Program, information processing method, and information processing device

Abstract

Provided are a program, etc., capable of acquiring each customer's preferences. A computer acquires, in accordance with the program, content data related to content selected by a user. Then, the computer outputs preference information related to the user's preferences for products or services on the basis of the acquired content data.

Description

Program, information processing method, and information processing device

[0001] The present disclosure relates to a program, an information processing method, and an information processing device.

[0002] Patent Literature 1 discloses a system that determines a product or category to recommend to a customer based on points that quantify the customer's preference for the product or service. With the technology disclosed in Patent Literature 1, when a customer purchases or selects a product or service, points are added not only for the product or service but also for products or services that belong to the same category as the product or service. This makes it possible to recommend products or services that reflect not only the product or service purchased or selected by the customer, but also the customer's preferences for the product or service category.

[0003] Patent No. 5944605

[0004] In Patent Document 1, a product or category to be recommended to a customer is determined based on points accumulated for each product according to the product that the customer has previously purchased, etc. Therefore, there is a problem in that it is not possible to determine a product or category to be recommended to a customer who has not accumulated points for each product.

[0005] In one aspect, an object is to provide a program etc. that can acquire the preferences of each customer.

[0006] A program according to one aspect causes a computer to acquire content data relating to content selected by a user, and output preference information relating to the user's product or service preferences based on the acquired content data.

[0007] In one aspect, preferences for each customer can be obtained.

[0008] 1 is a block diagram showing an example of the configuration of an information processing device. FIG. 1 is an explanatory diagram showing an example of a questionnaire screen regarding music content. FIG. 1 is an explanatory diagram showing an example of the configuration of a first learning model. FIG. 2 is an explanatory diagram showing an example of the configuration of a second learning model. FIG. 2 is an explanatory diagram showing an example of a questionnaire screen for generating training data. FIG. 3 is a flowchart showing an example of a processing procedure for generating a learning model. FIG. 4 is a flowchart showing an example of a processing procedure for estimating preferences for fashion brands. FIG. 5 is an explanatory diagram showing a comparison between a user's preferences for fashion brands and preferences for fashion brands estimated from the user's sensibilities for music. FIG. 6 is a flowchart showing an example of a processing procedure for estimating preferences for fashion brands in embodiment 2. FIG. 7 is an explanatory diagram showing an example of the configuration of a third learning model. FIG. 8 is an explanatory diagram showing an example of the record layout of a training data DB. FIG. 9 is an explanatory diagram showing an example of the configuration of a fourth learning model. FIG. 10 is a flowchart showing an example of a processing procedure for comparing the sensibilities of users who have a high preference for each fashion brand. FIG. 11 is an explanatory diagram showing an example of a screen.

[0009] Hereinafter, a program, an information processing method, and an information processing device according to the present disclosure will be described in detail with reference to the drawings illustrating embodiments thereof.

[0010] (Embodiment 1) An information processing device that estimates a user's tastes (preferences) for products or services based on content data related to content selected (specified) by the user will be described. When selecting desired content (preferred content) from various types of content, such as text, books, music (songs, etc.), videos (movies, television programs, plays, YouTube (registered trademark), etc.), photographs, manga, animations, illustrations, computer graphics, and games, users often make selections based on their own sensibilities. Furthermore, when purchasing various products and services, users select products and services to be purchased based on their own sensibilities. In this way, users select desired content and products and services to be purchased based on their own sensibilities. It is believed that there is a correlation between a user's tastes for various content and their tastes for various products and services. Therefore, by estimating a user's tastes based on data related to content selected by the user, for example, content selected or purchased by the user, the user's tastes (preferences) for various products or services can be estimated from the estimated sensibilities. It is desirable that the type of content used to estimate sensibility and the type of product or service whose preference is estimated from the estimated sensibility be a combination that has a strong correlation between the numerical values ​​of the sensibility factors estimated from the user's image or impression of the content and the numerical values ​​of the sensibility factors estimated from the user's image or impression of the product or service. The inventors of the present application have discovered that there is a strong correlation between the sensibility of a user when selecting their favorite music content and the sensibility of the user when selecting their favorite fashion brand. Therefore, this embodiment describes an information processing device that estimates a user's sensibility from content data related to the music content selected by the user, where the content is music content and the product or service is a fashion brand, and estimates the user's preferences (preferences) for multiple fashion brands from the estimated sensibility.

[0011] FIG. 1 is a block diagram showing an example configuration of an information processing device. The information processing device 10 is a computer capable of various information processing and information transmission / reception, such as a server computer, a personal computer, or a tablet terminal. The information processing device 10 includes a control unit 11, a memory unit 12, a communication unit 13, an input unit 14, a display unit 15, a reading unit 16, and the like, which are interconnected via a bus. The control unit 11 is configured using one or more processors, such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), a TPU (Tensor Processing Unit), or an AI chip (AI semiconductor). The control unit 11 executes various information processing and control processes to be performed by the information processing device 10 by appropriately executing a program P stored in the memory unit 12. Note that if the control unit 11 includes multiple processors, each process may be executed by a different processor.

[0012] The storage unit 12 includes RAM (Random Access Memory), flash memory, a hard disk, an SSD (Solid State Drive), etc. The storage unit 12 stores the program P (program product, computer program) executed by the control unit 11 and various data required for executing the program P. The storage unit 12 also temporarily stores data generated when the control unit 11 executes the program P. The storage unit 12 also stores learning models M1 and M2 that have learned training data through, for example, machine learning. The learning models M1 and M2 are expected to be used as program modules constituting artificial intelligence software. The learning models M1 and M2 perform predetermined calculations on input values ​​and output calculation results. The storage unit 12 stores data such as coefficients and thresholds of functions that define these calculations as the learning models M1 and M2. The storage unit 12 may be composed of multiple storage devices, and a portion of the storage unit 12 may be another storage device connected to the information processing device 10 or another storage device with which the information processing device 10 can communicate. Furthermore, instead of being configured to store the learning models M1 and M2 in the storage unit 12, the information processing device 10 may access a server that stores the learning models M1 and M2 and read them out.

[0013] The communication unit 13 is a communication module for performing processes related to wired or wireless communication, and transmits and receives information to and from other devices via a network. The network may be the Internet, a public telephone network, or a local area network (LAN) established within the facility where the information processing device 10 is installed. The input unit 14 accepts operation inputs from a user and sends control signals corresponding to the operation content to the control unit 11. The input unit 14 includes, for example, a keyboard, a mouse, and a microphone for voice input. The display unit 15 is a liquid crystal display or an organic electroluminescence (EL) display, and displays various information according to instructions from the control unit 11. A part of the input unit 14 and the display unit 15 may be integrated into a touch panel.

[0014] The reading unit 16 reads information stored in a portable storage medium 10a such as a CD (Compact Disc), a DVD (Digital Versatile Disc), a USB (Universal Serial Bus) memory, or an SD (Secure Digital) card. The program P and various data stored in the storage unit 12 may be read by the control unit 11 from the portable storage medium 10a via the reading unit 16 and stored in the storage unit 12. The program P and various data may be written to the storage unit 12 during the manufacturing stage of the information processing device 10, or may be downloaded by the control unit 11 from another device via the communication unit 13 and stored in the storage unit 12.

