.

[0069] Note that the process of S204 may be omitted, and the calculated representative attribute value and index value may be directly determined as data to be input to the large-scale language model. Also, instead of extracting attributes in S202, importance for input to the large-scale language model may be output. A similar effect can be achieved by instructing the large-scale language model to select attributes with high importance. Of course, limiting the attributes to be input as in S202 improves the quality of the output from the large-scale language model. Note that the calculation and use of index values ​​may not be necessary.

[0070] Fig. 9 is a diagram showing an example of a calculated representative attribute value and index. Fig. 9 shows an example of text in which a set of attribute names, representative attribute values, and index values ​​is listed. "Original Value" is the representative attribute value, and "Index Value" is the index value. In the example of Fig. 9, when the type of attribute value is category, the representative acquisition unit 58 outputs the category name as the representative attribute value.

[0071] The calculation of the representative attribute value and the index value may be performed by other methods. Fig. 10 is a flow diagram showing another example of the processing of the cluster attribute determination unit 55. The processing shown in Fig. 10 is performed for each cluster, and the cluster that is the target of the processing is referred to as the target cluster. The method shown in Fig. 10 is called the "Core Member Approach."

[0072] First, the importance calculation unit 57 calculates the average value Av(ft) of the contributions of all users for each of a plurality of attributes (S301). The users for which the average contributions are to be calculated may be users classified into the target cluster, or all users including those in other clusters.

[0073] Next, the importance calculation unit 57 calculates the correlation between the attribute value and the contribution for each of the multiple attributes in the target cluster (S302). The correlation for a certain attribute is the sum of the product of the contribution for that attribute and (attribute value - average value) for all users included in the target cluster. The correlation corresponds to the importance of the attribute for classifying multiple users into the target cluster. A positive correlation indicates that users with a larger attribute value for that attribute have a higher probability of belonging to the cluster.

[0074] The representative acquisition unit 58 sets the variable i to 1 and assigns all users in the target cluster to the user group (S303). The representative acquisition unit 58 calculates the difference d between the contribution of each user in the user group and the average value Av(ft) for the attribute with the i-th largest absolute value of the average value Av(ft) among the attributes with positive correlation (S304). The representative acquisition unit 58 also calculates the number N of users to be selected (S305). The number N of users is calculated by rounding down the decimal point of the quotient obtained by dividing the number of users belonging to the current user group by the absolute value of the average value Av(ft).

[0075] The representative acquisition unit 58 assigns the users having the Nth highest absolute value of the difference d among the users belonging to the user group to a new user group (S306). If the number of users belonging to the new user group is 100 or more (Y in S307), the representative acquisition unit 58 increments i by 1 (S308) and repeats the processing from S304 onwards for the new user group.

[0076] On the other hand, if the number of users belonging to the user group is less than 100 (N in S307), the attribute selection unit 59 extracts attributes with a positive correlation as targets for calculating representative attribute values ​​(S309), and the representative acquisition unit 58 calculates representative attribute values ​​and index values ​​for the target attributes based on the attribute values ​​of the users belonging to the user group (S310). The representative attribute values ​​and index values ​​may be calculated by using the attribute values ​​of the users belonging to the user group instead of the attribute values ​​of the users included in the target cluster in FIG. 8. Note that after the processing of S310, the attribute selection unit 59 may perform the processing of S204, i.e., limiting the attributes based on the index values.

[0077] In the method shown in Figure 10, correlation is calculated as importance based on contribution, and the attributes to be input to the large-scale language model are limited by the correlation. In the method shown in Figure 7, correlation may also be used as importance to limit the attributes. Furthermore, in the method shown in Figure 10, the representative attribute value is calculated not from all users included in the target cluster, but from the attribute values ​​of one or more users included in the target cluster who have similar contributions to the important attributes. This reduces noise among users belonging to the cluster, and the representative attribute value is calculated from users who have higher cluster characteristics.

[0078] When the representative attribute values ​​and index values ​​to be input to the large-scale language model are determined in S104, the personality output unit 56 outputs a description and an image of the personality of the user representing each cluster based on the representative attribute values ​​and index values ​​(S105). The large-scale language model is used to generate the description, and an image generation model is used to generate the image.

[0079] The process of S105 will be described in detail below with reference to a flowchart of FIG.

