JP7055300B2 - Information processing equipment, methods and programs - Google Patents

Description

The disclosure herein relates to information processing devices, methods and programs.

It is known to generate new data using a trained model trained using trained data.

As a method of obtaining such a trained model, a technique using deep learning such as a hostile network (GAN: Generative Adversarial Networks) and a variational self-encoder (VAE) is known. For example, Japanese Patent Application Laid-Open No. 2018-139071 (Patent Document 1) describes a learning method using a hostile network (GAN). The GAN consists of two networks, a generator and a discriminator. The discriminator identifies whether the input data is training data or is generated by the generator. The generator attempts to generate data similar to the training data. The generator tries to generate more elaborate fake data, and the discriminator learns to discriminate the fake data more accurately. By repeating the learning between the generator and the discriminator in this way, the generator can generate data that is indistinguishable from the training data.

Japanese Unexamined Patent Publication No. 2018-139071

With GAN, it is possible to generate an image that is at first glance indistinguishable from the training data. On the other hand, users have various hobbies and tastes. Therefore, the various images generated by GAN include those that do not suit the user's hobbies and tastes. When an image is to be generated by the generator in consideration of what the user wants, it is necessary to prepare a trained model that reflects these user's wishes in advance.

However, since it takes time to build the trained model, it also takes a long time to generate an image according to the user's wishes.
Therefore, there is a need for a technique that makes it even easier to generate an image according to the user's wishes.

According to one embodiment shown in the present disclosure, the information processing apparatus includes a trained generative model, a storage unit for storing a function for estimating a variable input to the generative model, and one or more computer processors. To prepare for. The one or more computer processors obtain the first input value associated with the user by executing a computer-readable instruction, and the estimation result by inputting the first input value to the estimation model. A step of estimating the first variable, a step of generating the first data by inputting the first variable into the generation model, and a step of presenting the generated first data to the user are performed. ..

According to one embodiment, the time for generating data according to the user's wishes can be shortened, and the user's operability is further improved.

It is a figure which shows the hardware configuration example of the terminal apparatus 10 of this embodiment. It is a figure which shows the functional structure of the terminal apparatus 10. It is a figure explaining the detail of the generation model learning module 1042. It is a figure which shows the data structure of the data held by the terminal apparatus 10. It is a figure which shows the process which the estimation model determination module 1043 obtains an estimation model 183. It is a figure which shows the screen example in this embodiment. It is a flowchart which shows the operation of a terminal apparatus 10.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In the drawings for explaining the embodiments, the same components are designated by the same reference numerals in principle, and the description is not repeated.

<Overview>
In the present embodiment, by giving a latent variable to a trained model for generating data (for example, image data), data corresponding to the latent variable is obtained, and the data is tagged by user input or the like. Set the information. By giving various latent variables to the trained model, data corresponding to the latent variables is generated. Therefore, (i) input latent variables, (ii) data generated by the trained model, and (iii) tags set in the data are accumulated in the database.

Here, it is assumed that a model for estimating a latent variable is prepared when tag information is specified. In this case, data can be obtained by estimating the latent variable based on the designated tag information and inputting the estimated latent variable into the trained model. When the user specifies the tag information by the input operation based on the hobby preference etc., it can be said that the data generated by the trained model based on the latent variable estimated by this reflects the user's hobby preference etc.

<Structure>
FIG. 1 is a diagram showing a hardware configuration example of the terminal device 10 of the present embodiment. The terminal device 10 is a device operated by each user. The terminal device 10 is realized by, for example, a stationary PC (Personal Computer), a laptop PC, a smartphone compatible with a mobile communication system, a mobile terminal such as a tablet, or the like. The terminal device 10 includes a processor 101, a memory 102, a storage unit 103, an input device 104, an output device 106, and a communication IF (Interface) 107.

The terminal device 10 is communicably connected to another device (not shown) such as a server via a network. The terminal device 10 is connected to the network by communicating with a communication device such as a wireless base station corresponding to a predetermined communication standard and a wireless LAN router corresponding to a wireless LAN standard such as IEEE802.11. The user can perform data input / output operations via the server by communicating with the server from, for example, a PC installed in the office or a mobile terminal used when going out. For example, the user can communicate with the server by the terminal device 10, refer to various data managed by the server, and input data regardless of whether he / she is staying in the office or going out.