[0015] In this embodiment, the information processing device 10 may be a multi-computer consisting of multiple computers, or may be a virtual machine virtually constructed by software within a single device. Furthermore, if the information processing device 10 is configured as a server computer, the information processing device 10 may be a local server installed in the facility where the information processing device 10 is located, or a cloud server connected via a network such as the Internet. The following description will be given assuming that the information processing device 10 is a single computer. Furthermore, the program P may be deployed on a single computer or at a single site, or may be distributed across multiple sites and executed on multiple computers interconnected via a network. Furthermore, the input unit 14 and the display unit 15 are not essential for the information processing device 10. The information processing device 10 may be configured to accept operations via a connected computer or to output information to be displayed to an external display device.

[0016] The information processing device 10 of this embodiment performs a process for estimating a user's fashion brand preferences based on responses to a questionnaire regarding the user's most favorite music content (songs). FIG. 2 is an explanatory diagram illustrating an example of a questionnaire screen regarding music content. The screen shown in FIG. 2 provides input fields for the song title and artist name of the user's most favorite song (most favorite song) or most frequently listened to song (most frequently listened song). The screen shown in FIG. 2 also displays multiple questionnaire items based on the Semantic Differential Method (SD) for the song entered in the song title and artist name input fields. For each questionnaire item, the user is asked to respond (input) the image or impression they have of the song using a number between 1 and 5. Note that the selectable values ​​for each questionnaire item are not limited to five (1 to 5) and can be any number. The number of questionnaire items can be, for example, approximately 50, but may be any number.

[0017] The information processing device 10 of this embodiment acquires responses to multiple questionnaire items regarding the user's most favorite music content via a questionnaire screen such as that shown in FIG. 2 . The information processing device 10 then estimates numerical values ​​for multiple affective factors representing the user's sensibilities from the acquired responses, and estimates the user's preferences for multiple fashion brands from the estimated numerical values ​​for the affective factors. The estimated preferences for each user's fashion brands can be used for advertising the fashion brands to each user. Therefore, the user's image or impression (answers to the questionnaire items) of the music content they like can be used as marketing information for the fashion brand. The information processing device 10 of this embodiment uses a first learning model M1 to estimate the numerical values ​​of the affective factors representing the user's sensibilities from the responses to the questionnaire items regarding music content, and uses a second learning model M2 to estimate the user's preferences for fashion brands from the numerical values ​​of each affective factor. In addition, the information processing device 10 may be configured to estimate a user's fashion brand preferences from questionnaire responses regarding music content selected by the user at their discretion, instead of the user's most preferred music content or the music content that the user listens to most frequently.

[0018] FIG. 3A is an explanatory diagram showing an example of the configuration of the first learning model M1, and FIG. 3B is an explanatory diagram showing an example of the configuration of the second learning model M2. The first learning model M1 is trained to input responses to questionnaire items related to music content selected by a user (e.g., the user's favorite music content) as content data. Based on the input content data, the first learning model M1 performs a calculation to estimate numerical values ​​for each of multiple types of affective factors related to the user's sensibilities, and output the calculated results. In the example of FIG. 3A, responses to multiple questionnaire items, including the first questionnaire item (vulgar, refined) (response to the first item) and the second questionnaire item (childlike, mature) (response to the second item) shown in FIG. 2, are input, and numerical values ​​for each of multiple pre-set affective factors are output. The affective factors used may be, for example, a gentle, soothing feeling, a cheerful feeling, a self-reliant feeling, an authentic feeling, a passionate feeling, a wild feeling, a straightforward feeling, a fresh feeling, a dazzling feeling, or any other feeling that can express the user's sensibilities. The affective factors may be some or all of these, but are not limited to these. The number of affective factors is, for example, about 13, but is not limited to this number. The numerical values ​​for the affective factors can be, for example, numerical values ​​between -1 and 1 (for example, numerical values ​​to two decimal places).

[0019] The first learning model M1 can be constructed using algorithms such as multiple regression analysis, SVM (Support Vector Machine), random forest, decision tree, XGBoost (eXtreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), neural network, and deep learning, or can be constructed using a combination of multiple algorithms. In addition to the above-mentioned algorithms, the first learning model M1 can also be constructed using dimensionality reduction algorithms such as factor analysis, principal component analysis (PCA), uniform manifold approximation and projection (UMAP), latent semantic indexing (LSI), singular value decomposition (SVD), linear discriminant analysis (LDA), independent component analysis (ICA), partial least squares (PLS), and t-distributed stochastic neighbor embedding (t-SNE). When the first learning model M1 is constructed using a dimensionality reduction algorithm, it is generated by unsupervised learning.

[0020] The first learning model M1 includes an input layer that receives questionnaire responses about music content, an output layer that outputs numerical values ​​for each emotional factor, and an intermediate layer that calculates output values ​​based on the input data. The input layer of the first learning model M1 has multiple input nodes, through which users receive questionnaire responses about music content. The intermediate layer of the first learning model M1 extracts features from the input data using various functions and thresholds, and outputs the extracted feature values ​​to the output layer. The output layer of the first learning model M1 has multiple output nodes, each associated with a pre-set emotional factor, such as a gentle, soothing feeling, a cheerful feeling, a self-reliant feeling, an authentic feeling, a passionate feeling, a wild feeling, a straightforward feeling, a fresh feeling, and a flashy feeling. Each output node outputs a numerical value for the associated emotional factor. With this configuration, the first learning model M1 outputs a user's numerical value for each emotional factor when a predetermined number of questionnaire responses are input.

[0021] The first learning model M1 is generated by machine learning using training data that associates training questionnaire responses (questionnaire responses regarding music content) with numerical values ​​(correct numerical values) of affective factors that represent the user's sensibilities and are calculated based on the questionnaire responses. The training data for the learning models M1 and M2 of this embodiment is generated based on the questionnaire responses from users. FIG. 4 is an explanatory diagram showing an example of a questionnaire screen for generating training data. The screen shown in FIG. 4 includes a questionnaire regarding fashion brand preferences in addition to the questionnaire regarding music shown in FIG. 2. The fashion brand questionnaire displays preset fashion brands, and is configured so that the user answers (inputs) their fashion brand preference for each fashion brand using a numerical value from 1 to 7.

[0022] Training data for the first learning model M1 is generated by associating numerical values ​​for each affective factor calculated based on questionnaire responses about songs acquired through a questionnaire shown in FIG. 4 with the questionnaire responses using a maximum likelihood estimation method, such as a preset linear equation. The linear equation for calculating the numerical values ​​of each affective factor has a coefficient for each affective factor set for each questionnaire response (numeric values ​​1 to 5 in the example of FIG. 4 ). The first learning model M1 learns to output the correct numerical value for each affective factor from the output node corresponding to each affective factor when the questionnaire responses included in the training data are input. In the learning process, the first learning model M1 performs calculations based on the input questionnaire responses to calculate output values ​​from each output node. The first learning model M1 then compares the calculated output values ​​of each output node with the correct numerical value and optimizes parameters used in the calculation process so that the two approximate each other. For example, parameters such as weights (coupling coefficients) between nodes in the first learning model M1 are optimized using backpropagation, steepest descent, or the like. This results in a first learning model M1 that outputs a numerical value for an affective factor that represents the sensibility of a certain user when the user inputs a questionnaire response regarding the user's music.

