JP6581263B1 - Information processing apparatus, information processing method, and program - Google Patents

Abstract

An environment for selecting an optimal product image to be presented to a user by simple processing is provided. A product image selection device includes presentation image information for a plurality of types of images presented to a user as a product image for a single product, user information of a user who has received presentation of a plurality of types of images, and a plurality of types. Information on selection results based on user preferences for images of the user, and a selection information management unit that links and manages information on each user, and information on each user is classified based on information on selection results of a plurality of users for a plurality of products A user cluster management unit that manages user clusters, a product cluster management unit that manages a product cluster in which each product is classified based on a user's answer, and a product is managed by the user cluster management unit and the product cluster management unit. Multiple product images to be presented to any target user belonging to the user cluster based on the relationship between the clusters Comprising a presentation image selection unit that selects from the presented candidate image. [Selection] Figure 3

Description

  The present invention relates to an information processing apparatus, an information processing method, and a program, and more particularly, to a technique for selecting an optimal image to be presented to a user from presentation candidate images.

When presenting product information in electronic commerce, an image about a product (product image) plays an important role in determining the user's willingness to purchase.
Therefore, when a product image is presented as product information, it is important to select an appropriate product image.
For example, in the following Patent Document 1, a user is made to answer a question related to a psychological test, the psychological characteristics of the user are analyzed based on the answer information, and the users are grouped. And the structure by which the template of the web page considered appropriate based on a group is selected, and the merchandise information containing a merchandise image is provided is disclosed. Here, the psychological characteristics include basic psychological characteristics related to the personality of the user and purchase-related psychological characteristics related to factors that are influenced when the user purchases the product. Basic psychological characteristics are generally accepted characteristics in the field of psychology, and include neural factors, extrovert factors, creative exploratory factors, diligence factors, and harmony factors. The purchase-related psychological characteristics include purchase attitude classification, public self-consciousness, and self-esteem.

JP 2003-346046 A

It is not easy to generate questions considering such psychological characteristics.
Therefore, an object of the present invention is to select an optimal product image to be presented to the user by simple processing.

An information processing apparatus according to the present invention includes presentation image information about a plurality of types of images presented to a user as a product image for one product, user information about a user who has received the presentation of the plurality of types of images, Information on selection results based on user preferences for types of images, and a selection information management unit that links and manages information, and information on each user is classified based on information on the selection results of a plurality of users for a plurality of products A user cluster management unit that manages a user cluster, a product cluster management unit that manages a product cluster in which each product is classified based on a user's answer, and the user cluster management unit and the product cluster management for a certain product Product images to be presented to any target user belonging to the user cluster based on the relationship between the clusters managed by And presenting an image selecting section for selecting from a plurality of presentation candidate image as an image, those having a.
A plurality of product images for a certain product are presented, and an image selected by the user is likely to be an image that the user feels attractive about the product. According to this configuration, such image selection information is managed in association with user information.

In the processing for selecting a presentation image to be presented to the target user from a plurality of presentation candidate images for a certain product, the presentation image selection unit of the information processing apparatus described above selects any type of image of the certain product by the target user. If there is no history of selection, a presentation candidate image may be selected as a presentation image to the target user based on a selection tendency in the user cluster to which the target user belongs with respect to the image of the certain product.
Thereby, even if there is no selection information of the target user for the presentation candidate image, appropriate image selection is performed based on selection information of other users belonging to the same user cluster.

The information processing apparatus further includes an image evaluation axis setting unit for setting an evaluation axis representing an image tendency for each product image, and the evaluation axis is a selection tendency in the user cluster in the presentation image selection unit. It may be used for the determination.
Accordingly, an evaluation axis that can be a factor for selecting each product image based on the selection tendency of each user with respect to the product image is appropriately set.

  The product cluster management unit of the information processing apparatus described above includes, for each product image, the same evaluation axis for images whose evaluation values based on the same evaluation axis determined by the image evaluation axis setting unit are equal to or greater than a first threshold. An image having an evaluation axis whose evaluation value is equal to or higher than a second threshold value that is higher than the first threshold value among the same evaluation axes included in the product image that is determined to be an image that has the same evaluation axis. The images may be determined as images having a similar tendency, and the presentation image selection unit may select the presentation image as a presentation image to the target user based on the similarity tendency.

The user cluster management unit of the information processing apparatus described above may generate the user cluster based on a combination of a plurality of the selection results.
Instead of creating a user cluster in advance, a user cluster is generated according to a combination of selected images actually selected by the user. Therefore, for example, a user cluster for a combination of selected images that no one can select is not generated in vain.

The user cluster management unit and / or the product cluster management unit of the information processing apparatus described above may generate a new cluster when determining that the relationship between individual data in the generated cluster is more than a predetermined value. Good.
When a certain variation is seen in the selection information of each user belonging to the same user cluster, that is, when the information is dispersed, the similarity of the tendency of the user cluster may be reduced. According to this configuration, reclassification of each user belonging to such a user cluster is performed.

The information processing method according to the present invention includes presentation image information about a plurality of types of images presented to a user as a product image for one product, user information of a user who has received the presentation of the plurality of types of images, Selection information management function for linking and managing information on selection results based on user preferences for types of images, and information on each user based on information on the selection results of a plurality of users for a plurality of products A user cluster management function for managing a user cluster, a product cluster management function for managing a product cluster in which each product is classified based on a user's answer, and the user cluster management function and the product cluster management for a certain product. In order to present to any target user belonging to the user cluster based on the relationship between the clusters managed by the function And presenting an image selection function that selects from a plurality of presentation candidate image as presentation image product images, the one in which the information processing device executes.
With this information processing method, it is possible to select an appropriate product image for the user with simple processing.
The program according to the present invention is a program that causes an information processing apparatus to execute a procedure corresponding to each of the above steps. A storage medium according to the present invention stores the above program. With these, the processing of the information processing apparatus described above is realized.

  According to the present invention, it is possible to select an appropriate product image for the user with a simple process.

It is explanatory drawing of the network containing the goods image selection apparatus of embodiment of this invention. It is a block diagram of the computer apparatus of an embodiment. It is explanatory drawing of a function structure of the product image selection apparatus of embodiment. It is a figure which shows the example of a question web page. It is a figure for demonstrating the example from which a different product image is selected for every user. It is a figure for demonstrating another example from which a different product image is selected for every user. It is a figure for demonstrating the example of the evaluation axis | shaft set to a product image. It is a figure for demonstrating the example of the product image information memorize | stored in product DB. It is a figure for demonstrating the user cluster information memorize | stored in cluster DB. It is a figure for demonstrating the goods cluster information memorize | stored in cluster DB. It is a flowchart of an answer acquisition process. It is a flowchart of a user clustering process. It is a figure for demonstrating calculation of the distance d. It is a flowchart of a product clustering process. It is a figure for demonstrating the evaluation tendency of a product image. It is a figure for demonstrating the selection tendency for every user cluster. It is a flowchart of a presentation image selection process. It is a flowchart of the modification of goods clustering processing. It is a flowchart of the modification 2 of goods clustering processing. It is a flowchart of a preference evaluation axis extraction process. It is a flowchart of the modification of a presentation image selection process. It is a figure which shows an example of the product image which each user selected. It is a flowchart which shows the modification 2 of a presentation image selection process. It is a flowchart which shows the modification 2 of a presentation image selection process.

Hereinafter, embodiments will be described with reference to the accompanying drawings.

<1. System configuration>
The configuration of the entire network system including the product image selection device 7 according to the present embodiment will be described with reference to FIG.
In the network system 1, various terminal devices can communicate with each other via a communication network 2.
The network system 1 includes an electronic commerce system 3, a user terminal 4, and a store terminal 5.

The configuration of the communication network 2 shown in FIG. 1 is not particularly limited. In addition to the Internet described above, for example, an intranet, an extranet, a LAN (Local Area Network), a CATV (Community Antenna TeleVision) communication network, A virtual private network, a telephone line network, a mobile communication network, a satellite communication network, etc. are assumed.
Various examples of the transmission medium constituting all or part of the communication network 2 are also assumed. For example, IEEE (Institute of Electrical and Electronics Engineers) 1394, USB (Universal Serial Bus), power line carrier, telephone line, etc., infrared, IrDA (Infrared Data Association), Bluetooth (registered trademark), 802.11 wireless It can also be used wirelessly, such as mobile phone networks, satellite lines, and digital terrestrial networks.

  The electronic commerce system 3 includes a shopping server 6 that provides various services used by a user to purchase a product, and a product image selection device 7 that selects a product image to be presented to the user.

The various functions provided by the shopping server 6 are, for example, sold by an administrator of the store (hereinafter referred to as a seller) when there are a plurality of stores that are members of electronic commerce operated using the system. For example, a function for registering product information (product information) to be registered, a function for changing registered product information, and the like. For this purpose, the shopping server 6 has a function of managing member store information, seller information, and product information.
Furthermore, the shopping server 6 orders a product from the seller when a user performs a purchase operation for a product or a function for searching for and presenting a product desired by the user from a group of products handled in electronic commerce. , Payment processing functions that mediate the exchange of payments when products are sold, functions for delivering products to each user, and functions for notifying users when products are purchased and products A function of notifying the seller of the user information of purchasing the product.
When a user purchases a product, information on the destination (address) of the product, credit card number and contact information (e-mail address, etc.) is required. The shopping server 6 has a function of managing user information in order to save time and labor for inputting such information every time a user purchases a product.

The shopping server 6 generates web page data in order to display a web page as a user interface for realizing the above various functions on another information processing device (the user terminal 4 or the store terminal 5). Send.
The web page data is, for example, a structured document file such as HTML (Hyper Text Markup Language) or XHTML (Extensible HyperText Markup Language). The structured document file describes text data such as product descriptions and image data such as product images, and their arrangement and display mode (character color, font, size, decoration, etc.).
Examples of the web page include a login page for allowing a user and a seller to input login information, a search result display page for displaying a product search result, and a product page on which product information is displayed.
The shopping server 6 includes a user / seller authentication function, a function for registering information in various databases, a function for acquiring information from various databases, and the like.

  The product image selection device 7 is an information processing device that selects a product image presented when providing information related to a product to a user. For example, the shopping server 6 transmits to the user terminal 4 a web page on which products as search results are listed in response to a user search request. A product image is posted on the web page together with product price information. The product image selection device 7 is a device that selects which product image is to be displayed when a plurality of product images to be posted on the search result web page are prepared together with price information and the like.

  Moreover, the electronic commerce system 3 may transmit recommendation information about a product that is of interest to the user. When a product image is added to such recommendation information, a selection process by the product image selection device 7 is performed.