The processor 101 is hardware for executing an instruction set described in a program, and is composed of an arithmetic unit, registers, peripheral circuits, and the like. The terminal device 10 is configured to include one or more computer processors.

The memory 102 is for temporarily storing a program, data processed by the program, or the like, and is, for example, a volatile memory such as a DRAM (Dynamic Random Access Memory).

The storage unit 103 is a storage device for storing data, and is, for example, a flash memory or an HDD (Hard Disc Drive).

The input device 104 is a device for receiving an input operation from a user (for example, a touch-sensitive device, a pointing device such as a mouse, a keyboard, a microphone, a motion sensor, etc.).

The output device 106 is a device (display, speaker, etc.) for presenting information to the user.

The communication IF 107 is an interface for inputting / outputting signals because the terminal device 10 communicates with an external device.

In the example of FIG. 1, the terminal device 10 is configured by a single computer, but is not limited to this, and may be configured by, for example, a plurality of computers connected via a network. ..

Further, although not shown, the terminal device 10 may include a position information sensor that detects the position of the terminal device 10 by a GPS (Global Positioning System) module or the like. The terminal device 10 can acquire the current location of the user who uses the terminal device 10 by the position information sensor. In addition to this, the terminal device 10 may include a vibrator for vibrating the terminal device 10 and notifying the user by the vibration.

Next, the functional configuration of the terminal device 10 will be described.
FIG. 2 is a diagram showing a functional configuration of the terminal device 10. As shown in FIG. 2, the terminal device 10A exhibits functions as a communication unit 107A, a storage unit 103A, and a control unit 101A.

The control unit 101A exhibits the functions shown as various modules when the processor of the terminal device 10A performs processing according to a program. Details of each module will be described later.

The storage unit 103A stores data and programs used by the terminal device 10. The storage unit 103A stores the generation model 181, the generation image database 182, the estimation model 183, and the user action log database 184.

The generative model 181 is a model obtained by a generative model learning module described later. In the present embodiment, the generative model 181 generates image data by inputting a latent variable.

The generated image database 182 is a database including latent variables given as inputs to the generated model 181, an image generated by the generated model 181 and tag information set for the image.

The estimation model 183 is a model obtained by the estimation model determination module 1043 described later.

The user action log database 184 is a database that records the contents of user actions as a log. Information on the content of the user's actions includes, for example, life log data acquired by a sensor device of a device possessed by the user, such as a terminal device 10 or a wearable device worn by the user. For example, the user's location information, the user's voice, the user's biometric information and other information are included in the life log data. In addition to this, the information on the content of the user's actions includes the history of the user accessing the website with a browser or the like. The access history includes, for example, when a user browses an EC (electronic commerce) site, the time during which the user browsed the website, the designation of a product, the purchase, and other information.

The communication unit 107A performs a process for the terminal device 10A to communicate with an external device.

The operation content acquisition module 1041 acquires the user's input operation to the input device 104 and determines the operation content.

The generative model learning module 1042 learns a model for generating data. The generation model learning module 1042 stores the trained model in the storage unit 103 as the generation model 181. Details will be described later, but in this embodiment, it is assumed that a model is obtained by using deep learning. Specifically, the generative model learning module 1042 uses a hostile network (GAN) to learn a model for generating image data.

The estimation model determination module 1043 determines a function for estimating a variable (latent variable) given as an input to the generative model 181. Details will be described later. The estimation model determination module 1043 stores the determined function in the storage unit 103 as an estimation model 183.

The latent variable estimation module 1044 utilizes the estimation model 183 to obtain variables to be input to the generative model 181. Although the details will be described later, the latent variable estimation module 1044 acquires the input value associated with the user and uses the acquired input value as the input of the estimation model 183 to input the latent variable to be input to the generation model 181. Obtainable.

The image generation module 1045 generates image data as data by inputting the latent variable obtained by the latent variable estimation module 1044 into the generation model 181.

The action log analysis module 1046 calculates a variable to be input to the latent variable estimation module 1044 based on the information of the content of the user's action accumulated in the user action log database 184. That is, since the information on the content that the user has acted on includes the information generated by various sensor devices and the information on the content that the user has acted on the website, these information are integrated and input to the latent variable estimation module 1044. Convert to a latent variable to do.