[0023] The first learning model M1 is configured to output a numerical value for each emotional factor, expressed as a continuous value between -1 and 1, for example. However, it may also be configured to determine any of the numerical values ​​selectable as the emotional factor. In this case, the first learning model M1 may be configured to have multiple output nodes corresponding to multiple numerical values ​​selectable as the emotional factor numerical value for each emotional factor, and each output node may output a confidence level for the associated numerical value. In this case, each output node may output an output value between 0 and 1 for each emotional factor, and the sum of the output values ​​from each output node is 1 (100%). In this configuration, when a questionnaire response regarding a song is input, the first learning model M1 outputs a confidence level for the numerical value assigned to each output node for each emotional factor. When using the first learning model M1 configured in this way, the information processing device 10 can estimate the numerical value of each emotional factor as the numerical value associated with the output node that outputs the largest output value (confidence level) for that emotional factor. Note that the first learning model M1 configured as described above may have a single output node that outputs the highest confidence value for each affective factor, instead of multiple output nodes that output confidence levels for each value. The first learning model M1 can also be configured using a dimensionality reduction algorithm. In this case, the first learning model M1 is trained to output values ​​for, for example, approximately 13 affective factors by performing factor analysis on training questionnaire responses (e.g., questionnaire responses of approximately 50 items regarding music content from 2,000 people) through unsupervised learning. The first learning model M1 is trained to output affective factors that can contribute to estimating fashion brand preferences while maintaining the information volume of the 50 questionnaire responses.

[0024] The second learning model M2 shown in FIG. 3B is trained to input numerical values ​​for multiple types of affective factors representing a user's sensibilities, perform calculations to estimate the user's preferences (preference information) for multiple types of fashion brands based on the input numerical values ​​for each affective factor, and output the calculation results. The numerical values ​​for the affective factors input to the second learning model M2 can be, for example, the numerical values ​​for the affective factors estimated using the first learning model M1 shown in FIG. 3A. The fashion brand preferences can be, for example, the numerical values ​​1 to 7 used in the questionnaire shown in FIG. 4. The second learning model M2 can also be constructed using algorithms such as multiple regression analysis, SVM, random forest, decision tree, XGBoost, LightGBM, neural network, and deep learning, or can be constructed by combining multiple algorithms.

[0025] The second learning model M2 includes an input layer to which numerical values ​​for multiple affective factors are input, an output layer that outputs preferences for each fashion brand, and an intermediate layer that calculates output values ​​based on the input data. The input layer of the second learning model M2 has multiple input nodes, through which numerical values ​​for the multiple affective factors are input. The intermediate layer of the second learning model M2 extracts features of the input data using various functions, thresholds, etc., and outputs the extracted feature values ​​to the output layer. The output layer of the second learning model M2 has multiple output nodes, each associated with a pre-set fashion brand, and each output node outputs a preference for the associated fashion brand. Specifically, the output layer of the second learning model M2 has multiple (here, seven) output nodes, each associated with a selectable numerical value (e.g., preference levels 1 to 7) for each fashion brand, and each output node outputs a confidence level for each associated preference. For each fashion brand, each output node outputs an output value between 0 and 1, with the sum of the output values ​​from each output node being 1 (100%). When using a second learning model M2 configured in this way, the information processing device 10 can estimate, for each fashion brand, the preference (e.g., any of 1 to 7) associated with the output node that outputs the largest output value (certainty) as the preference for that fashion brand. Note that the second learning model M2 configured in this way may be configured to have a single output node that outputs the preference with the highest certainty, instead of having multiple output nodes that output certainty for each preference for each fashion brand.

[0026] The second learning model M2 is generated by machine learning using training data that associates the numerical values ​​of each training affective factor with the fashion brand preferences (correct preferences) of the same user as the questionnaire responses used to calculate the numerical values ​​of the affective factors. Specifically, the training data for the second learning model M2 is generated by associating the numerical values ​​of each affective factor calculated based on the questionnaire responses about music with the fashion brand preferences (numeric values ​​1 to 7 in the example of FIG. 4) obtained through the questionnaire shown in FIG. 4. The second learning model M2 learns to output the correct preferences for each fashion brand from the output node corresponding to each fashion brand when the numerical values ​​of each affective factor included in the training data are input. In the learning process, the second learning model M2 performs calculations based on the input numerical values ​​of each affective factor and calculates an output value from each output node. The second learning model M2 then compares the calculated output value of each output node with the value corresponding to the correct preference for each fashion brand (1 for the output node corresponding to the correct preference, 0 for other output nodes), and optimizes the parameters used in the calculation process so that the two values ​​approximate each other. Again, parameters such as the weights (coupling coefficients) between nodes in the second learning model M2 are optimized using backpropagation, steepest descent, or the like. This results in a second learning model M2 that, when the numerical values ​​of the affective factors representing a user's sensibilities are input, outputs the user's preference for fashion brands.

[0027] The learning of the learning models M1 and M2 may be performed by the information processing device 10 or by another learning device. The trained learning models M1 and M2 generated by training on the other learning device are downloaded from the learning device to the information processing device 10 via a network or via the portable storage medium 10a and stored in the storage unit 12.

[0028] The following describes a process for generating learning models M1 and M2 by learning the training data described above. FIG. 5 is a flowchart showing an example of the process for generating learning models M1 and M2. The following process is executed by the control unit 11 of the information processing device 10 in accordance with a program P stored in the storage unit 12, but may also be executed by another learning device. In the following process, the control unit 11 first generates training data based on responses to the questionnaire shown in FIG. 4, and then uses the generated training data to train the learning models M1 and M2. It is assumed that the questionnaire responses used to generate the training data are provided by a large number of respondents and stored in a predetermined area (a predetermined DB) of the storage unit 12.

[0029] The control unit 11 of the information processing device 10 acquires the survey responses for one person stored in the storage unit 12 (S11). The survey responses may be read from the storage unit 12 or may be acquired, for example, via a network from a terminal of the survey respondent or a terminal that collects survey responses. The acquired survey responses include a survey response to music and a survey response to fashion brand preference.

[0030] The control unit 11 extracts questionnaire responses to songs from the questionnaire responses and calculates a value for each affective factor representing the respondent's sensibility based on the questionnaire responses to the songs (S12). Here, the control unit 11 calculates the value of each affective factor representing the respondent's sensibility (maximum likelihood estimate) using a maximum likelihood estimation method. Specifically, the control unit 11 calculates the value of the affective factor representing the respondent's sensibility using a linear equation with coefficients preset for each questionnaire item. The control unit 11 associates the extracted questionnaire responses to songs with the calculated values ​​(correct values) of each affective factor to generate training data to be used for training the first learning model M1, and stores the training data in the memory unit 12 (S13). The control unit 11 stores the generated training data, for example, in a training DB (not shown) for the first learning model M1 provided in the memory unit 12.

[0031] The control unit 11 extracts the survey responses regarding fashion brand preferences from the survey responses, associates the numerical values ​​of each affective factor calculated in step S12 with the extracted fashion brand preferences (correct preferences), generates training data to be used for learning the second learning model M2, and stores the data in the memory unit 12 (S14). Here too, the control unit 11 stores the generated training data, for example, in a training DB (not shown) for the second learning model M2 prepared in the memory unit 12.

[0032] The control unit 11 determines whether there are any unprocessed questionnaire responses stored in the memory unit 12 that have not been used in the training data generation process (S15). If it is determined that there are unprocessed questionnaire responses (S15: YES), the control unit 11 returns to the process of step S11 and performs the processes of steps S11 to S14 on the unprocessed questionnaire responses. The control unit 11 repeats the processes of steps S11 to S15 until it determines that there are no unprocessed questionnaire responses. As a result, training data used for training the learning models M1 and M2 is generated based on the questionnaire responses stored in the memory unit 12 and accumulated in the training DBs (the training DB for the first learning model M1 and the training DB for the second learning model M2).