[0080] First, the personality output unit 56 creates a first command text that causes the large-scale language model to generate a representative user description (S401). The first command text is a prompt to be input to the large-scale language model, and includes a command sentence and information on the attributes to be input. When using a large-scale language model specialized for generating descriptions, the command sentence does not need to exist. The information on the attributes to be input includes at least one of a representative attribute value and an index value. The first command text is then input to the large-scale language model, and its output (description) is obtained (S402).

[0081] FIG. 12 is a diagram showing an example of a first instruction text. Strictly speaking, FIG. 12 shows a template of the first instruction text, and the {behavior_data} portion is replaced with text listing a set of attribute names, representative attribute values, and index values ​​as shown in FIG. 9. In FIG. 12, the first instruction text is written in English, but it may be written in another language. Furthermore, the languages ​​used in examples of text input to a large-scale language model or image generation model and text output from a large-scale language model, which will be described later, may also be different. Furthermore, the first instruction text does not need to include an index value.

[0082] 12, the first command text includes text explaining the representative attribute value and index value, and text specifying the format of the description to be output. More specifically, the first command text includes instructions to output an age group instead of the age itself, instructions not to output internal information such as clusters, and instructions not to use an attribute that indicates a low probability when generating a description.

[0083] Fig. 13 is a diagram showing an example of an explanatory sentence generated based on the first command text. In Fig. 13, the explanatory sentence is output in English, but the personality output unit 56 may output the explanatory sentence in another language by adjusting the first command text.

[0084] When the explanatory text is acquired, the personality output unit 56 creates a second instruction text that causes the large-scale language model to generate situation text for generating an image of the representative user (S403).

[0085] Figure 14 is a diagram showing an example of the second command text. In actuality, the explanatory text shown in Figure 13 is set in area 81. When generating an image, it is better to specify a detailed scene, and accuracy of the image is not required compared to the explanatory text. Therefore, the second command text includes an instruction to output in the situation text clothing, colors, background, etc. that can be associated with the explanatory text.

[0086] Here, the second command text may include an instruction to generate situation text from the description, or may include an instruction to generate situation text directly from at least one of the representative attribute value and the index value. Furthermore, the generation of the situation text may be performed in multiple stages. For example, in the first stage, the personality output unit 56 may instruct the large-scale language model to generate text indicating a personality representative of the user, different from the description, from the representative attribute value and the index value, and in the second stage, the personality output unit 56 may instruct the large-scale language model to generate situation text from the generated text.

[0087] When the second command text is created, the personality output unit 56 inputs the second command text into the large-scale language model and obtains the output (situation text) (S404).The personality output unit 56 generates an instruction sentence based on the situation text, inputs the instruction sentence into the image generation model, and obtains the output (generated image) (S405).

[0088] 15 is a diagram showing an example of an instruction generated based on the second command text. In area 82, situation text is embedded.

[0089] Then, the personality output unit 56 outputs the explanatory text and image to the administrator as information indicating the representative personalities of the users belonging to the target cluster (S406).

[0090] 10, a description and an image describing a representative user of a cluster are generated from statistical data such as representative attribute values, and these are output to an administrator. Here, the personality output unit 56 does not necessarily need to use a large-scale language model or an image generation model. For example, the personality output unit 56 may simply output a list of representative attribute values ​​or index values ​​of selected attributes to an administrator.

[0091] Conventionally, user analysis using big data, especially segmentation, has relied on the intuition of administrators. The information processing system 1 in this embodiment can find user clusters based on various attributes that indicate user characteristics. Furthermore, the information processing system 1 can detect attributes that characterize clusters by analyzing the clustering results using the contribution of attributes based on explainable AI technology. Using this, the information processing system 1 can acquire and output information indicating the personalities of representative users included in the clusters.

[0092] Although a large-scale language model is used in this embodiment, there are no particular limitations on the scale of its implementation and the number of parameters. The present invention can be applied to machine learning models (language models) that handle natural language. [Explanation of symbols]

[0093] 1 Information processing system, 2 Large-scale language model system, 3 Image generation system, 11 Processor, 12 Storage, 13 Communication unit, 51 Classification unit, 52 Classification training unit, 53 Classification model, 54 Contribution calculation unit, 55 Cluster attribute determination unit, 56 Personality output unit, 57 Importance calculation unit, 58 Representative acquisition unit, 59 Attribute selection unit, 70 Attribute database, 81, 82 Area.