FIG. 3 is a diagram illustrating the details of the generative model learning module 1042. As shown in FIG. 3, the generation model learning module 1042 includes a generator 301, a training data set 302, and a classifier 303. As described above, the generative model learning module 1042 uses GAN as an example of a learning method for a hostile network, but is not limited to this. In FIG. 3, "X" is an input variable input to the classifier 303. “Y” is an output variable output by the classifier 303. "Z" is an input variable (latent variable) input to the generator 301.

The classifier 303 is trained so that it can discriminate whether the input variable X is the training data set 302 or the data (generated data) generated by the generator 301. In the example of FIG. 3, the classifier 303 learns so that the output Y becomes 0 with respect to the data generated from the generator 301. The classifier 303 learns so that the output variable Y becomes 1 with respect to the input variable X selected from the training data set 302. That is, in the classifier 303, the output variable Y is set to 0 when the input variable X is the data generated by the generator 301, and the output variable Y is set to 1 when the input variable X is that of the learning data set 302. As described above, the value of each parameter constituting the classifier 303 is learned.

On the other hand, the generator 301 is trained so that the classifier 303 generates data that cannot be distinguished from the training data of the training data set 302. In the example of FIG. 3, the generator 301 takes the noise z as an input and learns to generate data similar to the training data. For example, when the input variable X is the training data of the training data set 302, the value of each parameter constituting the generator 301 is learned so that the output variable Y becomes 0. By repeating the above learning, the identification accuracy of the classifier 303 is improved, and the generation accuracy of the generator 301 (the accuracy in which the data generated by the generator 301 is similar to the learning data of the learning data set 302) is improved.

The above learning is realized by solving the evaluation function shown in the following (Equation 1). In (Equation 1), "V" is an evaluation value, "D" is a parameter group constituting the classifier 303, "G" is a parameter group constituting the generator 301, E [·] is an expected value, and "x ~". "pdata" corresponds to a set of training data (input variable X) sampled from the data set. Further, "z ~ pz" corresponds to the latent variable Z, D (X) corresponds to the output variable Y when the input variable X is input, and G (Z) corresponds to the data generated when the latent variable Z is input. do.

The first term on the right side of the equation (Equation 1) corresponds to the evaluation value of the classifier 303, and the higher the identification accuracy of the classifier 303, the higher the value. The second term on the right side of the equation (Equation 1) corresponds to the evaluation value of the generator 301, and the classifier 303 erroneously recognizes the data generated by the generator 301 as the data of the learning data set 302. (That is, the more identification errors of the classifier 303), the higher the value.

Therefore, as the learning of the classifier 303 progresses, the first term on the right side of (Equation 1) becomes higher and the second term on the right side becomes lower. Further, as the learning of the generator 301 progresses, the first term on the right side of (Equation 1) becomes lower and the second term on the right side becomes higher.

<Data structure>
FIG. 4 is a diagram showing a data structure of data held by the terminal device 10.

As shown in FIG. 4, each record of the generated image database 182 includes an item "generated image ID", an item "latent variable at the time of image generation", and an item "tag".

The item "generated image ID" is identification information for specifying the image generated by the generated model 181.

The item "latent variable at the time of image generation" is a latent variable given as an input value when the generation model 181 generates an image specified by the item "generated image ID".

The item "tag" indicates the content of the tag information set for the image specified by the item "generated image ID".

The tag information may be information about an object included in the image. As described below, the tag information includes information on one or more items such as "ease of sale", "brand", "age", "color pattern", and "fashion system" of the object.
Objects included in the image include fashion items, daily necessities (electric appliances, etc.), works of art, and other items. The appearance of these items is an object of appreciation for the user, and is a motivation for purchasing.

The tag information may include information about the source of the object (brand name, sales company, etc.). The tag information includes information on when the object was created (date of production, time of sale, time of publication, etc.). The tag information includes information about the color of the object. As described above, the content set in the tag information may be an item referred to for making a purchase decision when the user purchases the object. For example, it is assumed that the user specifies a specific brand as tag information for the image generated by the generation model 181. In this case, the tag information may indicate the degree to which the user evaluates that the object included in the image is "a specific brand". Further, the tag information may include information on the fashion system. The fashion system is a category classified from the user's point of view, such as "casual", "pretty", "conservative", and "garish". In this case, the tag information may indicate the degree to which the user evaluates that the object contained in the image is suitable for a specific system.