[0033] If the control unit 11 determines that there are no unprocessed questionnaire responses (S15: NO), it uses the training data stored in the training DB as described above to train the learning models M1 and M2. For example, the control unit 11 reads one of the training data stored in the training DB for the first learning model M1 and performs a training process for the first learning model M1 based on the read training data (S16). Here, the control unit 11 inputs questionnaire responses for songs included in the training data into the first learning model M1 and obtains output values ​​output from the first learning model M1 in response to the input questionnaire responses. The control unit 11 compares the output values ​​of each output node output from the first learning model M1 with the correct numerical values ​​of each affective factor included in the training data, and trains the first learning model M1 so that the two values ​​approximate each other. In the training process, the control unit 11 optimizes parameters such as weights between nodes in the first learning model M1 using an error backpropagation algorithm, which sequentially updates the parameters from the output layer to the input layer.

[0034] The control unit 11 determines whether there is unprocessed training data stored in the training DB for the first learning model M1 that has not yet undergone learning processing (S17). If it determines that there is unprocessed training data (S17: YES), the control unit 11 returns to the processing of step S16 and performs learning processing on the first learning model M1 using the unprocessed training data. If it determines that there is no unprocessed training data (S17: NO), the control unit 11 reads one of the training data stored in the training DB for the second learning model M2 and performs learning processing on the second learning model M2 based on the read training data (S18). Here, the control unit 11 inputs the numerical values ​​of each affective factor included in the training data into the second learning model M2 and obtains output values ​​output from the second learning model M2 in response to the input of the numerical values ​​of each affective factor. The control unit 11 compares the output value of each output node output from the second learning model M2 with the correct preference for each fashion brand included in the training data, and trains the second learning model M2 so that the two are similar. Specifically, the control unit 11 compares the output value of each output node with a value corresponding to the correct preference for each fashion brand (1 for the output node corresponding to the correct preference, and 0 for the other output nodes), and trains the second learning model M2 so that the two are similar. Here, too, the control unit 11 optimizes parameters such as the weights between nodes in the second learning model M2 using an error backpropagation method, which sequentially updates them from the output layer to the input layer.

[0035] The control unit 11 determines whether there is unprocessed training data that has not yet been subjected to the learning process among the training data stored in the training DB for the second learning model M2 (S19). If it is determined that there is unprocessed training data (S19: YES), the control unit 11 returns to the process of step S18, and performs the learning process of the second learning model M2 using the unprocessed training data. If it is determined that there is no unprocessed training data (S19: NO), the control unit 11 ends the series of processes.

[0036] The above-described learning process generates a first learning model M1 that outputs the numerical values ​​of the emotional factors of a survey respondent when a questionnaire response about a song is input, and a second learning model M2 that outputs the survey respondent's fashion brand preference when the numerical values ​​of each emotional factor are input. In the above-described process, the training data generation process in steps S11 to S15 and the learning models M1 and M2 generation process in steps S16 to S19 may be performed by separate devices. The learning models M1 and M2 can be further optimized by repeatedly performing the learning process using the training data described above. Furthermore, by retraining already trained learning models M1 and M2 using the above-described learning process, learning models M1 and M2 with further improved discrimination accuracy can be generated. If the first learning model M1 is configured using a dimensionality reduction algorithm, in step S13, the control unit 11 stores the questionnaire response about the song as training data for the first learning model M1. Then, in step S16, the control unit 11 performs factor analysis on the training data (questionnaire responses) to learn the first learning model M1.

[0037] The following describes a process in which the information processing device 10 of this embodiment estimates a user's fashion brand preference based on the user's responses to a questionnaire about music. Fig. 6 is a flowchart showing an example of a process for estimating a user's fashion brand preference.

[0038] In this embodiment, a questionnaire such as that shown in Fig. 2 is conducted at any time, such as when a user starts using a music streaming service (music distribution service) or while using the service, or when the user accesses various sites via a network. The information processing device 10 estimates the user's fashion brand preferences from the responses to the questionnaire conducted at the arbitrary time.

[0039] The control unit 11 of the information processing device 10 acquires survey responses from any user (S21). The control unit 11 may acquire user information, including the survey respondent's contact information, along with the survey responses. The survey responses may be acquired in advance via the communication unit 13 or the input unit 14 and stored in the storage unit 12. In this case, the control unit 11 reads the survey responses from the storage unit 12. The control unit 11 may also acquire the survey responses via a network from the survey respondent's terminal, a terminal that collects survey responses, or the like. The control unit 11 estimates the numerical values ​​of each affective factor representing the respondent's sensibility based on the acquired survey responses about the music (S22). Here, the control unit 11 inputs the survey responses about the music into the first learning model M1 and acquires the output values ​​from each output node of the first learning model M1 as the numerical values ​​of the affective factors associated with each output node.

[0040] Next, the control unit 11 estimates the survey respondent's preference for each fashion brand based on the estimated values ​​of each affective factor (S23). Here, the control unit 11 inputs the values ​​of each affective factor into the second learning model M2 and acquires the preference for each fashion brand based on the output values ​​from each output node of the second learning model M2. Specifically, the control unit 11 acquires, for each fashion brand, the preference associated with the output node that outputs the largest output value (certainty) as the preference for that fashion brand. If the control unit 11 has acquired user information along with the survey responses, the control unit 11 associates the user information with brand information including the preference for each fashion brand estimated in step S23 and stores the associated information in the storage unit 12 (S24). Note that the control unit 11 may store at least one of the survey responses acquired in step S21 and the values ​​of each affective factor estimated in step S22 in the storage unit 12, in addition to the user information and fashion information.

[0041] The control unit 11 determines whether there are any unprocessed survey responses for which the user's fashion brand preference has not yet been estimated (S25). If it is determined that there are any unprocessed survey responses (S25: YES), the control unit 11 returns to the process of step S21 and performs steps S21 to S24 on the unprocessed survey responses. This allows the fashion brand preference of each user who responded to the music survey to be estimated. The control unit 11 repeats the process of steps S21 to S25 until it determines that there are no unprocessed survey responses. If it determines that there are no unprocessed survey responses (S25: NO), it ends the series of processes. This allows the user's fashion brand preference to be estimated based on the survey responses regarding music obtained from the user. The preference for each fashion brand estimated in step S23 is expressed as a number from 1 to 7 (1: very dislike, 7: very like), allowing the high and low preferences for each fashion brand to be compared.

[0042] The above-described process makes it possible to estimate a user's fashion brand preferences from their responses to a questionnaire about music content selected by the user (e.g., their most favorite music content). This makes it possible to provide users with, for example, promotions and advertisements about highly preferred fashion brands, product recommendation information, and to provide fashion brand retailers with, for example, information about users who have a high preference for each fashion brand. The user's preferred fashion brands can be estimated simply by answering a questionnaire about music, without the user having to be aware of their fashion brand preferences.

[0043] FIG. 7 is an explanatory diagram showing a comparison between a user's preference for a fashion brand and a preference for a fashion brand estimated from the user's sensitivity to music. The graph in FIG. 7 plots each user's actual preference for a fashion brand on the horizontal axis and the preference for a fashion brand estimated from the user's sensitivity (the numerical value of each sensitivity factor) estimated from a questionnaire response to a song selected by the user on the vertical axis. The graph plots each user's actual preference and the preference estimated from their sensitivity to music. Each user ranks multiple brands and assigns a score to each brand according to the ranking (the higher the ranking, the higher the score). The actual preference is calculated by subtracting the score of the second brand from the score of the first brand (the score of the first brand minus the score of the second brand). In other words, the actual preference shown on the horizontal axis of the graph in FIG. 7 indicates the difference between each user's preference for the two brands, with a higher preference for the first brand toward the right and a higher preference for the second brand toward the left. The preference estimated from each user's musical sensibility is calculated by the process of this embodiment, where the user's preference for multiple brands is estimated from the user's sensibility (the value of each sensibility factor) calculated from the user's responses to a questionnaire about music, and the ranking of the brands is determined from the user's preference, a score is assigned to each brand according to the ranking (the higher the ranking, the higher the score), and the value is calculated by subtracting the score of the second brand from the score of the first brand. That is, the estimated preference shown on the vertical axis of the graph in Figure 7 indicates the difference in the preference for the two brands estimated from each user's musical sensibility, with the higher the value, the higher the preference for the first brand, and the lower the value, the higher the preference for the second brand.