Claims

1. a classification means for classifying the plurality of users into a plurality of clusters based on attribute values ​​of a plurality of types of attributes stored in association with each of the plurality of users; a contribution calculation means for calculating a contribution of each of the plurality of types of attributes to classification into a target cluster, which is one of the plurality of clusters, for each of the plurality of users classified into the target cluster; a representative acquisition means for acquiring a mode or an average value of each attribute value of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster as a representative attribute value representing the target cluster for the attribute of that type; an output means for outputting information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and the contribution degree; An information processing system including:

2. 2. The information processing system according to claim 1, The method further includes a training means for training a machine learning model using training data including attribute values ​​of a plurality of types of attributes for each of the plurality of users and correct answer data indicating a cluster into which each of the plurality of users is classified, the contribution calculation means calculates, for each of the plurality of users classified into the target cluster, a contribution of each of the plurality of types of attributes to classification into the target cluster, based on the trained machine learning model and attribute values ​​of the plurality of types of attributes for each of the plurality of users classified into the target cluster; Information processing system.

3. 2. The information processing system according to claim 1, further comprising importance calculation means for calculating an importance of each of the plurality of types of attributes with respect to classification of the plurality of users into the target cluster based on the contribution degree calculated for the plurality of users classified into the target cluster; the output means outputs information indicating a representative personality of the user belonging to the target cluster based on the representative attribute value and the importance. Information processing system.

4. 4. The information processing system according to claim 3, further comprising an attribute selection means for selecting a part of the plurality of types of attributes based on the importance; the output means outputs information indicating a representative personality of the users belonging to the target cluster based on the representative attribute value of the selected attribute. Information processing system.

5. 4. The information processing system according to claim 3, the importance calculation means calculates, for each of the plurality of types of attributes, an average value of the contribution degrees calculated for the plurality of users classified into the target cluster as the importance; Information processing system.

6. 4. The information processing system according to claim 3, the importance calculation means calculates, for each of the plurality of types of attributes, a value indicating a correlation based on the contribution degree and the attribute value of the plurality of users classified into the target cluster, as the importance; Information processing system.

7. A classification means for classifying a plurality of users into a plurality of clusters based on attribute values ​​of a plurality of types of attributes stored in association with each of the plurality of users; a contribution calculation means for calculating a contribution of each of the plurality of types of attributes to classification into a target cluster, which is one of the plurality of clusters, for each of the plurality of users classified into the target cluster; a representative acquisition means for selecting one or more representative users representing the target cluster from the plurality of users based on the attribute values ​​of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster, and acquiring representative attribute values ​​representing the target cluster based on the attribute values ​​of the representative users for at least some of the plurality of types of attributes; an output means for outputting information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and the contribution degree; An information processing system including:

8. 8. The information processing system according to claim 7, further comprising importance calculation means for calculating an importance of each of the plurality of types of attributes with respect to classification of the plurality of users into the target cluster based on the contribution degree calculated for the plurality of users classified into the target cluster; The representative acquisition means calculating an average value of the contribution degrees calculated for the plurality of users for each of the plurality of types of attributes; selecting some of the users as one or more representative users based on the average values ​​and the contributions of the users classified into the target cluster for at least some of the attributes; generating a representative attribute value representing the target cluster based on the attribute values ​​of the one or more representative users for at least a portion of the plurality of types of attributes; the output means outputs information indicating a representative personality of the user belonging to the target cluster based on the representative attribute value and the importance. Information processing system.

9. 2. The information processing system according to claim 1, the representative acquisition means acquires a relative value indicating a relative relationship between the representative attribute value calculated for the plurality of types of attributes and an overall representative attribute value calculated for a group consisting of the plurality of clusters; the output means outputs information indicating a representative personality of the user belonging to the target cluster based on the contribution degree, the acquired relative value, and the representative attribute value. Information processing system.

10. 10. The information processing system according to claim 9, the representative acquisition means acquires relative values ​​indicating the likelihood that a plurality of users classified into the plurality of clusters belong to a category indicated by a representative attribute value of the target cluster, for an attribute indicating a category among the plurality of types of attributes; Information processing system.