The tag information may be information about a user's evaluation item for the object. For example, it includes an evaluation value input by the user for the image. As the evaluation value, for example, when the object included in the image tends to be purchased according to the user's hobbies and tastes such as fashion items, the degree to which the user evaluates the object (“easiness to sell””. , "It seems to be popular", etc.) may be quantified. Further, the evaluation value in the tag information may be a value in which the user's emotion is evaluated. For example, the user may be able to specify "like" for positively evaluating the object and "not so" for negatively evaluating the object for the image.

Further, the object included in the image may include a "product", a "person", an "animal" and the like. The tag information may be information indicating the degree to which the "product", "person", and "animal" are suitable for hobbies and tastes, or may be information on the evaluation value of emotions.

Each record of the user action log database 184 is configured to include the item "site action log", the item "SNS action log", and the item "life log data" in association with the user identification information (user ID). Has been done.

The item "site action log" shows a log of actions taken by a user visiting a website using a browser or the like and within the website (for example, an EC site). For example, the "site action log" includes information on the timing of accessing a website, information on a page viewed by a user on the website, information on a page or image on which the user made a purchase on the website, and the like. ..

The item "SNS action log" indicates a log of actions taken by the user in the SNS (Social Network Service). The action log includes, for example, the user browsing the information of another user, the user taking an action on data such as an image (distribution of pressing "Like"), and the like.

The item "life log data" is data of various life logs acquired by the terminal device 10 or the wearable device. For example, user position information, voice information, and the like are included. The life log information also includes data generated by the user operating the terminal device 10 or the like. For example, when the user executes an application for creating a memo of the terminal device 10, the content of the memo can be the information of the life log.

FIG. 5 is a diagram showing a process in which the estimation model determination module 1043 obtains the estimation model 183.

As shown in the state of FIG. 5A, first, an image is generated by inputting a latent variable to the generation model 181 (result of training the generator 301). Here, it is assumed that the latent variable has a value of N (N is an integer of 1 or more) dimension. Tag information is set for the image generated by the generative model 181.

As shown in the state of FIG. 5B, the latent variable and the tag information are stored in association with each other in the generated image database 182. For example, the user's evaluation value (whether or not the product included in the image is liked, etc.) may be used as the tag information. The state of FIG. 5B conceptually shows the generated image database 182 as described above. As shown in the state of FIG. 5B, (i) the latent variable and (ii) the tag information to be set are associated with each other for the plurality of latent variables. An estimation model 183 is obtained by performing deep learning, multiple regression analysis, or the like on these data sets.

FIG. 5 shows an example of obtaining an estimation model 183 by using a one-dimensional value as the tag information, but the tag information may be a vector having a multidimensional value. For example, the tag information may have a plurality of items and may have an evaluation value for each item. For example, the tag information has an evaluation value related to a fashion brand and an evaluation value related to an age (for example, as an item related to an article such as a fashion item, a plurality of items such as "specific brand-likeness" and "announced age". It may have an evaluation value of another concept such as (item). Further, the tag information has a value corresponding to the first brand and a value corresponding to the second brand (for example, "first brand-likeness" and "second brand" as items related to goods. Multiple items such as "likeness") may have the same concept but different types of evaluation values.

FIG. 6 is a diagram showing a screen example in this embodiment.
The screen example of FIG. 6A shows a situation in which a user's evaluation is accepted for an image generated by giving a latent variable Z as an input value to the generation model 181. As a result, in the generated image database 182, the user's evaluation can be held as tag information in association with the latent variable and the image identification information.

In the screen example of FIG. 6A, as an example, a case where the output device 106 is a display and the input device 104 is a touch-sensitive device is shown. The terminal device 10 displays the image display unit 106A, the evaluation value display unit 106B, and the evaluation input unit 106C on the output device 106 (display).

The image display unit 106A displays an image generated by inputting a latent variable into the generation model 181.

The evaluation value display unit 106B displays the evaluation value input by the user.