Further, in order to select an appropriate product image for each user, the product image selection device 7 performs user clustering to generate a user cluster UC, and performs product clustering to generate a product cluster SC.
Further, the product image selection device 7 performs processing for generating a question for performing each clustering and transmission processing for transmitting the question to the user.

In order to realize the various functions described so far, the electronic commerce system 3 sells a user DB (Database) 50 in which user information is stored, a product DB 51 in which information on products handled in electronic commerce is stored, and products. It includes a store DB 52 that stores information on sellers (stores), a selection information DB 53 that stores selection information for user questions, and a cluster DB 54 that stores information related to user clusters UC and product clusters SC. .
In addition, a web page DB (not shown) in which web page data of various web pages is stored is managed.
These DBs can communicate with the shopping server 6 and the product image selection device 7 of the electronic commerce system 3 via a network such as a LAN, for example, and such a network is particularly limited like the communication network 2. It is not a thing.

  The user terminal 4 is an information processing apparatus used by the user to receive various services provided by the electronic commerce system 3. Specifically, it is used when browsing product information, purchasing products, or the like.

  The store terminal 5 is an information processing device used on the store side that provides product information to the user by registering the product sold by itself in the electronic commerce system 3. The store terminal 5 performs transmission / reception processing for registering, deleting, and changing product information with respect to the electronic commerce system 3.

The user terminal 4 and the store terminal 5 are, for example, a PC (Personal Computer), a feature phone, a PDA (Personal Digital Assistants) having a communication function, or a smart device such as a smartphone or a tablet terminal.

<2. Hardware configuration of computer device>

The hardware configuration of the computer device that constitutes each device (shopping server 6, user terminal 4, store terminal 5, user DB 50, product DB 51, store DB 52, selection information DB 53, cluster DB 54) including the product image selection device 7 As shown in FIG. A CPU (Central Processing Unit) 101 of each computer apparatus executes various processes according to a program stored in a ROM (Read Only Memory) 102 or a program loaded from a storage unit 108 to a RAM (Random Access Memory) 103. To do. The RAM 103 also appropriately stores data necessary for the CPU 101 to execute various processes.
The CPU 101, ROM 102, and RAM 103 are connected to each other via a bus 104. An input / output interface 105 is also connected to the bus 104.
An input unit 106, an output unit 107, a storage unit 108, and a communication unit 109 are connected to the input / output interface 105.
The input unit 106 includes a keyboard, a mouse, a touch panel, and the like.
The output unit 107 includes a display made up of an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube), an organic EL (Electroluminescence) panel, etc., and a speaker.
The storage unit 108 includes an HDD (Hard Disk Drive), a flash memory device, or the like.
The communication unit 109 performs communication processing and inter-device communication via the network 2.
A media drive 110 is also connected to the input / output interface 105 as necessary, and a removable medium 111 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is appropriately mounted, and information can be written to the removable medium 111. Reading is performed.

In such a computer device, data and programs are uploaded and downloaded by communication by the communication unit 109. Data and programs can be exchanged via the removable medium 111.
The CPU 101 performs processing operations based on various programs, whereby each device such as the product image selection device 7 (shopping server 6, user terminal 4, store terminal 5, user DB 50, product DB 51, store DB 52, selection information DB 53, Necessary information processing and communication as the cluster DB 54) are executed.
Note that the information processing devices constituting each device (shopping server 6, user terminal 4, store terminal 5, user DB 50, product DB 51, store DB 52, selection information DB 53, cluster DB 54) such as the product image selection device 7 are shown in FIG. Such a computer apparatus is not limited to a single configuration, and a plurality of computer apparatuses may be configured as a system. The plurality of computer devices may be systemized by a LAN or the like, or may be arranged at a remote place by a VPN using the Internet or the like. The plurality of information processing apparatuses may include an information processing apparatus as a server group (cloud) that can be used by the cloud computing service.

Each function of the information processing apparatus is a function realized by processing executed by the CPU 101 in accordance with the program in the information processing apparatus. However, all or some of the processes described below may be realized by hardware.
Further, when each function is realized by software, each function need not be realized by an independent program. Processing of a plurality of functions may be executed by one program, or one function may be realized by cooperation of a plurality of program modules.
Each function may be distributed among a plurality of information processing apparatuses. Furthermore, one of the functions may be realized by a plurality of information processing apparatuses.

<3. Functional configuration of product image selection device>

A specific configuration of the product image selection device 7 will be described with reference to FIG.
The product image selection device 7 includes a selection information management unit 7a, a user cluster management unit 7b, a product cluster management unit 7c, a presentation image selection unit 7d, and an evaluation axis setting unit 7e.

  The selection information management unit 7a performs a process of presenting the user with a questionnaire-type question for selecting which product image is preferred for a plurality of product images prepared as images for the same product. Also, a process for managing the user's answer to the question as user selection information is performed. Specifically, a process of associating the information of the user who is the answerer of the question with the selection information and storing it in the selection information DB 53 is performed.

A questionnaire-type question presented by the selection information management unit 7a is transmitted to the user terminal 4 as web page data, for example. The question may be sent to a user who wishes in exchange for giving points that can be used in electronic commerce by the electronic commerce system 3, or for all users who use the electronic commerce system 3. It is also possible to manage only the answer information (selection information) of the user who sent the e-mail with the link information of the web page and made the answer. Further, link information to a web page for answering a question may be set on the top page of the electronic commerce provided by the electronic commerce system 3 to prompt the user to answer. Furthermore, you may make it answer this question in the case of user registration performed when a user starts utilization of electronic commerce.
Alternatively, a user who has purchased a product may be caused to select a product image for the product on the purchase screen of the product, and the selection information for each product may be accumulated.

With reference to FIG. 4, the example of the web page (henceforth a question web page) on which the question of the questionnaire format shown to a user was published is demonstrated.
As shown in FIG. 4A, on one question web page 10, two product images 11 and 11 for one product (wallet in the figure) are presented together with a word prompting an answer. Selection buttons 12 and 12 are provided below each product image 11. The user selects the product image 11 that he / she wants to buy from the two product images 11 and 11 and presses a selection button 12 placed below.
When any one of the selection buttons 12 is pressed, the question web page 10 for the next product (clothes case) is presented to the user (FIG. 4B). In this way, each time an answer is made, a question web page for obtaining selection information about the next product is presented to the user.

Since one question web page targets one product, for example, even when the user stops answering in the middle, selection information about the product that has been answered so far can be obtained. In addition, since the question web page includes images, the amount of communication is considered to be relatively large. Even in such a case, the amount of information (number of images) transmitted to the user when the answer is stopped in the middle is less than that of transmitting a question web page with 30 questions at a time. I'll do it.
A question web page in which questions about a plurality of products are provided may be generated and sent to the user.

The selection information management unit 7a acquires user selection information by various methods. The number of products handled in electronic commerce is enormous, and it is difficult to obtain selection information for all products for one user. Therefore, you may make it obtain the selection information about 10-30 goods by making it answer the question comprised about 10-30 questions.
Then, by changing the combination of products to be questioned for each user, as many products as possible may be targeted by any user.

The product that is the subject of the question presented to the user may be changed to include the new product when the new product is handled in electronic commerce, for example. Accordingly, even a relatively new product can be appropriately classified into a product cluster, and an appropriate product image can be presented to the user.
Moreover, the question with respect to a user may be produced | generated by the administrator who manages an electronic commerce system, for example, and may be produced | generated by the manager of the store which has opened a store in electronic commerce.
Specifically, flag information indicating whether or not it is the subject of the question is linked to each product ID and stored in the product DB 51, and when presenting the question to the user, by referring to the flag The product to be questioned may be selected and presented to the user. Of course, you may provide separately DB which memorize | stores the question information shown to a user.

  The product image is stored in the product DB 51, for example. In addition to the product DB 51, a product image DB for storing product images may be provided.

The user cluster management unit 7b classifies the users based on the selection information, and stores the classification information (for example, information that can identify which cluster the user is classified into and information on the classification criteria) in the cluster DB 54. And manage. That is, users are clustered from the viewpoint of what kind of product image is preferred. In the following description, a user cluster is referred to as a user cluster, and is described as user clusters UC1, UC2,.
Specific processing for clustering will be described later.

The product cluster management unit 7c classifies products based on user selection information. That is, products are not classified based on the product genre, but are classified based on the selection tendency of product images. Therefore, there is a possibility that even products with completely different product genres belong to the same cluster.
In the following description, a cluster attached to a product is referred to as a product cluster, and described as product clusters SC1, SC2,.
Specific processing for clustering will be described later.

The presentation image selection unit 7d performs a process of selecting a product image to be presented to the user from candidate product images (presentation candidate images) based on the user cluster UC and the product cluster SC. Some examples of scenes in which product images are selected are shown.
First, FIG. 5 shows an example in which the selected product image is displayed on the search result list page presented on the user terminal 4 when the user performs a product search.
On the search result list page, for example, a product image 13, a product name 14, a price 15, a seller name (store name) 16, an evaluation value 17, and the like are displayed for each product extracted as a search result. The product image 13 displays an image selected by the user cluster UC and the product cluster SC. That is, even if the same product is sold by the same store, the image displayed as the product image 13 differs depending on the user.

FIG. 6 shows an example in which the selected product image is displayed on the product page in which detailed product information is described.
In the product page, for example, in addition to a plurality of product images 13A and 13B, product name 14, price 15, seller name 16, and evaluation value 17, a detailed description column 18 in which a detailed description of the product is displayed, when purchasing Various operation buttons 20, 20,... Such as a number input field 19 for inputting the number of items, an operation button for adding a product to a cart (shopping basket), and an operation button for registering a product as a favorite. Is displayed.
One of the images for the same product is displayed on the plurality of product images 13A and 13B, but in this example, the product image selected as the product image 13A arranged above which is more easily noticed by the user. To be elected. That is, the product image 13 that is different depending on the user is displayed as the product image 13A arranged above.

  The evaluation axis setting unit 7e sets an evaluation axis for a plurality of product images prepared for the same product. When there are a plurality of product images prepared for the same product, the user selects one of them based on the difference between the product images. Therefore, the difference is set as an evaluation branch point for the product image, and at least one difference is set as an evaluation axis. One evaluation axis may be set for each difference, or one evaluation axis may be set for a plurality of differences.

Specifically, this will be described with reference to FIG.
FIG. 7A shows an example of a product image for showing the appearance of the product and a product image for showing the contents of the product. The difference between the two product images is whether the product can be seen or the contents can be seen. Therefore, the evaluation axis given to the two product images shown in FIG. 7A is the “appearance / content evaluation axis”. Whether the user prefers a product image whose appearance can be confirmed or a product image whose contents can be confirmed depends on the user. The set evaluation axis is used in various processes described later.