[0044] 7 shows that there is a strong correlation between a user's actual preference for fashion brands and the preference for fashion brands estimated from the user's sensibility to music. Therefore, in this embodiment, the user's sensibility to music is expressed as numerical values ​​of multiple emotional factors based on the user's responses to a questionnaire about a song selected by the user, and the user's preference for fashion brands is estimated using the numerical values ​​of the multiple emotional factors that represent the user's sensibility, making it possible to accurately estimate the user's preference (taste) for fashion brands.

[0045] As described above, since users make various selections based on their own sensibilities, a configuration may be adopted in which a user's sensibilities are estimated based on the user's preferences (preferences) for various content, not limited to music content. Furthermore, a configuration may be adopted in which a user's preferences (preferences) for various products and services, not limited to fashion brands, are estimated based on the user's sensibilities. For example, a configuration may be adopted in which the numerical values ​​of the sensibility factors representing the user's sensibilities are estimated based on questionnaire responses to one or more of content, such as video content (movies, television programs, plays, YouTube, etc.), text, books, photographs, manga, animations, illustrations, computer graphics, and games. Furthermore, the questionnaire responses used to estimate the sensibility may include, in addition to questionnaire responses regarding the user's most preferred music content as shown in FIG. 2 , questionnaire responses regarding the user's own musical sensibilities or musical values, questionnaire responses regarding the image and impression of songs or artists that affect the score (numerical value) of the sensibility factors, and sorting results of a predetermined number of songs, artists, melody lines, etc., that affect the score of the sensibility factors, in order of preference.

[0046] Furthermore, based on the user's sensibility estimated as described above, the system may be configured to estimate the user's preferences (tastes) for various products and services, including buildings (e.g., restaurants, shopping centers, shopping malls, landmarks), places, cities, travel destinations (e.g., tourist spots), and types of pets. Furthermore, based on the user's sensibility (values ​​of the sensibility factors), the system may be configured to estimate the user's preferences for people (e.g., coworkers, friends, and people of the opposite sex). In this case, it is possible to estimate compatibility with other people based on the user's preferences with other people. It is preferable that the combination of content used to estimate the sensibility (values ​​of the sensibility factors) and a product or service whose user's tastes (tastes) are estimated using the estimated sensibility be a combination in which there is a strong correlation between the values ​​of the sensibility factors estimated from questionnaire responses to the content and the values ​​of the sensibility factors estimated from similar questionnaire responses to the product or service.

[0047] In the above-described process, the process of estimating the numerical values ​​of the affective factors representing the respondent's sensibilities from the questionnaire responses about music using the first learning model M1 and / or the process of estimating the respondent's fashion brand preferences from the numerical values ​​of the affective factors using the second learning model M2 are not limited to being performed locally by the information processing device 10. For example, these processes may be performed by a server that provides the learning models M1 and M2. For example, the information processing device 10 may be configured to transmit the received questionnaire responses to the server and receive the numerical values ​​of the affective factors estimated from the questionnaire responses at the server. Furthermore, the information processing device 10 may be configured to transmit the numerical values ​​of the affective factors representing the respondent's sensibilities to the server and receive the respondent's fashion brand preferences estimated from the numerical values ​​of the affective factors at the server.

[0048] (Embodiment 2) An information processing device will be described that estimates the numerical values ​​of affective factors that represent a user's sensibility based on songs included in a playlist provided by a music streaming service (music distribution service), instead of the questionnaire responses to songs in the above-mentioned embodiment 1. The information processing device of this embodiment has a configuration similar to that of the information processing device 10 of embodiment 1 shown in Figure 1, and therefore a description of the configuration will be omitted.

[0049] FIG. 8 is a flowchart showing an example of a processing procedure for estimating a fashion brand preference level according to the second embodiment. The processing shown in FIG. 8 is the processing shown in FIG. 6 , except that steps S31-S32 are added instead of step S21, and step S33 is added instead of step S25. Explanations of steps identical to those in FIG. 6 will be omitted. In this embodiment, the control unit 11 of the information processing device 10 acquires a playlist used by a user on a music distribution service (S31). For example, if the information processing device 10 is a server providing a music distribution service, the control unit 11 can acquire the playlist from the user's terminal by transmitting and receiving information to and from the user's terminal via a network. Furthermore, if the user's terminal is capable of transmitting playlists via a network, the control unit 11 may acquire the playlist from the user's terminal via the network. The playlist includes the song title and artist name of each song. Note that the control unit 11 may also acquire information about songs registered as favorites on the music distribution service and information about songs included in the playback history (listening history) instead of the playlist.

[0050] The control unit 11 estimates responses to questionnaire items such as those shown in FIG. 2 based on the acquired playlist (S32). For example, the control unit 11 associates each song itself or song information such as the tempo, rhythm, pitch, and melody line of the song with an answer (e.g., a numerical value from 1 to 5) for each questionnaire item that a user who likes the song is likely to answer. The control unit 11 then extracts trends in the songs included in the playlist or in the song information such as the tempo, rhythm, pitch, and melody line of the song, and identifies each questionnaire answer corresponding to the extracted trends in the song or song information, thereby estimating the survey answers of the user using the playlist. Furthermore, if the playlist includes a number of times played (listened to), the control unit 11 may extract the song with the most number of plays and estimate the survey answers of the user by identifying each questionnaire answer associated with this song or the song information of that song.

[0051] The control unit 11 then executes the processes from step S22 onwards. As a result, in this embodiment as well, it is possible to estimate the numerical values ​​of each affective factor that represents the sensibility of the user of the playlist based on the questionnaire responses estimated from the playlist, and to estimate the preference of the user of the playlist for each fashion brand. The control unit 11 of this embodiment determines whether or not there are any unprocessed playlists (S33), and if it determines that there are (S33: YES), it returns to step S31, and if it determines that there are no unprocessed playlists (S33: NO), it ends the process.

[0052] Through the above-described processing, in this embodiment, the user's sensibilities are expressed as numerical values ​​of multiple sensibility factors based on the playlists used by the user, and the user's fashion brand preferences can be estimated based on the user's sensibilities. Therefore, it is possible to estimate the user's preferences for fashion brands that match the user's sensibilities (tastes). Furthermore, in this embodiment, the user's fashion brand preferences are estimated based on the playlists used by the user on a music distribution service, so the user only needs to use the music distribution service, and no additional operational burden is placed on the user.