The evaluation input unit 106C accepts the input of the evaluation value from the user. As shown in the figure, the evaluation input unit 106C sets the ease of selling (probably fashionable) for the object (fashion item in the illustrated example) displayed on the image display unit 106A as an evaluation item. There is. The user inputs the ease of selling (whether it seems to be popular or not) by the slide bar 106D. The user slides the icon 106E of the slide bar 106D with the finger 106F. In the illustrated example, the terminal device 10 accepts a value in a certain range as an evaluation value from the user. The terminal device 10 accepts the input of the evaluation value according to the position on the slide bar 106D as the user moves the icon 106E left and right on the slide bar 106D, and displays the evaluation value in the evaluation value display unit 106B. Display on.

The terminal device 10 updates the generated image database 182 by receiving an operation of determining the evaluation value of the target evaluation item (for example, “ease of sale”) from the user, and associates it with the identification information of the image. (That is, the evaluation value of the evaluation item is retained as tag information in association with the latent variable when the image is generated by the generation model 181).

The screen example of FIG. 6B shows a situation in which a user specifies an input value, a latent variable is estimated by the estimation model 183, and an image is generated by the estimated latent variable. The terminal device 10 displays the adjustment image display unit 106G, the evaluation value display unit 106H, the evaluation value input unit 106J, the registration reception unit 106L, and the information acquisition unit 106M on the output device 106.

The terminal device 10 adjusts the evaluation value to be input by the user moving the icon 106E on the slide bar 106K displayed on the evaluation value input unit 106J by the finger 106F. The evaluation value display unit 106H is an area for displaying the evaluation value input by the user.

The terminal device 10 estimates the latent variable by the estimation model 183 each time in response to the operation of changing the evaluation value by the user. The terminal device 10 generates an image using the estimated latent variable as an input value of the generation model 181 and displays the generated image on the adjustment image display unit 106G. As a result, the user can confirm the image displayed on the image display unit 106A every time the input of the evaluation value is updated on the slide bar 106K.

The registration reception unit 106L accepts an operation for the user to save the image displayed on the adjustment image display unit 106G. As a result, the terminal device 10 associates the identification information of the image displayed on the adjusted image display unit 106G with the evaluation value input by the user and the latent variable estimated by the estimation model 183, for example, the generated image database 182. To hold it. The terminal device 10 may determine the estimation model 183 by the estimation model determination module 1043 in response to the update of the generated image database 182 in this way.

The information acquisition unit 106M acquires a value to be input to the estimation model 183 based on the user's life log information. The terminal device 10 reads out the item "life log data" of the user action log database 184 in response to the user's operation to the information acquisition unit 106M, and refers to the read data of various life logs, and the user's preference. Calculate the index value for. As a result, the terminal device 10 can estimate the latent variable by inputting the index value into the estimation model 183, and display the image on the adjusted image display unit 106G by the estimated latent variable and the generation model 181.

The screen example of FIG. 6C shows a phase in which an image is generated by estimating a latent variable by an estimation model 183, as in the screen example of FIG. 6B. Compared with the screen example of FIG. 6 (B), in the screen example of FIG. 6 (B), the evaluation value input by the user is a one-dimensional value, whereas in the screen example of FIG. 6 (C), there are many. It differs in that it is a dimensional value.

The terminal device 10 displays an evaluation value input unit 106P and an evaluation value display unit 106N, which are different from the screen example of FIG. 6B.

The terminal device 10 displays the input reception image 106Q that accepts the input of the multidimensional value from the user in the evaluation value input unit 106P. In the screen example of FIG. 6C, as an example, a three-dimensional (x, y, z) evaluation value is accepted from the user. The terminal device 10 displays a figure whose value in each dimension can be adjusted as the input reception image 106Q. In the illustrated example, the input reception image 106Q has a hexagonal shape. In the hexagonal shape, input of a three-dimensional value is accepted by connecting the vertices. For example, the horizontal direction indicates the input of the value of "x", the upward-sloping direction indicates the input of the value of "y", and the downward-sloping direction indicates the input of the value of "z". The user can input a three-dimensional evaluation value by moving the icon 106E with the input reception image 106Q by the finger 106F.

The terminal device 10 displays the multidimensional input value input by the user in the evaluation value input unit 106P on the evaluation value display unit 106N.