FIG. 7B is an example in which whether or not text information is included in the product image is different. Therefore, the evaluation axes given to the two product images shown in FIG. 7B are “text presence / absence evaluation axes”.
FIG. 7C is an example of a product image in which the background color of the product is different. The evaluation axes given to the two product images shown in FIG. 7C are “background color cold / warm color evaluation axes”.

FIG. 7D is an example of a product image in which the number of products is different. The evaluation axes given to the two product images shown in FIG. 7D are “single product multiple evaluation axes”.
FIG. 7E is an example of a product image in which whether or not a model appears is different. The evaluation axes given to the two product images shown in FIG. 7E are “model presence / absence evaluation axes”.

In addition to the example shown in FIG. 7, many examples of evaluation axes are conceivable. A plurality of evaluation axes may be given to two product images. For example, if one product image does not show the model and text information is included, and the other product image is an image of the model with the product in hand, and text information is not included, There are two evaluation axes given based on the difference: “model presence / absence evaluation axis” and “text presence / absence evaluation axis”.

<4. Example of information stored in DB>

[4-1. Product image information]
A DB capable of transmitting / receiving data to / from the product image selection device 7 will be described with a specific example.
First, a product image for each product stored in the product DB 51 will be described with reference to FIG.

The product DB 51 stores information related to products (product ID, product name, product page URL information, seller name, etc.) and a plurality of product images.
The product image includes an image for explaining the usage mode of the product, but only the product image targeted in the present embodiment is extracted and shown in FIG.
The product image targeted in the present embodiment is a product image that is selected by the user on the question web page presented to the user, and is a product image (that is, the above-described presentation candidate image) that separates the images presented by the user. is there. Accordingly, an image that is similarly presented to any user, such as an image for explaining how to use the product, is not a target of the present embodiment.

  FIG. 8 shows product images a1, a2, d1, d2, m1, and m2, which are presentation candidate images for three products (products A, D, and M). In each product, one product image is a “first product image” and the other is a “second product image”. Of course, three or more presentation candidate images may be prepared for one product.

  The first product image a1 and the second product image a2 for the product A (wallet) are product images for showing the appearance and contents of the product. That is, the evaluation axis assigned to the first product image a1 and the second product image a2 is the “appearance / content evaluation axis”.

  The first product image d1 and the second product image d2 for the product D are product images for showing the appearance and contents of the product. That is, the evaluation axis given to the first product image d1 and the second product image d2 is also the “appearance / content evaluation axis”.

  The first product image m1 and the second product image m2 for the product M are product images that differ in whether or not text information is included. Therefore, the evaluation axis given to the first product image m1 and the second product image m2 is a “text presence / absence evaluation axis”.

As described above, the plurality of product images for each product stored in the product DB 51 include product images (that is, presentation candidate images) for changing the image presented by the user. Of course, depending on the product, the same product image may be presented to all users. In that case, there is no presentation candidate image, and no evaluation axis is given to each product image. Moreover, the process which selects a presentation image from a presentation candidate image according to the user cluster UC and the goods cluster SC is not performed.

[4-2. User cluster information]
Next, the user cluster UC information stored in the cluster DB 54 will be described with reference to FIG.
The user cluster information is stored in association with reference answer information and affiliation user information for a user cluster No that can uniquely identify one user cluster UC.

Here, the standard answer information will be described. The user cluster UC is classified so that users who have similar preferences about the product image, that is, users who have obtained similar selection information on the question web page are in the same cluster. However, if classification is performed so that only users whose selection information matches are the same user cluster UC, the number of user clusters UC becomes enormous.
Therefore, in the present embodiment, a reference answer is set for each user cluster UC, and a user whose selection information is not significantly different from the reference answer, that is, a user whose selection information is a fixed distance from the reference answer is assigned to the same cluster. An example of classification is given below.

The user cluster UC1 having the user cluster No. UC1 includes the user U1 and the user U2. As described above, since the user U1 and the user U2 are users set at a certain distance from the reference answer, the answer information for the same product is similar.
Here, how to set the standard answer will be described.
For example, when the user U1 has answered the question web page for the products A to J, the standard answer for the products A to J is set in the user cluster UC1 based on the selection information obtained for the user U1. That is, for the products A, B, C, D, G, H, and I, the first product image x1 (product images a1, b1, c1, d1, g1, h1, i1) is set as the standard answer, and the products E, F , J, the second product image x2 (product images e2, f2, j2) is set as the standard answer.

  In this state, when the user U2 has answered the question web pages for the products A to E and the products J to L, based on the answer information, the items about the products K and L that cannot be set by the selection information of the user U1. A standard answer is set. That is, for the product K, the second product image k2 is the standard answer, and for the product L, the first product image 11 is the standard answer.

As described above, the reference answer may be set based on selection information of a plurality of users to which the user belongs. Of course, the reference answer may be set using only selection information of one user.
In addition, since the selection information is classified into the same user cluster UC for users whose distance is a fixed distance from the reference answer, the selection information for the product A of the user U2 may be the second product image a2.
Specific processing for setting the standard answer will be described later.

[4-3. Product cluster information]
Information on the product cluster SC stored in the cluster DB 54 will be described with reference to FIG.
The product cluster information is stored by associating a common evaluation axis and a product image selection tendency with respect to a product cluster No. that uniquely identifies one product cluster SC.

A common evaluation axis will be described. For example, the presentation candidate images prepared for the product A are the first product image a1 and the second product image a2 (see FIGS. 4 and 8), and the evaluation axis set for the two product images is “ “Appearance / Content Evaluation Axis” (see FIGS. 7 and 8).
When the presentation candidate images for two different products have the same evaluation axis, it is possible to infer the preference for the other product from the user's preference for one product. A product cluster SC is a collection of such products in the same group. Accordingly, products belonging to the same product cluster SC have the same evaluation axis (common evaluation axis).

  As the product image selection tendency, the user cluster UC that has selected the first product image x1 and the user cluster UC that has selected the second product image x2 are stored separately. Also stored is whether the first product image x1 is a product image for “appearance” or a product image for “content”.

  Specifically, the product cluster SC1 in which the product cluster number is SC1 is the user cluster UC (UC1 to UC3) that has selected the first product image x1 and the second product in the presentation candidate images of each product belonging to the product cluster SC1. The user cluster UC (UC4 to UC7) that has selected the image x2 is stored. This information means that for the products A, D, G, H,... Belonging to the product cluster SC1, which of the first product image x1 and the second product image x2 is selected depends on the user cluster UC. However, the same tendency is observed for each user cluster UC. That is, the user cluster UC1 that has selected the first product image x1 for the product A has a strong tendency to select the first product image x1 for the product D as well.

  When focusing on each user belonging to the user cluster UC1, since the user cluster UC1 includes a user whose distance is fixed from the reference answer as described above, a certain product is different from many people. Some users prefer product images.

  As can be understood from FIG. 10, not all products having a common evaluation axis belong to the same product cluster SC. Specifically, the product clusters SC1 and SC2 are different product clusters even though the common evaluation axis is the “appearance / content evaluation axis”.

In this regard, consider a general user X belonging to the user cluster UC2. The user X likes the product image a1 that can confirm the appearance of the product A. On the other hand, for the product B, the product image b2 whose contents can be confirmed is preferred. In such a case, the tendency of the favorite product image varies depending on the product. If such products A and B are included in the same product cluster, when a product image for the product B is presented based on the selection information of the product A, a product image different from the user's preference is presented. The possibility increases.
In view of such a problem, different product clusters SC are generated based on the selection tendency for each user cluster UC even if the common evaluation axis is the same.

  The product cluster SC may be set so as to allow a slight difference in the selection tendency of the user cluster UC. For example, the product clusters SC1 and SC2 in FIG. 10 have the same selection tendency except for the user clusters UC2 and UC5 among the user clusters UC1 to UC7. That is, the selection tendency of the user cluster UC of 5/7 (= 71.4%) is the same. Therefore, even if a single product cluster in which the product clusters SC1 and SC2 are combined is generated, a product image that is different from the product image preferred by the user is rarely presented, so it can be considered that there is no problem.

In addition, as described above, since the products are classified into the product clusters SC based on the user's selection tendency with respect to the product image, the products that are completely different in genre, that is, seemingly have no relationship, are the same. It belongs to the product cluster SC. Such product classification is difficult without using a classification method such as that of the present embodiment based on the user's selection tendency, and is based on a common point that a person cannot be aware of (that is, unconscious).

<5. Various processing examples>

[5-1. Response acquisition process]
As described above, in the present embodiment, information (selection information) that allows the user to select a product image is acquired, and clustering processing based on the information is performed.
Here, a process executed to acquire user selection information will be described with reference to FIG.

The following processes are executed by the product image selection device 7 using the function of the selection information management unit 7a, for example.
In step S101, the product image selection device 7 executes processing for specifying the target user.
For example, when the response acquisition process shown in FIG. 11 is performed for a new user newly registered to use the electronic commerce provided by the electronic commerce system 3, the target user is the new user. In addition, when each user periodically sends and answers the question web page information shown in FIG. 4, the target user is a user to whom the question web page is transmitted. Moreover, you may collect reply information for the point grant which can be utilized by electronic commerce as consideration. In that case, a user who is a collection target of answer information is a target user.

The product image selection apparatus 7 generates a question list in step S102. The question list is, for example, a question about a product that has not been presented to the target user. This can be realized by storing the presented question information in the DB for each user.
Further, it may be a question for a product newly handled by the electronic commerce system 3. In that case, whenever a new product handled by the electronic commerce system 3 increases, an answer acquisition process (FIG. 11) for the new product may be executed.

  The product image selection device 7 sends the question list in step S103. Specifically, information on the question web page (see FIG. 4) on which the question list generated in step S102 is posted is transmitted to the user terminal 4 of the target user. Note that the link information of the question web page may be transmitted to the user terminal 4. In addition, when link information is posted on a personal page for a target user who uses electronic commerce operated by the electronic commerce system 3 together with a word prompting an answer to the question list, and the target user selects the link information The information on the question web page may be transmitted to the user terminal 4 for the first time. That is, instead of executing the process of step S103 immediately after performing the process of step S102, the process of step S103 may be executed after a certain amount of time such as several hours or days.

  In step S104, the merchandise image selection device 7 performs a process of acquiring response information. Specifically, the user's answer information for the question is acquired. The answer information may be acquired every time the user selects an image for one question, or information on selected images that the user has performed for a plurality of questions may be acquired at a time.

In step S105, the product image selection device 7 performs processing for storing the acquired answer information as selection information in the selection information DB 53. Thereby, information (that is, selection information) indicating which image the target user has selected for each product is stored in the selection information DB 53, and is used in various processes described later.