[0053] In this embodiment, too, the system may be configured to estimate a user's sensibility based on playlists, playback history, favorite registrations, or other user preferences for various content, such as video content in a video streaming service, rather than song playlists in a music distribution service. In this case, responses to each questionnaire item are associated with various content or information about each content, and each questionnaire response corresponding to content or information about content registered in the user's playlists, playback history, favorite registrations, or other user preferences may be identified. Furthermore, the system may be configured to estimate a user's sensibility (preferences) for various products and services based on the user's sensibility estimated from the user's preferences based on playlists, playback history, favorite registrations, or other user preferences for various content. In this embodiment, too, the process of estimating the values ​​of the affective factors representing the user's sensibility from song-related questionnaire responses (questionnaire responses estimated from playlists) using the first learning model M1 and / or the process of estimating the user's fashion brand preferences from the values ​​of the affective factors using the second learning model M2 are not limited to being performed locally by the information processing device 10. These processes may be performed by a server that provides the learning models M1 and M2.

[0054] In this embodiment, the processing is the same as that of the first embodiment, and the same effects as those of the first embodiment can be obtained, except for the processing of estimating the user's responses to questionnaire items related to songs (content) based on the songs (content) registered in the user's playlist, playback history, or favorites, and estimating the numerical values ​​of the affective factors that represent the user's sensibilities from the estimated questionnaire responses. Furthermore, in this embodiment, the modified examples described in the first embodiment above can also be applied as appropriate.

[0055] (Embodiment 3) The information processing device 10 of the above-described embodiments 1 and 2 is configured to estimate a user's fashion brand preferences from the user's questionnaire responses or questionnaire responses estimated from the user's playlist using a first learning model M1 and a second learning model M2. In this embodiment, an information processing device is described that estimates a user's fashion brand preferences from the user's questionnaire responses or questionnaire responses estimated from the user's playlist using a single third learning model. The information processing device of this embodiment has a configuration similar to that of the information processing device 10 of embodiment 1 shown in FIG. 1 , and therefore a detailed description of the configuration will be omitted. Note that the information processing device 10 of this embodiment stores a third learning model M3 in the storage unit 12 instead of the first learning model M1 and the second learning model M2.

[0056] 9 is an explanatory diagram showing an example configuration of the third learning model M3. Similar to the input data of the first learning model M1, the third learning model M3 receives inputs such as questionnaire responses regarding music content selected by the user (e.g., the user's most preferred music content). Based on the input data, the third learning model M3 performs calculations to estimate the user's preferences (preference information) for each of multiple fashion brands. Similar to the output data of the second learning model M2, the third learning model M3 is trained to output the user's preferences for fashion brands. The third learning model M3 can be configured using algorithms such as multiple regression analysis, SVM, random forest, decision tree, XGBoost, LightGBM, neural network, and deep learning. It may also be configured by combining multiple algorithms, such as combining the first learning model M1 and the second learning model M2 of embodiments 1 and 2.

[0057] The third learning model M3 includes an input layer to which questionnaire responses about music content are input, similar to the first learning model M1, an output layer to which preferences for each fashion brand are output, similar to the second learning model M2, and an intermediate layer to calculate output values ​​based on the input data. With this configuration, when questionnaire responses about music are input, the third learning model M3 outputs a confidence level for each associated preference level from each output node for each fashion brand. When using this third learning model M3, the information processing device 10 acquires, for each fashion brand, the preference level (e.g., any one of 1 to 7) associated with the output node that outputs the largest output value (confidence level) as the preference level for that fashion brand.

[0058] The third learning model M3 is generated by machine learning using training data that associates questionnaire responses about training songs with the respondents' preferences for fashion brands (correct preferences). The training data for the third learning model M3 is generated by associating questionnaire responses about music content obtained through the questionnaire (numbers 1 to 5 for each questionnaire item in the example of FIG. 4 ) with preferences for fashion brands obtained through the questionnaire (numbers 1 to 7 in the example of FIG. 4 ). The preferences used in the training data may be preferences for each fashion brand estimated from the questionnaire responses using the first learning model M1 and the second learning model M2 of the first embodiment. The third learning model M3 learns to output correct preferences for each fashion brand from the output node corresponding to each fashion brand when the questionnaire responses included in the training data are input. During the learning process, the third learning model M3 performs calculations based on the input questionnaire responses and calculates output values ​​from each output node. The third learning model M3 then compares the calculated output value of each output node with a value corresponding to the correct preference for each fashion brand (1 for the output node corresponding to the correct preference, 0 for other output nodes), and optimizes the parameters used in the calculation process so that the two values ​​approximate each other. Again, parameters such as the weights between nodes in the third learning model M3 are optimized using backpropagation, steepest descent, or the like. This results in a third learning model M3 that outputs a user's preference for fashion brands when a user enters a questionnaire response about a song. The learning of the third learning model M3 may also be performed by the information processing device 10 or another learning device.

[0059] The information processing device 10 of this embodiment, configured as described above, can execute processing similar to that shown in FIG. 6 . In the information processing device 10 of this embodiment, the control unit 11 does not perform the processing of step S22, but instead estimates the survey respondent's preference for each fashion brand based on the survey responses about music acquired in step S21 (S23). Here, the control unit 11 inputs the survey responses to a third learning model M3 and acquires the preference for the fashion brand associated with each output node based on the output values ​​from each output node of the third learning model M3. Specifically, the control unit 11 estimates the preference for each fashion brand as the preference (e.g., any one of 1 to 7) associated with the output node that outputs the largest output value (confidence level) for each fashion brand.

[0060] By the above-described process, in this embodiment, too, a user's fashion brand preferences can be estimated from the user's responses to a questionnaire about the user's most preferred music content. Thus, the user can estimate the user's preferences for fashion brands that match the user's sensibilities (tastes) simply by answering a questionnaire about the music they have selected (e.g., their favorite music).

[0061] The configuration of this embodiment is applicable to the information processing device 10 of the above-described embodiments 1 and 2, and similar effects can be obtained even when applied to the information processing device 10 of embodiments 1 and 2. Furthermore, the modified examples described in the above-described embodiments 1 and 2 can also be applied to this embodiment as appropriate.

[0062] (Embodiment 4) A modified example of the third learning model M3 of embodiment 3 will be described. The information processing device of this embodiment has the same configuration as the information processing device 10 of embodiment 3, except that a fourth learning model is stored in the storage unit 12 instead of the third learning model M3. Note that the information processing device 10 of this embodiment stores in the storage unit 12 a DB (hereinafter referred to as a training data DB) that stores data that can be used as training data for learning the fourth learning model.

[0063] FIG. 10 is an explanatory diagram showing an example of a record layout of a training data DB. The training data DB shown in FIG. 10 includes a data ID column, an artist column, a song column, a song questionnaire response column, an affective factor column, a brand questionnaire response column, etc., and stores information about each user in association with a data ID assigned to data collected from each user. The artist column, song column, song questionnaire response column, and brand questionnaire response column store the artist name, song title, responses to each questionnaire item related to the song (e.g., a numerical value from 1 to 5), and questionnaire responses regarding preferences for each fashion brand (e.g., a numerical value from 1 to 7) of favorite songs, which the user responded to through the questionnaire shown in FIG. 4, for example. The affective factor column stores a numerical value for each affective factor calculated by a predetermined calculation based on the song questionnaire response, or a numerical value for each affective factor estimated using the first learning model M1 in embodiment 1. In addition to the questionnaire responses regarding the preference for each fashion brand, the brand questionnaire response column may also store the preference for each fashion brand estimated for each user using the second learning model M2 in embodiment 1. When estimating the preference for each fashion brand using the second learning model M2, the numerical values ​​of each affective factor input to the second learning model M2 may be numerical values ​​for each affective factor calculated by a predetermined calculation based on the questionnaire responses regarding music, or may be numerical values ​​of each affective factor estimated from the questionnaire responses regarding music using the first learning model M1.