As described above, in the screen example of FIG. 6B and the screen example of FIG. 6C, an example in which the user inputs the evaluation value by the slide bar 106K and the evaluation value input unit 106P has been described. In addition to this, the input value to be input to the estimation model 183 may be determined by accepting a predetermined operation from the user without inputting the evaluation value by the user.

The screen example of FIG. 6D shows a situation in which the terminal device 10 presents a plurality of data such as images to the user and accepts an operation selected by the user to estimate the user's hobbies and tastes. .. The terminal device 10 determines the input value given to the estimation model 183 by predicting the tendency of the user's hobbies and tastes in this way.

As shown in the screen example of FIG. 6D, the terminal device 10 displays a plurality of images (images 106R1, 106R2, 106R3, 106R4) on the output device 106. The image is an image of an article such as a fashion item. These images are given classifications such as brand name, age of production, fashion system, etc. in advance.
In this way, when presenting a sample to the user and estimating a tendency such as a user's hobbies and tastes, music may be presented to the user as a sample. For example, the music presented to the user is classified into categories, and the tendency of the user's hobbies and tastes can be estimated based on the classification given to the music selected by the user. Further, the content of a magazine (fashion magazine, etc.) may be presented to the user. Since a magazine is usually produced by defining a customer segment, it is possible to estimate a tendency such as a user's hobbies and preferences based on the information given to the magazine content selected by the user. In addition, information on a famous person (entertainer, celebrity) may be presented to the user. Based on the information given to these persons (age, fashion tendency, content being produced, etc.), it is possible to estimate the tendency of the user's hobbies and tastes.

As described above, as the terminal device 10 repeats the process of presenting a plurality of images to the user and accepting the selection from the user, the terminal device 10 shows the tendency of the user's hobbies and tastes based on the classification given to the selected image. Can be estimated. The terminal device 10 determines an input value to be input to the estimation model 183 by performing a predetermined process based on the tendency data such as the user's hobbies and preferences estimated in this way.

<Operation>
FIG. 7 is a flowchart showing the operation of the terminal device 10.

In step S701, the terminal device 10 (operation content acquisition module 1041) displays on the output device 106 a UI (User Interface) component that receives an input value to be input to the estimation model 183 from the user. Corresponds to displaying the evaluation value input unit 106J of the screen example of FIG. 6B.

In step S705, the terminal device 10 (latent variable estimation module 1044) inputs the input value specified by the user into the estimation model 183. As a result, the terminal device 10 acquires the estimation result of the latent variable.

In step S709, the terminal device 10 (image generation module 1045) generates an image by using the latent variable acquired in step S705 as an input of the generation model 181.

In step S713, the terminal device 10 displays the generated image and the input value specified by the user on the output device 106. It corresponds to displaying the adjustment image display unit 106G and the evaluation value display unit 106H of the screen example of FIG. 6B.

<Modification example>
(1) Learning method of the generative model 181 In the above embodiment, the learning method of the generative model 181 has been described with reference to FIG. 3 and the like. Here, as learning data, an image of a specific brand may be used in an article such as a fashion item. This allows the generative model 181 to generate an image that resembles an article of a particular brand.
In addition, images of a plurality of brands may be used as learning data. As a result, the generative model 181 can generate an image in which a plurality of brands are mixed.

(2) Aspect for Providing an Image Generation Method For example, the image generation method in the present embodiment may be provided on a website such as an EC site. For example, in the case of an EC site that sells fashion items, in response to the user performing a predetermined action on the EC site, an evaluation value is input from the user as shown in the screen example of FIG. 6 (B). An image may be displayed on the adjusted image display unit 106G. By the user performing an operation on the registration reception unit 106L in the screen example of FIG. 6B, information on the fashion item desired by the user can be acquired on the EC site side.

Here, as predetermined actions, for example, the user inputs a keyword to search for an item, the user adds an item to the cart when purchasing the item, and the user performs a payment process for purchasing the item. That, etc. are included. The EC site may display a screen that accepts the input of the evaluation value from the user as in the screen example of FIG. 6B by lottery when the user performs these predetermined actions. This can make the purchase experience on the EC site surprising and motivate you to continue to visit the EC site.

<Additional Notes>
The matters described in the above embodiments are added below.