[5-2. User clustering process]
User clustering processing for classifying users into user clusters UC will be described with reference to FIG.
Note that the series of processing illustrated in FIG. 12 is executed by the product image selection device 7 using the function of the user cluster management unit 7b, for example.
The timing for executing the user clustering process is, for example, when user selection information is acquired (that is, after the answer acquisition process shown in FIG. 11 is executed). Further, even if a user has already been classified into any user cluster UC, when a new selection information of the user is acquired, a series of processes shown in FIG. 12 is performed as a process for reclassifying the target user. You may do.

First, in step S201, the product image selection device 7 specifies a user who is a target of clustering this time as a target user.
In step S <b> 202, the product image selection device 7 executes a process of acquiring selection information about the target user from the selection information DB 53.

Subsequently, in order to determine which user cluster UC the target user belongs to, a process of checking each user cluster UC that has already been generated is executed.
Therefore, the product image selection device 7 determines whether or not there is an unprocessed user cluster UC in step S203.
If no user cluster UC has been generated or if all user cluster UCs have been confirmed but there is no user cluster UC suitable for the target user to belong to, the process proceeds to step S204. In step S204, the product image selection device 7 generates a new user cluster UC in which the target user selection information is set as the reference answer. Of course, the target user belongs to the new user cluster UC.
Information about the generated new user cluster UC is stored in the cluster DB 54.

  When there is an unprocessed user cluster UC (step S203: Yes), the product image selection device 7 performs a process of selecting one of the user clusters UC as a target user cluster in step S205, and in subsequent step S206. Get the reference answer for the target user cluster. For example, the standard answer is stored in the cluster DB 54 as a part of the user cluster information shown in FIG.

In step S207, the product image selection device 7 calculates the distance d. The distance d is a value indicating the degree of difference between the target user selection information and the reference answer of the target user cluster acquired in step S206.
Specifically, this will be described with reference to FIG.
13A and 13B show the reference answer of the target user cluster, the target user selection information, and the difference information for each product. For example, in the example shown in FIG. 13A, for the product A, both the standard answer and the selection information are the product image a1, and it can be said that the selection tendency of the target user and the target user cluster is similar.
For product B, the standard answer is product image b1, while the selection information is product image b2, so the difference information is “different”. The distance d is for counting the number of products whose difference information is “different”. In the range of products A to G, the distance d = 2.

As another method for calculating the distance d, a distance d is a numerical value obtained by dividing the number of items set to “different” by the total number of items (number of items set to “same” or “different”). It is possible.
For example, in the example illustrated in FIG. 13A, for products A to G, since 2/7 = 0.286, the distance d = 0.286. Further, the distance d may be set to a difference rate by further multiplying by 100. Specifically, distance d = difference rate = 28.6%.

Note that the reference answer of the target user cluster and the selection information of the target user are not necessarily targeted for the same product. This will be specifically described with reference to FIG. 13B.
In FIG. 13B, selection information about the product B is not obtained. That is, it may occur when a question about the product B is not provided on the question web page presented to the target user, or when a question is not answered.

In such a case, even if both pieces of information are compared, it is not suitable for determining whether or not the selection tendency of the target user is along the target user cluster. That is, the difference information is neither the same nor different, and is described as “−”.
In addition, the product S is in a state where there is no reference answer information. Also in this case, the difference information is described as “−”.
Further, the product T is in a state in which neither the standard answer nor the selection information is present.
The distance d between the standard answer and the selection information shown in FIG. 13B may be 1 (different number) or 1/7 = 0.143 when calculated based on the products A to G, S, and U. Alternatively, the difference rate may be 14.3%.

After calculating the distance d, the product image selection device 7 performs a process of determining whether or not the distance d is smaller than the distance threshold THd in step S208.
The distance threshold THd is, for example, a threshold for comparison with the distance d as the difference rate, and is a numerical value such as 20%. Of course, a numerical value such as “2” may be used as the threshold for the distance d obtained by counting the number of items (product number) for which the difference information is “different”.

That the distance d is smaller than the distance threshold THd indicates that the reference answer of the target user cluster and the selection information of the target user are similar, that is, the target user may be classified into the target user cluster. Judgment can be made.
In this case (step S208: Yes), the product image selection device 7 performs a process of adding the target user to the target user cluster in step S209.

Further, the distance d being equal to or greater than the distance threshold THd indicates that the reference answer of the target user cluster is not similar to the target user selection information, that is, the target user can be classified into another user cluster. In other words it can be desirable.
In this case (step S208: No), the product image selection device 7 executes the processes of steps S203 to S208 again. That is, for each of the other user clusters UC, the processes of steps S205 to S208 are performed, or a process of generating a new user cluster UC for classifying the target user (step S204).

[5-3. Product clustering process]
Product clustering processing for classifying products into product clusters SC will be described with reference to FIG.
The product clustering process illustrated in FIG. 14 is a process executed by the product image selection device 7 using the function of the product cluster management unit 7c, for example.
The execution timing of the product clustering process is, for example, the timing when selection information of a predetermined number of users is obtained for a product that is newly handled by the electronic commerce system 3. In other words, the product clustering process is not executed until user selection information for the target product is obtained.

In step S301, the product image selection device 7 specifies the target product N for which the answer acquisition process has been completed. That is, for example, a product for which a predetermined number or more of selection information is obtained is specified as the target product N.
In step S302, the product image selection device 7 sets an evaluation axis of the target product N.
As described above, all products belonging to the same product cluster SC have a common evaluation axis (see FIG. 10). Therefore, in order to classify products, it is necessary to grasp what the evaluation axis is.
The process of step S302 is, for example, a process of associating information on the evaluation axis input (or selected) by the operator with the target product N. In addition, by performing image processing on the product images n1 and n2 of the target product N, elements that are candidates for the evaluation axis may be automatically determined and presented to the operator. In this case, the evaluation axis selected by the operator is linked to the target product N.

Subsequently, the product image selection device 7 performs processing for searching for an appropriate product cluster SC to which the target product N belongs by comparing the characteristics of the product cluster SC that has already been generated with the characteristics of the target product N.
For this purpose, first, the product image selection device 7 determines in step S303 whether there is an unprocessed product cluster. If there is no unprocessed product cluster SC, an appropriate product cluster SC for classifying the target product N has not been found so far. Alternatively, a case where there is no unprocessed product cluster SC even when the generated product cluster SC does not exist yet.
In this case (step S303: No), the product image selection device 7 performs a process of generating a new product cluster SC to which the target product N belongs in step S304. The new product cluster SC is stored in the cluster DB 54.

On the other hand, if there is an unprocessed product cluster SC (step S303: Yes), the product image selection device 7 selects one target product cluster SCn in step S305, and acquires a common evaluation axis in step S306.
Then, in step S307, the product image selection device 7 determines whether or not the evaluation axis of the target product N and the common evaluation axis of the target product cluster SCn are the same. As described above, since products that do not have a common evaluation axis do not belong to the same product cluster SC, if the determination result is “No” in step S307, the target product cluster SCn is changed again. Steps S303 to S307 are executed.

On the other hand, when the determination result is “Yes” in step S307, the minimum condition for the target product N to belong to the target product cluster SCn is cleared. The product image of the product belonging to the target product cluster SCn and the product images n1 and n2 of the target product N are the same evaluation axis equipped images.
In this case, in step S308, the product image selection device 7 determines whether or not the evaluation tendency of the product images n1 and n2 of the target product N is the same as the evaluation tendency of the product images of the target product cluster SCn.

Here, the evaluation tendency of the product images n1 and n2 and the evaluation tendency of the product image of the target product cluster SCn will be described with reference to FIG.
For example, it is assumed that the target product N is a product V (wallet), and the product W (watermelon) is included in one of the products belonging to the target product cluster SCn.
At this time, the evaluation axes set for the product V are an “appearance / content evaluation axis” and a “text presence / absence evaluation axis” (see FIG. 15A). The evaluation axes set for the product W are also the “appearance / contents evaluation axis” and the “text presence / absence evaluation axis” (see FIG. 15B).
In this case, the product V and the product W may belong to the same product cluster SC.

Here, attention is paid to the product images v1 and v2 for the product V and the product images w1 and w2 for the product W.
The product image v1 is “appearance and no text”, and the product image v2 is “content and text”.
On the other hand, the product image w1 is “look and have text”, and the product image w2 is “content and no text”.

  In this state, when a certain user selects the product image v1 and the product image w1, the product image v1 and the product image w1 that can confirm the “appearance” of the product are the same user by focusing on the element “appearance”. Therefore, it can be said that it is preferable to classify the products V and W into the same product cluster SC.

  However, when focusing on the element “text”, the user selects a product image w1 from which text information about the product W can be obtained, while the product V includes a simple product that does not include text information. The image v1 is selected. From this, the product V and the product W can be regarded as products having different tendency of favorite product images, and it seems inappropriate to classify them into the same product cluster SC.

  In this way, in the case where the suitability of classification into the same product cluster SC is determined depending on what element is focused on, that is, the product V and the product W shown in FIG. 15 are the same product. It is determined that it is not appropriate to classify the cluster SC.

An example in which the evaluation tendency of the product images n1 and n2 and the evaluation tendency of the product image of the target product cluster SCn are matched will be described with reference to FIGS. 15A and 15C.
FIG. 15C shows product images y1 and y2 for the product Y (watermelon).
The product image y1 is “appearance and no text”, and the product image y2 is “content and text”. That is, the product image v1 for the product V shown in FIG. 15A and the product image y1 for the product Y shown in FIG. 15C are product images including the same elements. Moreover, the product image v2 for the product V shown in FIG. 15A and the product image y2 for the product Y shown in FIG. 15C are also product images including the same elements. That is, when a user selects the product image v1 and the product image y1, regardless of whether the element is “look” or “text”, the above-described wrinkles occur. Therefore, it can be determined that it is appropriate to classify into the same product cluster SC.
In the process of step S308 shown in FIG. 14, it is determined that such product V (FIG. 15A) and product Y (FIG. 15C) have the same evaluation tendency of product images.
That is, the evaluation axis set in the product image is acquired from the DB, and further, the common evaluation axis of the target product cluster SCn is acquired and it is determined whether or not the evaluation tendency is the same. It is determined whether or not there is a possibility of classifying products for the product image.

  Note that if the product images n1 and n2 have one evaluation axis or if the target product cluster SCn has one common evaluation axis, the above-described wrinkle cannot occur, so the process of step S308 is performed. It is not necessary to perform.