[0064] FIG. 11 is an explanatory diagram showing an example configuration of the fourth learning model M4. Instead of questionnaire responses about music content in the third learning model M3, the fourth learning model M4 uses content data about music content selected (specified) by the user as input data. The content data includes, for example, song titles and artist names of music content that the user likes or frequently listens to. The fourth learning model M4 is trained to perform calculations to estimate the user's preferences (preference information) for each of multiple fashion brands based on the input content data and output the user's preferences for fashion brands. The fourth learning model M4 can be constructed using algorithms such as multiple regression analysis, SVM, random forest, decision tree, XGBoost, LightGBM, neural network, and deep learning, or may be constructed using a combination of multiple algorithms.

[0065] The fourth learning model M4 includes an input layer to which content data is input, an output layer that outputs preferences for each fashion brand, and an intermediate layer that calculates output values ​​based on the input data. The input layer of the fourth learning model M4 has, for example, two input nodes, and song titles and artist names of music content that the user likes or frequently listens to are input via each input node. The input layer of the fourth learning model M4 may be configured to input either song titles or artist names. The intermediate layer of the fourth learning model M4 extracts features of the input data using various functions and thresholds, and outputs the extracted feature values ​​to the output layer. The output layer of the fourth learning model M4 has multiple output nodes, each associated with a pre-set fashion brand, and each output node outputs a preference (e.g., one of 1 to 7) for the associated fashion brand. Specifically, the output layer of the fourth learning model M4 has, for each fashion brand, a softmax function (not shown) and multiple (seven) output nodes associated with selectable preference values ​​(e.g., 1 to 7). Each output node outputs a confidence level for each associated preference. For each fashion brand, each output node outputs an output value of, for example, 0 to 1, with the sum of the output values ​​from each output node being 1 (100%). The fourth learning model M4 configured in this way can estimate the preference for each fashion brand based on the preference level (e.g., any of 1 to 7) associated with the output node that outputs the largest output value (confidence level). By using the fourth learning model M4 configured in this way, the information processing device 10 can identify a user's preference for each fashion brand based on the output values ​​output from the fourth learning model M4 when the information processing device 10 inputs content data for music content selected by the user. In addition, the fourth learning model M4 may be configured to have a single output node that outputs the preference with the highest certainty for each fashion brand, instead of having multiple output nodes that output the certainty for each preference.

[0066] The fourth learning model M4 is generated by machine learning using training data that associates content data related to training music content with the preferences (correct preferences) for each fashion brand of users who selected the music content. The training data for the fourth learning model M4 is generated using data stored in a training data DB in the memory unit 12. Specifically, the training data for the fourth learning model M4 is generated by associating survey responses to fashion brands with the artist names and song titles in each piece of data stored in the training data DB. In the learning process, when the artist names and song titles (content data) included in the training data are input, the fourth learning model M4 performs calculations based on the input data and calculates output values ​​from each output node. Then, for each fashion brand, the fourth learning model M4 compares the calculated output values ​​of each output node with values ​​corresponding to the correct preferences (1 for the output node corresponding to the correct preferences and 0 for other output nodes) and optimizes parameters used in the calculation process so that the two values ​​approximate each other. Here, parameters such as the weights between nodes in the fourth learning model M4 are optimized using backpropagation, steepest descent, etc. This results in a fourth learning model M4 that, when content data of a song preferred by a certain user is input, outputs a confidence level for each preference from the output node of each fashion brand. Learning of the fourth learning model M4 may also be performed by the information processing device 10 or another learning device.

[0067] The content data, which is input data for the fourth learning model M4, may include information such as the song titles and artist names of music content preferred by the user, as well as information on the tempo, rhythm, pitch, and melody lines of the music content. It may also include the results of the user sorting multiple songs, artists, or melody lines that influence the user's sensibilities in order of preference. The input data for the fourth learning model M4 may also be music information for songs (music content) registered in playlists, playback history, or favorites used in music distribution services. The fourth learning model M4 configured in this way is obtained by storing the content data described above in a training data database and learning using the training data, which is the content data stored in the training data database and questionnaire responses to the corresponding fashion brands.

[0068] When using the fourth learning model M4 configured as described above, the information processing device 10 can execute processing similar to that shown in FIG. 6 . In the information processing device 10 of this embodiment, the control unit 11 does not perform the processing of step S22, but instead estimates the user's preference for each fashion brand based on the artist names and song titles of the user's favorite songs among the responses to the song-related questionnaire obtained in step S21 (S23). Here, the control unit 11 inputs the artist names and song titles (content data) of the user's favorite songs into the fourth learning model M4 and acquires the preference for each fashion brand based on the output values ​​from each output node of the fourth learning model M4. Specifically, the control unit 11 identifies the output node that outputs the largest output value among the output values ​​from each output node for each fashion brand, and identifies the preference associated with the identified output value as the preference for that fashion brand.

[0069] By using the above-described process, when the fourth learning model M4 is used, a user's preference for fashion brands can be estimated from the artist names and song titles of the user's favorite songs. In this embodiment, it is not necessary to obtain responses to a questionnaire such as that shown in FIG. 2 in step S21; only the artist names and song titles of the user's favorite songs need to be obtained. Furthermore, the artist names and song titles of the user's favorite songs may be extracted, for example, from a music distribution service's playlist, playback history, or favorites. In this case, the user can estimate their preference for fashion brands that match their sensibilities (tastes) simply by using a music distribution service. The modifications described in the above-described embodiments 1 to 3 can also be applied to this embodiment.

[0070] (Embodiment 5) In the above-described embodiments 1 to 4, a user's fashion brand preference is estimated from the user's responses to a music questionnaire or information about songs selected by the user. Therefore, it is possible to provide product or service recommendation information taking into account the user's fashion brand preference. In addition, this embodiment describes an information processing device that can compare, for each of a plurality of fashion brands, the sensibilities (values ​​of the sensibility factors) of users who have a high preference for each fashion brand. The information processing device of this embodiment has a configuration similar to that of the information processing device 10 of embodiment 1 shown in FIG. 1, and therefore a description of the configuration will be omitted.

[0071] FIG. 12 is a flowchart showing an example of a process for comparing the sensitivities of users with a high preference for each fashion brand, and FIGS. 13A and 13B are explanatory diagrams showing example screens. The information processing device 10 of this embodiment can execute a process similar to that shown in FIG. 5 . In step S14 of FIG. 5 , the numerical values ​​for each emotional factor representing the sensitivities of the survey respondent are associated with the survey respondent's preferences for the fashion brand and stored in the storage unit 12. In this embodiment, the numerical values ​​for each emotional factor and the survey respondent's preferences for the fashion brand (i.e., training data for the second learning model M2) stored in the storage unit 12 are used to compare the sensitivities of users with a high preference for each fashion brand. If the information processing device 10 has the training data DB shown in FIG. 10 stored in the storage unit 12, the numerical values ​​for each emotional factor and the survey responses to the fashion brands stored in the training data DB may also be used.

[0072] The control unit 11 of the information processing device 10 selects one of the fashion brands (S41) and obtains the numerical values ​​of each emotional factor of the survey respondents who have a high preference for the selected fashion brand (S42). Here, the control unit 11 extracts the survey respondents who have the highest preference for the selected fashion brand from among multiple fashion brands and reads the numerical values ​​of each emotional factor of each extracted survey respondent from the storage unit 12. The control unit 11 then calculates the average value of each emotional factor based on the read numerical values ​​of each emotional factor of each survey respondent (S43). The control unit 11 stores the average value of each emotional factor in association with the fashion brand information in the storage unit 12. The control unit 11 determines whether there are any unprocessed fashion brands for which the above-described processing has not been performed (S44). If it is determined that there are unprocessed fashion brands (S44: YES), the control unit 11 returns to the processing of step S41, selects one of the unprocessed fashion brands (S41), and performs the processing of steps S42 and S43 for the selected fashion brand. The control unit 11 repeats the processing of steps S41 to S44 until it determines that there are no unprocessed fashion brands, thereby obtaining the average value of the numerical values ​​of each emotional factor for the survey respondents with the highest preference for each fashion brand.