11. 10. The information processing system according to claim 9, the output means outputs information indicating a representative personality of the user belonging to the target cluster based on the representative attribute value of the attribute selected from the plurality of types of attributes based on the contribution degree and the acquired relative value. Information processing system.

12. 2. The information processing system according to claim 1, the output means inputs, into a language model, instructions to generate a sentence indicating a personality, the instructions including at least some representative attribute values ​​of the plurality of types of attributes for the target cluster, and outputs information indicating a representative personality of a user belonging to the target cluster based on an output of the language model in response to the input. Information processing system.

13. 13. The information processing system according to claim 12, further comprising an attribute selection means for selecting a part of the plurality of types of attributes based on the degree of contribution; the output means inputs, to a language model, an instruction including a representative attribute value of the selected attribute, to generate a sentence indicating a personality, and outputs information indicating a representative personality of a user belonging to the target cluster based on an output of the language model in response to the input. Information processing system.

14. 10. The information processing system according to claim 9, the output means inputs, into a language model, instructions to generate a sentence indicating a personality, the instructions including relative values ​​of at least some of the multiple types of attributes for the target cluster, and outputs information indicating a representative personality of a user belonging to the target cluster based on an output of the language model in response to the input. Information processing system.

15. 2. The information processing system according to claim 1, The output means inputting an instruction based on the representative attribute value and the contribution degree into a language model to generate a sentence for generating an image showing a personality; inputting an instruction to generate an image based on the output of the language model in response to the input into an image generation model, and outputting information including the image output from the image generation model as information indicating a representative personality of users belonging to the target cluster; Information processing system.

16. An information processing system including a classification means, a contribution calculation means, a representative acquisition means, and an output means, a step of classifying the plurality of users into a plurality of clusters based on attribute values ​​of a plurality of types of attributes stored in association with each of the plurality of users; the contribution calculation means calculates, for each of a plurality of users classified into a target cluster that is one of the plurality of clusters, a contribution of each of the plurality of types of attributes to classification into the target cluster; a step in which the representative acquisition means acquires, as a representative attribute value representing the target cluster for each of the types of attributes, a mode or an average value of the attribute values ​​of each of the multiple types of attributes stored in association with each of the multiple users classified into the target cluster; a step in which the output means outputs information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and the contribution degree; An information processing method including:

17. a classification means for classifying the plurality of users into a plurality of clusters based on attribute values ​​of a plurality of types of attributes stored in association with each of the plurality of users; a contribution calculation means for calculating, for each of a plurality of users classified into a target cluster that is one of the plurality of clusters, a contribution of each of the plurality of types of attributes to classification into the target cluster; a representative acquisition means for acquiring a mode or an average value of each of the attribute values ​​of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster as a representative attribute value representing the target cluster for the attribute of that type; and an output means for outputting information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and the contribution degree; A program that allows a computer to function as a

18. An information processing system including a classification means, a contribution calculation means, a representative acquisition means, and an output means, a step of classifying the plurality of users into a plurality of clusters based on attribute values ​​of a plurality of types of attributes stored in association with each of the plurality of users; the contribution calculation means calculates, for each of a plurality of users classified into a target cluster that is one of the plurality of clusters, a contribution of each of the plurality of types of attributes to classification into the target cluster; the representative acquisition means selecting one or more representative users representing the target cluster from the plurality of users based on the attribute values ​​of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster, and acquiring representative attribute values ​​representing the target cluster based on the attribute values ​​of the representative users for at least some of the plurality of types of attributes; a step in which the output means outputs information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and the contribution degree; An information processing method including:

19. A classification means for classifying a plurality of users into a plurality of clusters based on attribute values ​​of a plurality of types of attributes stored in association with each of the plurality of users; a contribution calculation means for calculating, for each of a plurality of users classified into a target cluster that is one of the plurality of clusters, a contribution of each of the plurality of types of attributes to classification into the target cluster; a representative acquisition means for selecting one or more representative users representing the target cluster from the plurality of users based on the attribute values ​​of the plurality of types of attributes stored in association with each of the plurality of users classified into the target cluster, and acquiring representative attribute values ​​representing the target cluster based on the attribute values ​​of the representative users for at least some of the plurality of types of attributes; and an output means for outputting information indicating a representative personality of a user belonging to the target cluster based on the representative attribute value and the contribution degree; A program that allows a computer to function as a

Images