In step S309, the product image selection device 7 determines whether the selection tendency of each user cluster UC for the product image of each product belonging to the target product cluster SCn and the selection tendency of each user cluster UC for the product images n1, n2 are the same. Determine whether.
Specifically, description will be made using a product A belonging to the product cluster SC1 and an unclassified product E (see FIG. 16).
The common evaluation axis of the product cluster SC1 is the “appearance / contents evaluation axis”, and the evaluation axis set in the product images e1 and e2 of the product E is also the “appearance / contents evaluation axis”. ing. In addition, the first product image is a product image (a1, e1) in which “appearance” can be confirmed, and the second product image is a product image (a2, e2) in which “content” can be confirmed. Yes.

As for the selection tendency of the user clusters UC1 to UC7 for the product A belonging to the product cluster SC1, the user clusters UC that have selected the first product image a1 are UC1 to UC3.
The user clusters UC that have selected the second product image a2 are designated as UC4 to UC7.
On the other hand, the user cluster UC that has selected the first product image e1 for the unclassified product E is UC2, UC3, UC6, UC7, and the user cluster UC that has selected the second product image e2 is UC1, UC4 and UC5. That is, although the product E and the product A have the same evaluation axis, the selection tendency of the product image is different, and if it is classified into the same product cluster SC, it is not optimal from one selection information. There is a high possibility that one product image will be selected.
Therefore, as shown in FIG. 10, it is desirable that the product E is classified into a product cluster SC different from the product cluster SC1 to which the product A belongs.

The process of step S309 is a process for determining whether or not the selection tendency of each user cluster UC for the product image matches in this way.
If it is determined that the selection tendencies match, the product image selection device 7 performs a process of adding the target product N to the target product cluster SCn in step S310. That is, a process for updating information stored in the cluster DB 54 is performed.
On the other hand, if it is determined that the selection tendencies do not match, the product image selection device 7 executes each process of steps S303 to S309 again.

[5-4. Presentation image selection process]
A presentation image selection process for determining which product image to present as a presentation image when a product image of a certain product is presented to the target user will be described with reference to FIG.
This process is executed, for example, when recommending a product to the target user or selecting a product image to be presented as a search result.
The presentation image selection process is executed by the product image selection device 7 using the function of the presentation image selection unit 7d, for example.

In step S401, the product image selection device 7 performs processing for specifying the target user Un and the target product In. Note that the target user Un and the target product In may already be determined before executing the series of processes shown in FIG.
Subsequently, the product image selection device 7 acquires selection information of the target user Un in step S402. At this stage, without considering the user cluster SCn to which the target user Un still belongs, processing is performed based only on information directly answered by the user Un.

In step S403, the product image selection device 7 checks whether or not there is selection information of the target user Un for the target product In.
If the target user Un has made a response for the product images in1 and in2 of the target product In (step S403: Yes), the product image selection device 7 is based on the selection information of the target user Un in step S404. One of the product images in1 and in2 is selected as a presentation image.
Thereby, it is possible to reliably select a product image that the user likes.

On the other hand, if the target user Un has not made an answer for the product images in1 and in2 of the target product In (step S403: No), the product image selection device 7 is the same product cluster SCn as the target product In in step S405. The presence / absence of selection information of the target user Un for other products belonging to the item is confirmed.
If the response of the target user Un is directly obtained for another product having a selection tendency similar to that of the target product In, the product image of the target product In is selected based on the response result (selection information). Is possible. That is, if the determination result is “Yes” in step S405, the product image selection device 7 is based on selection information for the product image of another product of the target user Un (can be regarded as a substitute image) in step S406. The product image for the target product In is selected as a presentation image.
Thereby, possibility that the target user Un's favorite product image will be selected can be increased.

On the other hand, if the response of the target user Un is directly obtained for another product having a selection tendency similar to that of the target product In (step S405: No), the product image selection device 7 is the target user Un in step S407. The reference answer of the same user cluster UCn is acquired.
Note that selection information of other users belonging to the user cluster UCn may be acquired instead of acquiring the reference answer in step S407.

  Next, the optimal product image for the target user Un is selected using the acquired reference answer. That is, in step S408, the product image selection device 7 determines whether or not there is a reference answer for the target product In. The reference answer of the user cluster does not include information about all products like the product S shown in FIG. 13B. In step S408, it is determined whether there is a reference answer for the target product In.

When there is a standard answer for the target product In (step S408: Yes), it is possible to grasp an approximate trend as to what product image the user cluster UCn to which the target user Un belongs for the target product In.
Accordingly, in step S409, the product image selection device 7 performs a process of selecting the product image as a presentation image based on the reference answer for the target product In.
Thereby, possibility that the optimal product image suitable for object user Un's liking will be selected increases.

  If there is no standard answer for the target product In (step S408: No), the product image selection device 7 in step S410 is a standard for other products similar to the target product In (ie, belonging to the same product cluster SCn). Determine whether there is an answer.

When there is a standard answer for a similar other product (step S410: Yes), the other user whose image preference is similar to that of the target user Un is similar to the target product In. Based on the result of answering about, processing for selecting an appropriate product image for the target user Un is performed. That is, in step S411, the product image selection device 7 selects a product image as a presentation image based on the standard answer for other products.
Thereby, there is no answer information (selection information) made by the target user Un for the target product In, and further, reply information (selection information) made by the target user Un for other products whose image selection tendency is similar to the target product In. Even if there is not, it is possible to increase the possibility that an appropriate product image is selected for the target user Un based on answer information (selection information) about other products of other users.

  When there is no standard answer for similar other products (Step S410: No), the product image selection device 7 selects one of the product images in1 and in2 of the target product In as a presentation image in Step S412. I do. In this process, the product images in1 and in2 may be selected at random, or may be selected based on the reference answers of other user clusters Un that are close to the user cluster Un to which the target user Un belongs and the reference answers. Good.

  By performing each process of steps S401 to S412 shown in FIG. 17, one product image for the target product In is selected for the target user Un.

In addition, although the example using the selection tendency of the user cluster UCn to which the target user Un belongs has been described, the selection information of other users belonging to the user cluster UCn may be used without using the reference answer of the user cluster UCn. In that case, for example, it is conceivable that each process of steps S407 to S411 loops by the number of other users belonging to the user cluster UCn. That is, there is a possibility that appropriate information cannot be obtained from one other user. Moreover, you may grasp | ascertain the selection tendency of the user cluster UCn by taking the majority vote of several other users.

<6. Modification>

[6-1. Modification of product clustering process]
A modification of the product clustering process will be described with reference to FIG.
Note that description of processes similar to the processes illustrated in FIG. 14 is omitted as appropriate.

  The product image selection device 7 executes steps S301 to S306, sets the evaluation axis of the target product N, and performs a process of examining each product cluster SC suitable for the target product N to belong to. Go. Then, after acquiring the common evaluation axis of the target product cluster SCn, whether or not the first evaluation threshold or more has the same evaluation axis among the common evaluation axis of the target product cluster SCn and the evaluation axis of the target product N Determine.

  For example, it is assumed that the product images n1 and n2 of the target product N have 10 evaluation axes, and five common evaluation axes of the target product cluster SCn are set. At this time, the process branches depending on how many of the five common evaluation axes of the target product cluster SCn are set in the product images n1 and n2 of the target product N. For example, when “60%” is set as the first threshold TH1, the product images n1 and n2 only need to have three evaluation axes out of the five common evaluation axes.

If not provided (in the case of No at step S311), the product image selection device 7 returns to the process of step S303.
On the other hand, if it is provided (step S311: Yes), the product image selection device 7 determines that the product images n1 and n2 and the product image of the product belonging to the target product cluster SCn are the same evaluation axis provided image. Then, the process proceeds to step S312.

For example, consider an example in which the product images n1 and n2 of the target product N include all the five evaluation axes common to the target product cluster SCn in 10 evaluation axes.
In this case, in step S312, the product image selection device 7 determines whether the evaluation axis having the same image tendency among the five common evaluation axes is equal to or greater than the second threshold value TH2 (for example, 80%). To do. In other words, in this example, among the five common evaluation axes, four or more evaluation axes that are 80% branch depending on whether or not the tendency of the image is similar.
For example, in the product image v1 in FIG. 15A and the product image w1 in FIG. 15B, there are two identical evaluation axes (“appearance / contents evaluation axis” and “text presence / absence evaluation axis”). However, there is one evaluation axis. That is, 50% of the evaluation axes have the same image tendency.
On the other hand, in the product image v1 in FIG. 15A and the product image y1 in FIG. 15C, there are two identical evaluation axes, and there are also two evaluation axes with the same image tendency. That is, 100% of the evaluation axes have the same image tendency.

Returning to the description of FIG.
When the evaluation axis having the same image tendency among the five common evaluation axes is less than the second threshold TH2, the product image selection device 7 executes the process of step S303 again.
On the other hand, if it is equal to or greater than the second threshold TH2, the product image selection device 7 determines that the product images n1 and n2 of the target product N and the product images of the products belonging to the target product cluster SCn are similar evaluation images, and continues. The process proceeds to step S313.

In step S313, the product image selection device 7 selects selection information for both product images n1 and n2 of the target product N and products of the product that belong to the target product cluster SCn and belong to the same user cluster UC. It is determined whether the selection tendency of the users having the third threshold value TH3 or more among the users having the same is the same.
For example, there are 100 users belonging to a certain user cluster UC. Among them, selection information about the product images n1 and n2 of the target product N and selection information about the product image of any product belonging to the target product cluster SCn. Suppose that there are 50 users who have both stored. In this case, if the selection tendency of the third threshold TH3 (for example, 70% = 35) among the 50 users is the same, the target product N may be classified into the target product cluster SCn. Can be considered. It should be noted that the process of step S313 is performed for all user clusters UC, and the target product N is selected as the target product cluster when the user selection tendency of a certain number (third threshold TH3) or more is similar for each user cluster UC. It is desirable to determine that it is classified as SCn.

When the determination result of step S313 is “No”, the product image selection device 7 returns to the process of step S303.
On the other hand, when the determination result is “Yes”, the product image selection device 7 performs the process of step S310 and executes the process of adding the target product N to the target product cluster SCn.

[6-2. Modification Example 2 of Product Clustering Process]
In the second modification of the product clustering process, an example in which product clustering is performed for each user cluster UC will be described. Here, an example is given in which each product is classified into three product clusters SC. The three product clusters SC are a first product cluster SC1, a second product cluster SC2, and a third product cluster SC3.
This will be specifically described with reference to FIG.

The product image selection device 7 selects a user cluster UC to be processed (target user cluster UC) in step S501, and subsequently selects one product in step S502. The selected product is the target product N.
Next, the product image selection device 7 acquires selection information about the target product N of each user belonging to the target user cluster UC from the selection information DB 53 in step S503.
The product image selection device 7 classifies the target product N into one of the first product cluster SC1 to the third product cluster SC3 based on the acquired selection information.