(Appendix 1) According to one embodiment, an information processing apparatus (10) is provided.
The information processing apparatus (10) includes a generative model (181) that has been trained, a storage unit (103A) that stores an estimation model (183) for estimating variables to be input to the generative model, and one or more computers. It includes a processor (101A). One or more computer processors obtain the first input value associated with the user by executing a computer-readable instruction (1041, 106J, 106P) and the first input value for the estimation model. Is generated to estimate the first variable as an estimation result (1044) and to generate the first data by inputting the first variable into the generation model (1045). The step (106G) of presenting the first data to the user is performed.

(Appendix 2) In (Appendix 1), the storage unit is configured to store the variable input to the generation model and the tag information associated with the data generated by inputting the variable. (182). One or more computer processors perform a step of determining an estimation model (10443, FIG. 5) by analyzing a plurality of stored variables and tag information, which is based on input values for the tag information. Estimate the variable.

(Appendix 3) In (Appendix 2), the generation model generates image data in response to the input of variables, and the tag information is information about an object generated as image data or the object. Contains at least one of the information about the user's evaluation items for (182).

(Appendix 4) In (Appendix 3), the tag information is information about the object, the object includes fashion items, daily necessities, works of art and other articles, and the information about the object includes the source of the article. It contains at least one of information about, when the article was created, and information about the color of the article (182).

(Appendix 5) In (Appendix 3), the tag information is information about the user's evaluation item for the object, and the information about the user's evaluation item for the object is information indicating the degree of conformity with the user's hobbies and tastes, the user. It contains at least one of the information on the evaluation value of the emotion of (182).

(Appendix 6) In (Appendix 5), the object includes a product, a person, or an animal, and information on a user's evaluation item regarding the object is the product, a person, or an animal included in the generated image data. It includes at least one of information indicating the degree of suitability for hobbies and tastes and information on evaluation values of emotions (182).

(Appendix 7) In any one of (Appendix 4) to (Appendix 6), one or more computer processors perform a step of learning the generated model based on the training data (1042), and in the learning step, one. Alternatively, in a step of learning a generation model using image data of a plurality of specific brand products as training data and acquiring a first input value, information about the brand is accepted as the first input value.

(Appendix 8) In any of (Appendix 1) to (Appendix 7), in the step of acquiring the first input value, the user is presented with the first interface for accepting the input of the first input value. Then, in the estimation step (106J, 106P), the input result for the first interface is input to the estimation model.

(Appendix 9) In (Appendix 8), the first input value includes information on a plurality of input items, and in the step of acquiring the first input value, the user is informed of the first input for the plurality of input items. In the step of accepting input of a value (106P) and estimating, the first variable is estimated in response to the first input value including information of a plurality of input items.

(Appendix 10) In (Appendix 9), in the step of acquiring the first input value, an image (106Q) that accepts multidimensional input is presented to the user as the first interface.

(Appendix 11) In any of (Appendix 8) to (Appendix 10), in the step of presenting, the first interface (106J) for accepting the input of the first input value and the first data to be generated. (106G), an input value (106H), and a second interface (106L) for accepting an operation for confirming the first data by the user are presented to the user on one screen, and one or more computer processors. Performs a step of associating the input value with the first data as tag information and storing it in the storage unit by accepting the operation of confirming (182).

(Appendix 12) In any one of (Appendix 1) to (Appendix 11), one or more computer processors performs a step of acquiring a log of user actions (184), and a step of acquiring a first input value. In, the first input value is acquired based on the log of the user's action.

(Appendix 13) In any of (Appendix 1) to (Appendix 12), predetermined information is presented to the user by the third interface (106R1, 106R2, 106R3, 106R4), and based on the content of the response from the user. And get the first input value.

(Appendix 14) In any of (Appendix 1) to (Appendix 13), one or more computer processors perform a step of learning a generative model based on the training data (1042), and in the learning step, the data. The generative model is trained by a learning method of a hostile network including a generator (301) that generates the data, a training data (302), and a discriminator (303) that identifies the data generated by the generator.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited to the above embodiments. Further, the above-described embodiment can be variously improved or modified without departing from the gist of the present invention. Further, the above embodiments and modifications can be combined.