[0073] If the control unit 11 determines that there are no unprocessed fashion brands (S44: NO), it displays, for each fashion brand, the average values ​​of the numerical values ​​of each affective factor corresponding to each fashion brand calculated in step S43 and stored in the memory unit 12 (S45). For example, the control unit 11 generates a screen as shown in FIG. 13A and displays it on the display unit 15. The screen of FIG. 13A displays the numerical values ​​of the first to fifteenth affective factors (average values ​​of the numerical values ​​of each affective factor) for each fashion brand. This allows the average affective feelings (numerical values ​​of each affective factor) of users with a high preference for each fashion brand to be displayed in a graph. The graph shown in FIG. 13A allows the affective feelings (numerical values ​​of each affective factor) of users who prefer each fashion brand to be compared. For example, in the example of FIG. 13A, it can be seen that the affective feelings of users who prefer the first and third brands are similar. It can also be seen that the affective feelings of users who prefer the fourth brand are not similar to the affective feelings of users who prefer the first to third brands.

[0074] Next, the control unit 11 selects two of the fashion brands (S46) and obtains the numerical values ​​of each emotional factor of the survey respondents who have a high preference for the two selected fashion brands (S47). Again, the control unit 11 extracts the survey respondents with the highest preference for each of the two selected fashion brands, reads the numerical values ​​of each emotional factor of each extracted survey respondent from the storage unit 12, and calculates the average value of each emotional factor. The control unit 11 then calculates the similarity between the average values ​​of the emotional factors between the two fashion brands (S48). The similarity may be, for example, a correlation coefficient or cosine similarity, but other similarity measures may also be used. The control unit 11 stores the similarity between the average values ​​of each emotional factor in the storage unit 12, corresponding to each pair of fashion brands.

[0075] The control unit 11 determines whether there are any unprocessed pairs of fashion brands (two fashion brands) for which the calculation process of the similarity of affective factors has not been performed (S49), and if it determines that there are any unprocessed pairs of fashion brands (S49: YES), the control unit 11 returns to the process of step S46. The control unit 11 then selects an unprocessed pair of fashion brands (S46) and performs the processes of steps S47 to S48 for the selected pair of fashion brands. The control unit 11 repeats the processes of steps S46 to S49 until it determines that there are no unprocessed pairs of fashion brands. In this way, it is possible to obtain, for each pair of fashion brands, the similarity of affective factors (sensibility similarity) for the survey respondents with the highest preference for each fashion brand.

[0076] If the control unit 11 determines that there are no unprocessed fashion brand pairs (S49: NO), it displays the similarity of the affective factors corresponding to each pair of fashion brands calculated in step S48 and stored in the memory unit 12 for each pair (S50). For example, the control unit 11 generates a screen such as that shown in FIG. 13B and displays it on the display unit 15. The screen of FIG. 13B displays the similarity of the affective factors for each pair of fashion brands. This makes it possible to display a list of the similarity of the sensibilities (values ​​of each affective factor) of users who have a high preference for each fashion brand for each pair of fashion brands. The table shown in FIG. 13B makes it possible to grasp the similarity of the sensibilities (values ​​of each affective factor) held by users who prefer each fashion brand.

[0077] Through the above-described processing, in this embodiment, it is possible to grasp the similarity in sensibilities of users who like each fashion brand (users who have a high preference for each fashion brand). Therefore, for example, if a fashion brand that a user has a high preference for can be estimated from the user's responses to a questionnaire about music, it is possible to estimate, in addition to the estimated fashion brand, other fashion brands that the user is likely to like from the similarity in sensibilities of users who like each fashion brand. Therefore, it is possible to estimate not only the fashion brands that a user has a high preference for, but also other fashion brands that the user is likely to like, and this can be used for marketing information.

[0078] The configuration of this embodiment is applicable to the information processing devices 10 of the above-described embodiments 1 to 4, and similar effects can be obtained even when applied to the information processing devices 10 of the above-described embodiments 1 to 4. Furthermore, the modified examples described in the above-described embodiments 1 to 4 can also be applied to this embodiment as appropriate.

[0079] The embodiments disclosed herein are to be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims.

[0080] The features described in each of the above-mentioned embodiments can be combined with each other. Furthermore, the independent claims and dependent claims described in the claims can be combined with each other in any and all combinations, regardless of the reference format. Furthermore, while the claims use a format in which a claim references two or more other claims (multiple claim format), this is not limited to this format. A multiple claim (multi-multi claim) that references at least one other multiple claim may also be used.

[0081] REFERENCE SIGNS LIST 10 Information processing device 11 Control unit 12 Memory unit 13 Communication unit 14 Input unit 15 Display unit M1 First learning model M2 Second learning model M3 Third learning model M4 Fourth learning model

Claims

1. A program for causing a computer to execute a process of acquiring content data related to content selected by a user and outputting preference information regarding the user's preference for products or services based on the acquired content data.

2. The program according to claim 1, wherein the acquired content data is input to a first learning model trained to output values for each of a plurality of types of sentiment factors representing the user's sentiment when the content data is input, values for each of the plurality of types of sentiment factors are obtained, and the values obtained from the first learning model for each of the plurality of types of sentiment factors are input to a second learning model trained to output preference information regarding the preference for products or services when the values for each of the plurality of types of sentiment factors are input, and the preference information is obtained.

3. The program according to claim 1, wherein the acquired content data is input to a third learning model trained to output preference information regarding the preference for products or services based on the sentiment of the user who selected the content when the content data is input, and the preference information is obtained.

4. The program according to any one of claims 1 to 3, wherein the content data includes information regarding content or an artist selected by the user.

5. The program according to any one of claims 1 to 3, wherein the content data includes information regarding content or an artist included in a playlist or viewing history provided by a content streaming service.

6. The program according to any one of claims 1 to 3, wherein the preference information includes preference levels for a plurality of brands.

7. The program according to claim 1, wherein the content is music content selected by the user, and when content data related to the music content is input, the acquired content data is input to a fourth learning model trained to output preference levels for a plurality of brands based on the sentiment of the user who selected the music content, and the preference levels for the plurality of brands are obtained.

8. The program according to claim 6, wherein the computer executes a process of obtaining, for each of the plurality of brands, a value for each of a plurality of aesthetic factors representing the aesthetics of users who prefer each brand, and outputting the values for the plurality of aesthetic factors for each of the plurality of brands.

9. The program according to claim 6, wherein the computer executes a process of obtaining, for each of the plurality of brands, a value for each of a plurality of aesthetic factors representing the aesthetics of users who prefer each brand, calculating a similarity of the values for the plurality of aesthetic factors in two of the plurality of brands, and outputting the similarity between the two brands.

10. An information processing method in which a computer executes a process of obtaining content data related to content selected by a user and outputting preference information related to the preference of the user for a product or service based on the obtained content data.

11. An information processing apparatus having a control unit, wherein the control unit obtains content data related to content selected by a user and outputs preference information related to the preference of the user for a product or service based on the obtained content data.

Images