  For example, in step S504, the product image selection device 7 determines whether or not a user having a fourth threshold value or more among users belonging to the target user cluster UC has selected the same product image for the target product N.

This process will be described with a specific example.
Assume that 100 users belong to the target user cluster UC. In addition, two product images n1 and n2 are prepared for the target product N.
When 90 users who are equal to or greater than a fourth threshold value (for example, 85%) out of 100 users have selected the product image n1, the determination process in step S504 is “Yes” determination. Become.
In this case, many users have selected the same image for the product image of the target product N. That is, there is a product image that matches the preferences of users belonging to the target user cluster UC.

  When “Yes” determination is made in step S504, the product image selection device 7 classifies the target product N into the first product cluster SC1 in step S505.

When it is determined as “No” in step S504, the product image selection device 7 determines in step S506 that a user who is equal to or more than the fifth threshold among users belonging to the target user cluster UC selects the same product image for the target product N. Determine whether or not.
The fifth threshold value is smaller than the fourth threshold value. For example, the fourth threshold value is “85%”, and the fifth threshold value is “70%”.
When “Yes” determination is made in step S506, the selection of the belonging user is somewhat varied with respect to the product images n1 and n2 for the target product N, but is somewhat biased to one product image. For example, when 75% of the users have selected the product image n1 and 25% have selected the product image n2, the determination in step S506 is “Yes”.

If the determination result in step S506 is “Yes”, the product image selection device 7 classifies the target product N into the second product cluster SC2 in step S507.
On the other hand, if the determination result in step S506 is “No”, the product image selection device 7 classifies the target product N into the third product cluster SC3 in step S508.
The target product N classified in the third product cluster SC3 is a product in which variation in the selection tendency of each user with respect to the product image prepared for the product is seen.

  After executing the processes of steps S505, S507, and S508, the product image selection device 7 returns to the process of step S502 and selects one next product. If there is no other product to be clustered, the series of processes shown in FIG.

By executing each process shown in FIG. 19, three product clusters SC1 to SC3 are generated for the target user cluster UC, and each product is classified.
In addition, in the first product cluster SC1, products for which the selection tendency of the users belonging to the target user cluster UC substantially match are classified.
In the second product cluster SC2, products whose belonging user's selection tendency matches to some extent are classified. The degree of coincidence of selection tendencies for the products classified into the second product cluster SC2 is lower than that of the products classified into the first product cluster SC1.
In the third product cluster SC3, products whose belonging user selection tendency varies are classified. For example, products for which the user who selected the product image n1 and the user who selected the product image n2 are both 50% are classified.

After the products are classified into the first product cluster SC1 to the third product cluster SC3, the product image selection device 7 executes a preference evaluation axis extraction process shown in FIG. In the preference evaluation axis extraction process, a common element of products belonging to each product cluster SC is extracted as a preference evaluation axis (that is, a preference evaluation axis).
Note that the example shown in FIG. 20 is a process of extracting a preference evaluation axis for the first product cluster SC1. The preference evaluation axes of the second product cluster SC1 and the third product cluster SC3 can also be extracted by performing processing equivalent to the processing shown in FIG.

In the preference evaluation axis extraction process, the product image selection device 7 selects one product (for example, product A) from the products belonging to the first product cluster SC1 in step S601.
In step S <b> 602, the product image selection device 7 acquires information on product images (for example, product images a <b> 1 and a <b> 2) of the selected product from the product DB 51.

In step S603, the product image selection device 7 sets an evaluation axis for the product image based on the difference for the acquired product images (for example, two product images). That is, as described above, the evaluation axis is an element that can be a basis for the selection when the user selects one image from two product images according to preference.
The evaluation axis set for the product image, that is, the evaluation axis set for the selected product may be one or plural. For example, the evaluation axes set for the product images a1 and a2 may be two evaluation axes of “appearance / content evaluation axis” and “text presence / absence evaluation axis”.

In step S604, the product image selection device 7 determines whether there is an unprocessed product that belongs to the first product cluster SC1, that is, a product that is not a processing target in step S603.
If there is an unprocessed product, the product image selection device 7 returns to step S601, selects one of the unprocessed products, and performs both processes of steps S602 and S603.

  On the other hand, if there is no unprocessed product, the product image selection device 7 proceeds to step S605 and performs a process of extracting an evaluation axis that should be a preference evaluation axis of the first product cluster SC1. Specifically, a process of extracting evaluation axes set for a plurality of products as preference evaluation axes of the first product cluster SC is performed.

Here, there are several conditions for the evaluation axis to be the preference evaluation axis.
For example, as products belonging to the first product cluster SC1, there are products A, B, C, D, and E.
An example in which product images a1, a2, b1, b2, c1, c2, d1, d2, e1, and e2 are linked as product images will be described.
The evaluation axis set for the product A based on the product images a1 and a2 is the “appearance / content evaluation axis”, and the evaluation axis set for the product D based on the product images d1 and d2 is also the “appearance / contents evaluation axis” (For example, FIG. 8). Furthermore, the product images a1 and d1 are product images for showing the “look” of the product, and the product images a2 and d2 are product images for showing the “contents” of the product.

For example, when most of the users belonging to the target user cluster UC select the product image a1 and the product image d1, an analysis that a product image that can confirm “appearance” is preferred as a common preference of users belonging to the target user cluster UC. Results are obtained. Such an evaluation axis is extracted as a preference evaluation axis of the first product cluster SC1.
However, for other products (for example, product E) belonging to the first product cluster SC1, most users belonging to the target user cluster UC are “contents” even though a product image that can confirm “appearance” is prepared. "Appearance / Content Evaluation Axis" is not extracted as a preference evaluation axis because it cannot be said that "Appearance / Content Evaluation Axis" can be a common selection factor for users belonging to the target user cluster UC. .

It should be noted that the number of products (for example, product E) for which the product image capable of confirming “content” is selected with respect to the number of products (for example, products A and D) for which the product image for which “appearance” can be confirmed is selected If the number is extremely small, it can be recognized that the product image that can confirm “appearance” is preferred as the preference tendency of the target user cluster UC, so “appearance / contents evaluation axis” is extracted as the preference evaluation axis. May be.
It should be noted that “appearance / content evaluation axis” need not be set for all of the products A, B, C, D, and E.
Moreover, you may comprise so that the evaluation axis extracted as a preference evaluation axis of 1st goods cluster SC1 may not be extracted as a preference evaluation axis of other goods cluster SC. That is, when a certain evaluation axis is extracted as a preference evaluation axis, it may be extracted only as a preference evaluation axis of one product cluster SC.

  In this way, at step S605, at least one evaluation axis is extracted as a preference evaluation axis. The preference evaluation axis is an evaluation axis representing the preference of each user belonging to the target user cluster UC. That is, when there are two product images having a preference evaluation axis, it is possible to guess which product image is likely to be selected by each user belonging to the target user cluster UC.

Moreover, since the preference evaluation axis is extracted beyond the product group frame set manually such as the product genre, each user belonging to the target user cluster UC has an essential and instinct. The tendency of preference for product images is easily reflected. By selecting a product image based on such a preference evaluation axis (for example, a modification of a presentation image selection process described later), it is possible to present a product image that is instinctively favored to the user.

[6-3. Modification of presented image selection process]
A modification of the presentation image selection process will be described. This selection process is a process of performing image selection using the product cluster clustered by the modification 2 of the product clustering process.
Specifically, this will be described with reference to FIG. Note that part of the processing is the same as the presentation image selection processing described with reference to FIG. 17, and therefore the same step numbers as those in FIG.

  The product image selection device 7 identifies the target user and the target product in step S401, acquires the target user selection information in step S402, and determines whether the target user has selection information for the target product in step S403. If there is target user selection information for the target product, the product image selection device 7 proceeds to step S404 and selects a product image based on the target user selection information. Thereby, the product image presented to the user is determined.

  If there is no selection information for the target product of the target user, the product image selection device 7 determines whether there is selection information of another user belonging to the same user cluster UC as the target user in step S420. . Since other users belonging to the same user cluster UC as the target user are users with similar selection tendencies for the product image, the other users are described as “similar users”.

If selection information of other users exists, the product image selection device 7 performs a branching process according to the product cluster SC in step S421. Specifically, when the product cluster SC to which the target product belongs is the first product cluster SC1 (in the case of “n = 1” in the figure), the product image selection device 7 selects the target product in step S422. A product image to be presented to the user is selected from the presentation candidate images based on the selection information of other users.
Each product belonging to the first product cluster SC1 is a product in which the selection tendency of the target user and the similar user is substantially the same. Therefore, it is preferable to select a product image from the presentation candidate images based on selection information of similar users.

If the product cluster SC to which the target product belongs is the second product cluster SC2 (in the case of “n = 2” in the figure), the product image selection device 7 responds to the second product cluster SC2 in step S423. Perform the selection process. Each product belonging to the second product cluster SC2 is a product in which the selection tendency of the target user and the similar user are matched to some extent. Accordingly, in step S423, a product image selected by many similar users is selected as a presentation image to the target user. However, in consideration of the fact that the products belonging to the second product cluster SC2 have some variation in selection tendency compared to the products belonging to the first product cluster SC1, the product images are randomly selected from the presentation candidate images at an appropriate frequency. May be selected.
For example, image selection based on the selection tendency of similar users is performed at a rate of 80%, and image selection contrary to the selection tendency of similar users is performed at a rate of 20%.

When the product cluster SC to which the target product belongs is the third product cluster SC3 (in the case of “n = 3” in the figure), the product image selection device 7 responds to the third product cluster SC3 in step S424. Perform the selection process.
For example, since each product belonging to the third product cluster SC3 is a product in which variation in the selection tendency of each user with respect to the product image is seen, a product image is selected at random from the presentation candidate images and used as a presentation image to the user. .

  After executing one of steps S422, S423, or S424, the product image selection device 7 ends the series of processing shown in FIG.

If it is determined in step S420 that there is no selection information of another user, the product image selection device 7 executes a process of setting an evaluation axis for the product image of the target product in step S425.
As described above, the evaluation axis is set based on the difference between the product images, and is an element that can be a motivation for the user to select an image. A plurality of evaluation axes may be set for the product image of the target product.

In step S426, the product image selection device 7 identifies an appropriate product cluster as the affiliation destination of the target product from the first product cluster SC1 to the third product cluster SC3.
Specifically, the evaluation axis for the target product set in step S425 and the preference evaluation axis for each product cluster SC extracted in FIG. 20 are compared to identify an appropriate product cluster.