Claims (16)

A storage unit for storing a trained generative model and an estimation model for estimating variables to be input to the generative model, and one or more computer processors.
The one or more computer processors may execute computer-readable instructions.
The step to get the first input value associated with the user,
A step of estimating the first variable as an estimation result by inputting the first input value to the estimation model, and
The step of generating the first data by inputting the first variable into the generative model, and
An information processing apparatus that performs a step of presenting the generated first data to the user. The storage unit is configured to store a variable to be input to the generation model and tag information associated with the data generated by inputting the variable.
The one or more computer processors
By analyzing the plurality of stored variables and the tag information, a step of determining the estimation model is performed.
The information processing apparatus according to claim 1, wherein the estimation model estimates the variable according to an input value related to the tag information. The generative model generates image data in response to the input of variables.
The information processing apparatus according to claim 2, wherein the tag information includes at least one of information about an object generated as image data and information about a user's evaluation item for the object. The tag information is information about the object, and is
The object includes fashion items, daily necessities, or works of art.
The information processing apparatus according to claim 3, wherein the information regarding the object includes at least one of information regarding the source of the article, information about the time when the article was generated, and information regarding the color of the article. The tag information is information about a user's evaluation item for the object, and is
The information processing apparatus according to claim 3, wherein the information regarding the user's evaluation item regarding the object includes at least one of information indicating the degree of conformity with the user's hobbies and tastes and information on the evaluation value of the user's emotions. The object includes a product, a person, or an animal.
The information regarding the user's evaluation item regarding the object is at least one of information indicating the degree to which the product, person or animal contained in the generated image data is suitable for the hobby and taste, and information on the evaluation value of the emotion. The information processing apparatus according to claim 5, comprising the above. The one or more computer processors
Based on the training data, perform steps to train the generative model,
In the learning step, the generation model is trained using image data of one or more specific brand products as training data.
The information processing apparatus according to any one of claims 4 to 6, which receives information about the brand as the first input value in the step of acquiring the first input value. In the step of acquiring the first input value, the user is presented with a first interface for accepting the input of the first input value.
The information processing apparatus according to any one of claims 1 to 7, wherein in the estimation step, an input result for the first interface is input to the estimation model. The first input value includes information on a plurality of input items, and includes information on a plurality of input items.
In the step of acquiring the first input value, the user is accepted to input the first input value for a plurality of input items.
The information processing apparatus according to claim 8, wherein in the estimation step, the first variable is estimated in response to the first input value including information of the plurality of input items. The information processing apparatus according to claim 9, wherein in the step of acquiring the first input value, an image that accepts a multidimensional input is presented to the user as the first interface. In the step presented, the first interface for accepting the input of the first input value, the generated first data, the first input value, and the first data are described. A second interface for accepting an operation confirmed by the user is presented to the user on one screen, and the user is presented with the second interface.
The one or a plurality of computer processors receive the operation to be determined, and perform a step of associating the first input value with the first data as tag information and storing it in the storage unit, according to claim 8. The information processing apparatus according to any one of 10. The one or more computer processors perform the step of acquiring a log of the user's actions.
The information processing apparatus according to any one of claims 1 to 11, wherein in the step of acquiring the first input value, the first input value is acquired based on the log of the user's action. The third interface according to any one of claims 1 to 12, wherein predetermined information is presented to the user and the first input value is acquired based on the content of the response from the user. Information processing device. The one or more computer processors
Based on the training data, perform steps to train the generative model,
In the learning step, the generation model is trained by a learning method of a hostile network including a generator that generates data and a discriminator that identifies the training data and the data generated by the generator. The information processing apparatus according to any one of claims 1 to 6 . A method performed by a computer
The computer comprises a trained generative model and a storage unit for storing an estimation model for estimating variables input to the generative model, and one or more computer processors.
The method is such that the one or more computer processors execute computer-readable instructions.
The step to get the first input value associated with the user,
A step of estimating the first variable as an estimation result by inputting the first input value to the estimation model, and
The step of generating the first data by inputting the first variable into the generative model, and
A method comprising performing a step of presenting the generated first data to the user. A program run by a computer
The computer comprises a trained generative model and a storage unit for storing an estimation model for estimating variables input to the generative model, and one or more computer processors.
The program can be applied to the one or more computer processors.
The step to get the first input value associated with the user,
A step of estimating the first variable as an estimation result by inputting the first input value to the estimation model, and
A step of generating the first data by inputting the first variable into the generative model, and
A program for executing a step of presenting the generated first data to the user.

Images