For example, if the evaluation axis set for the target product includes the preference evaluation axis of the first product cluster SC1, and does not include the preference evaluation axis of the second product cluster SC2 and the third product cluster SC3, the target product It is determined that the appropriate product cluster SC to which is belongs is the first product cluster SC1.
In addition, the evaluation axis set for the target product includes one preference evaluation axis for the first product cluster SC1, three preference evaluation axes for the second product cluster SC2, and the third product cluster SC3. If the preference evaluation axis is not included, it is determined that the appropriate product cluster SC to which the target product belongs is the second product cluster SC2.

In step S427, the product image selection device 7 performs a branching process using the product cluster SC identified in step S426.
When the identified product cluster SC is the first product cluster SC1 (in the case of “n = 1” in the figure), the product image selection device 7 performs a selection process according to the first product cluster SC1 in step S428. Since each product belonging to the first product cluster SC1 is a product in which the selection tendency of the target user and the similar user is substantially the same, a product image is selected from the presentation candidate images based on the selection information of the similar user.

  When the product cluster SC to which the target product belongs is the second product cluster SC2 (in the case of “n = 2” in the figure), the product image selection device 7 responds to the second product cluster SC2 in step S429. Perform the selection process. Specifically, image selection based on the selection tendency of similar users is performed at a rate of 80%, random image selection is performed at a rate of 20%, and the selection trend of similar users is basically followed and lower. Perform random image selection etc. with frequency.

When the product cluster SC to which the target product belongs is the third product cluster SC3 (in the case of “n = 3” in the figure), the product image selection device 7 responds to the third product cluster SC3 in step S430. Perform the selection process.
For example, since each product belonging to the third product cluster SC3 is a product in which variation in the selection tendency of each user with respect to the product image is seen, a product image is selected at random from the presentation candidate images and used as a presentation image to the user. .

By executing the processes in steps S428, S429, and S430, a product image to be presented to the user is selected from the presentation candidate images. After executing the processes of steps S428, S429, and S430, the product image selection device 7 ends the series of processes shown in FIG.

[6-4. Modification Example 2 of Presentation Image Selection Process]
A modified example 2 of the presentation image selection process will be described.
In this modification, the relationship between product images is evaluated using an “evaluation value”, and the product image to be presented to the user is selected after considering the relationship between product images.

First, selection information of eight users U1 to U8 for the product images a1 and a2 for the product A and the product images b1 and b2 for the product B is shown in FIG.
As shown in FIG. 22, the four users U1 to U4 have selected the product image a1 for the product A and the product image b2 for the product B.
For the user U5, the product image a1 is selected for the product A, and the product image b1 is selected for the product B.
In addition, two users U6 and U7 select the product image a2 for the product A and select the product image b1 for the product B.
For the user U8, the product a1 is selected for the product A, but there is no selection information about the product B.

Here, the evaluation values of the product image a1 and the product image b2 will be described.
According to FIG. 22, there are six users U1 to U5 and U8 who have selected the product image a1. Among the six people, there are five users U1 to U5 who have selection information about the product B.
Of the users (users U1 to U5) who have selected the product image a1 and have selection information about the product B, four (users U1 to U4) have selected the product image b2. . At this time, the numerical value indicating the similarity of the selection tendency of the product image a1 and the product image b2 is calculated as 4/5 × 100 = 80, for example. That is, 80% of users select the same product image.

Next, consider the product image b2.
There are four users who have selected the product image b2, and among them, there are four users who have selected the product image a1 for the product A. That is, the numerical value indicating the similarity of the selection tendency of both product images is 4/4 × 100 = 100. That is, all the users who have selected the product image b2 have selected the product image a1.

The evaluation value representing the similarity between the product images a1 and b2 may be an average value (that is, 90) of the two numerical values (100 and 80 in the example) calculated above, or the lower one (that is, 80). ).
In this way, an evaluation value representing the similarity between product images is calculated. It can be considered that the higher the evaluation value, the more similar the selection tendency of the two product images. In other words, it is estimated that a user who likes one product image also likes the other product image with high probability.

In the process of evaluating the relationship between product images using an evaluation value, an image having a high evaluation value is determined as a similar tendency image.
For example, “70” is set as the first evaluation threshold Te1, and “90” is set as the second evaluation threshold Te2.

Since product images having an evaluation value equal to or greater than the first evaluation threshold Te1 (for example, 70) tend to be somewhat similar in selection tendency, there are some same evaluation axes that cause the user to select an image. Then it can be estimated. For example, there are cases where the product image is selected as a monotone image as a factor, and cases where the product image is selected as a colorful image.
Such product images, that is, product images having an evaluation value equal to or higher than the first evaluation threshold 70, are recognized as “same evaluation axis equipped images”.

  Further, it can be seen that the product images having the evaluation axis equal to or higher than the second evaluation value Te2 (for example, 90) have a very strong tendency that the user who selects one product image selects the other product image. It is strongly inferred that such product images are product images that have some common points or are similar product images. For this reason, such a product image “similar tendency image” is recognized.

  The recognition of the same evaluation axis provided image and the similar tendency image may be performed for each user cluster UC. In other words, a certain product image a1 and another product image b2 may be similar images in the user cluster UC1, but may not be similar in selection tendency in other user clusters UC2. For such a phenomenon, an evaluation value is calculated for each user cluster UC. As a result, a high value (for example, 95) is calculated for the user cluster UC1, whereas a low value (for example, 40) is calculated for the user cluster UC2. Can happen if you are.

Next, processing for selecting a product image to be presented to the user based on the evaluation value between product images will be described with reference to FIGS.
Note that part of the processing is the same as the presentation image selection processing described with reference to FIG. 17, and therefore the same step numbers as those in FIG.

  The product image selection device 7 identifies the target user and the target product in step S401, acquires the target user selection information in step S402, and determines whether the target user has selection information for the target product in step S403. If there is target user selection information for the target product, the product image selection device 7 proceeds to step S404 and selects a product image based on the target user selection information. Thereby, the product image presented to the user is determined.

If there is no selection information for the target product of the target user, the product image selection device 7 determines whether there is selection information of another user belonging to the same user cluster UC as the target user in step S420. .
When selection information of another user exists, the product image selection device 7 selects a product image to be presented to the user from the presentation candidate images based on the selection information of the other user with respect to the target product in step S422.
That is, the product image to be presented to the target user is determined based on the selection information of other users who have a similar selection tendency as the target user.

On the other hand, if there is no other user's selection information about the target product, the product image selection device 7 determines in step S440 whether there is a product image that has a relationship between any of the presentation candidate images and a similar tendency image. The similarity tendency image is a product image for at least another product different from the target product.
When there is a product image that has a relationship between a similar tendency image and one of the presentation candidate images, it is possible to select the presentation candidate image based on the selection tendency with respect to the similar tendency image.

In other words, if “Yes” determination is made in step S440, the product image selection device 7 determines in step S441 whether or not there is an image selected by the target user among product images having a relationship of similar tendency images. If there is an image selected by the target user, the product image selection device 7 selects a similar tendency image similar to the selected image from the presentation candidate images in step S442. Thereby, the product image presented to the user is determined.
That is, it is presumed that there are two product images with similar selection tendencies, and a target user who has a track record of selecting one by preference is highly likely to prefer the other product image.

On the other hand, when there is no user selection information for the product image that is a similar tendency image (step S441: No determination), the processing proceeds to FIG. 24A. That is, in step S443, the product image selection device 7 determines whether there is an image selected by another user belonging to the same user cluster UC as the target user among product images having a similar tendency image relationship.
In the case of “Yes” determination in step S443, that is, when selection information of another user whose selection tendency is similar to the target user can be used, the product image selection device 7 is similar to the image selected by the other user in step S444. A similar tendency image is selected from the presentation candidate images. Thereby, the product image presented to the user is determined.

  On the other hand, in the case of “No” determination in step S443, the product image selection device 7 proceeds to the processing to step S445 in FIG. In step S440, the process proceeds to step S445 even when there is no product image that has a similar tendency image relationship to any of the presentation candidate images.

  In step S445, the product image selection device 7 determines whether or not there is a product image having the same evaluation axis provided image as any of the presentation candidate images. That is, when there is no product image whose similarity with the presentation candidate image is strongly estimated, it is determined whether or not there is a product image (that is, an image with the same evaluation axis) that has a certain degree of similarity with the presentation candidate image.

If it is determined in step S445 that the same evaluation axis provided image exists, the product image selection device 7 selects the target user from among the product images having the same evaluation axis provided image in any one of the presentation candidate images in step S446. It is determined whether there is information.
When there is selection information of the target user for the same evaluation axis provided image, the product image selection device 7 selects an image having the relationship between the image having the selection information and the same evaluation axis provided image from the presentation candidate images in step S447.

On the other hand, if it is determined in step S446 that there is no user selection information, the product image selection device 7 advances the processing to step S448 in FIG. 24B. That is, in step S448, the product image selection device 7 determines whether there is an image selected by another user belonging to the same user cluster UC as the target user among the product images having the same evaluation axis equipped image relationship. .
When there is a product image selected by another user having a similar selection tendency to the target user among images having the same evaluation axis, it is possible to select a product image to be presented to the target user from the presentation candidate images.
That is, in step S449, the product image selection device 7 selects an image with the same evaluation axis corresponding to the image selected by the other user from the presentation candidate images. Thereby, the product image to be presented to the user is determined.

On the other hand, if there is no other user's selection information for the same evaluation axis provided image in step S448, the product image selection device 7 proceeds to the process of step S412 in FIG. In addition, also when it determines with "No" by step S445, it progresses to the process of step S412.
In step S412, the product image selection device 7 selects one product image of the target product. The selection method may be random or other methods, for example.

By executing any one of the processing of step S404, the processing of step S442, the processing of step S444, the processing of step S447, the processing of step S449, or the processing of step S412, the product image to be presented to the user is included in the presentation candidate images. Selected from.
After the product image to be presented to the user is determined, the product image selection device 7 ends the series of presentation image selection processes shown in FIGS.

In addition, according to this example, as shown in FIG. 22, it is digitized how much the user who selected the product image a1 has selected the product image b2, and based on this numerical value (namely, evaluation value). The similarity between the product image a1 and the product image b2 is determined. That is, it is not determined what evaluation axis each product image has.
Accordingly, a common evaluation axis may be set by performing image analysis on the product images having high evaluation values and determined to be similar tendency images, in the above-described specific example, the product images a1 and b2. The common evaluation axis setting process is performed by, for example, the evaluation axis setting unit 7e.

  For example, when the product images a1 and b2 are both monochrome images, it can be determined that the monochrome image is an important element (evaluation axis) in image selection for some users. In such a case, the element is set as a common evaluation axis. The common evaluation axis can be used when selecting a product image to be presented to the user. For example, instead of performing the process shown in FIG. 23, it is determined whether there is a product image having a common evaluation axis in the presentation candidate images, that is, whether there is a black and white image. In some cases, the product image is selected as an image to be presented to the user.

In addition, although the product images a1 and b2 are black and white images, if the product images a2 and b1 are also black and white images, it is obvious that the product images a1 and b2 are not selected because they are black and white images. It is. Therefore, it is desirable not to select such an element as a common evaluation axis.

[6-4. Other variations]
In addition, although the product clustering process for classifying the product into the product cluster SC has been described above, it may be considered that a specific product is not classified into the product cluster SC. For example, when almost all of the product images a1 and a2 for the specific product A select the product image a1, the product A may not be clustered. That is, there is no problem if the product image a1 is presented to any user. Thereby, the processing burden for executing product clustering is reduced.

  The reference answer of the user cluster UC has been described with reference to an example in which the user cluster UC is set based on selection information of one user when the user cluster UC is generated. As an alternative, the majority of the selection information of users belonging to the same user cluster UC may be taken and the result reflected in the standard answer. This prevents minority selection information in the user cluster UC from becoming a reference answer.

In the example described above, an example has been given in which the selection information of each user belonging to the user cluster UC may be slightly different. That is, if the distance d from the reference answer is smaller than the distance threshold THd, they are classified into the same user cluster UC. Therefore, there is a possibility that there is a user who has selection information different from the standard answer for each product.
Therefore, a specific product may be configured not to be classified into the same user cluster UC unless the user has the same selection information as the reference answer.
For example, it is conceivable that there are many opportunities to recommend users to popular products. When there is a user who has selection information different from the standard answer for such a specific product, if a product image is selected based on the standard answer, a non-optimal product image is selected.
In order to prevent such a situation, a specific product may not be classified into the same user cluster UC unless the distance d from the reference answer is 0.

  The first threshold TH1 described above may be a first predetermined ratio, the second threshold TH2 may be a second predetermined ratio, and the third threshold TH3 may be a third predetermined ratio.

In the user clustering process described above, an example has been described in which a new user cluster can be generated when the distance d from the reference answer is equal to or greater than the distance threshold THd.
As a modification of the user clustering process, two user clusters are prepared for classifying each user belonging to the user cluster when the distribution of the selection information of each user belonging to a certain user cluster exceeds a certain value. Each user may be classified into the two user clusters.

In the case of obtaining answer information from the user, an example of transmitting dedicated web page data on which a question is posted has been described, but other examples are also conceivable. For example, when a user takes some action, one question may be answered. Specifically, when the user purchases a product, a web page for confirming the order contents is generated and presented to the user. One question may be placed on the order content confirmation page, and the user may be made to answer the question to obtain answer information. In addition, instead of pressing the order confirmation button, the web page data may be configured to select one of the favorite product images.
In addition, there is a case where an image is selected in order to determine whether or not various operations are performed by a human hand. Instead, one of the product images of products for which response information is to be acquired is selected. You may do it. For example, by instructing the user such as “Select one of your favorite product images and enter the number of products in the image”, let the user select the desired product image. At the same time, it may be determined whether or not the operation being performed is actually performed by a human hand.

<7. Summary>

As described in the above-described various examples, the product image selection device 7 includes the presentation image information about a plurality of images presented to the user as product images (product images a1 and a2) for a product (product A). A selection information management unit 7a that manages user information of users who have received presentation of the plurality of images in association with user selection information for the plurality of images, and a user whose users are classified based on the selection information A user cluster management unit 7b that manages the cluster UC, a product cluster management unit 7c that manages a product cluster SC in which products are classified based on user responses, and a user cluster UC and a product cluster SC that are presented to the user. A presentation image selection unit 7d that selects a product image for presentation as a presentation image.
A plurality of product images for a certain product are presented, and an image selected by the user is likely to be an image that the user feels attractive about the product. According to this configuration, such image selection information is managed in association with user information.
Thus, a database for automatically selecting a product image is configured when there is a product to be appealed to a certain user.
Further, by classifying users using the user cluster UC, even when the user's preference is unknown, the optimum product image is selected based on the preferences of other users belonging to the same user cluster UC. Is possible.
Then, by classifying the products using the product cluster SC, even if the product is unknown to the target user, the product image can be selected based on the target user's preference for other products belonging to the same product cluster SC. It becomes possible.
In other words, by performing clustering for users and products, it is possible to obtain a product image that is optimal for the user through simple processing (processing with low calculation costs) without using sophisticated processing with high calculation costs such as machine learning. It becomes possible to select.

Further, as described in the example of the presentation image selection process (particularly, steps S408 and S409), the presentation image selection unit 7d selects a presentation image to be presented to the target user Un from the presentation candidate images in1 and in2 for a certain product In. In the processing to be performed, when a selection image of another user belonging to the same user cluster UCn as the target user Un is included in the presentation candidate image, the presentation candidate image may be selected as a presentation image to the target user Un.
Thereby, even if there is no selection information of the target user Un for the presentation candidate image, appropriate image selection is performed based on the selection information of other users belonging to the same user cluster UCn.
Therefore, an appropriate image can be presented to the target user Un. Further, since there is no need for the selection information of the target user Un itself, the storage area for storing the selection information can be reduced.

Furthermore, as described in the example of the presentation image selection process (particularly, steps S410 and S411), the presentation image selection unit 7d selects a presentation image to be presented to the target user Un from the presentation candidate images in1 and in2 for a certain product In. In the processing to be performed, if a substitute image for another product belonging to the same product cluster SCn as a certain product In is included in a selection image of another user belonging to the same user cluster UCn as the target user Un, Among them, an image having the same tendency as that of the substitute image may be selected as a presentation image to the target user Un.
As a result, when there is no selection information of the target user Un for the presentation candidate image and there is no selection information of other users belonging to the same user cluster UCn, the other product that belongs to the same product cluster SCn as the product In When there is selection information of another user, a product image for a product is selected based on the tendency of the product image for another product selected by another user.
That is, even if the selection information about the product image of a certain product In is poor, if a certain product In is clustered, the possibility of selecting an appropriate presentation candidate image can be increased. Thereby, even if there is little information amount about selection information, selection of an appropriate product image can be performed. That is, it is possible to effectively use the storage area and reduce the necessary storage capacity.

In addition, as described in the example of the answer acquisition process, the selection information management unit 7a of the product image selection device 7 sets one set of a plurality of sets of images (see FIGS. 4A and 4B). You may manage the combination of the selection image which the user selected according to the presentation that one image is selected for every.
For example, the user is classified according to the selection of the product images from the product images a1 and a2 for the product A, the product images b1 and b2 for the product B, and the product images c1 and c2 for the product C.
Since the clustering is performed more finely than the clustering of users based on the selection information selected by the user for the product image for a single product, the presented image is based on the selection information of other users belonging to the same user cluster UC. When is selected, it is possible to increase the possibility that an image that matches the preference of the target user is selected.

Furthermore, as described in the example of the user clustering process, the user cluster management unit 7b of the product image selection device 7 may generate the user cluster UC based on the combination.
The user cluster UC is not generated in advance, but the user cluster UC is generated according to the combination of selected images actually selected by the user. Therefore, for example, a user cluster UC for a combination of selected images that no one chooses is not generated in vain.
That is, it is possible to save waste of resources such as the storage area and processing capability of the information processing apparatus (product image selection apparatus 7).

Furthermore, as described in the example of the user cluster information shown in FIG. 9, the user cluster management unit 7b may generate the user cluster UC using the combination of the selected images as a reference answer.
In one user cluster UC, the users belonging to the user cluster UC may not have exactly the same preference, but may allow a certain degree of preference difference. In such a case, it is difficult to set an ideal model of the preference of the user cluster UC. In particular, when generating user clusters UC according to a combination of actual selection images of the user, as compared with the case of classifying users after generating user clusters UC in advance, what kind of preference model is used. It is difficult to determine whether to be a UC preference model. However, according to this configuration, when a new user cluster UC is generated, a combination of selected images actually selected by the user is set as a reference answer (that is, an ideal model) of the user cluster UC, which is easy. .
That is, since it is not necessary to calculate the median value or the average value of the preference for each user cluster UC, it is possible to reduce the processing load on the information processing apparatus (product image selection apparatus 7).

In addition, as described in the example of the user clustering process, the user cluster management unit 7b has a combination of selected images in which the difference (distance d) is set to be equal to or greater than each reference answer of the generated user cluster UC. When a new event occurs, a user cluster UC using the new combination as a reference answer may be generated.
When a combination of user-selected images has a certain distance from any of the existing user clusters UC, a new user cluster UC is generated, so that a user cluster UC having a similar tendency is generated. Is prevented.
In addition, the combination of user-selected images that causes new user cluster generation is used as a reference answer for the new user cluster UC, thereby eliminating the annoyance of setting the preference model as described above. It is possible to reduce the processing load of the information processing apparatus (product image selection apparatus 7).

Then, as described in the example of the functional configuration of the product image selection device 7, an evaluation axis setting unit 7 e that sets an evaluation axis that represents an image tendency for each image may be provided.
For example, when there are an image showing only a product and an image showing a model with the product as an image of a product, the “model presence / absence evaluation axis” is set to the image. By performing user clustering based on such an evaluation axis and user selection information, it is possible to appropriately grasp what preference model the user cluster UC is configured by. .
In addition, when a product image is presented to the user, it is possible to determine which image matches (or is suitable for) the user's preference based on the evaluation axis set in the image.

In addition, as described in the modification example of the product clustering process (see FIG. 18), the product cluster management unit 7c uses two product images having different target products for images having the same evaluation axis as the first threshold value or more. Determines that the two product images are images having the same evaluation axis, and among the same evaluation axes included in the two product images that are the same evaluation axis provided image, When the two tendencies are the same, the two merchandise images are determined to be similar evaluation images, and the two merchandise images are used to obtain selection information about the two merchandise images that are users belonging to the same user cluster. Among users who are presented with images, when a user with a third threshold or higher selects both images that are similar evaluation images, the target products of the two images are classified into the same product cluster. May
For example, it is assumed that two evaluation axes are assigned to each of the product image a1 for the product A and the product image b1 for the product B. Among the two evaluation axes, there is the same evaluation axis, and there is also an evaluation axis assigned to only one image. According to this configuration, it is determined whether a common evaluation axis is assigned to the first threshold value or more (for example, half or more) among the two evaluation axes.
Thus, it is first determined whether or not to determine whether or not the product images a1 and b1 belong to the same product cluster in the first place. That is, if there is no common (identical) evaluation axis, it is not necessary to determine whether or not the product clusters are the same. Therefore, by determining whether or not to determine whether or not to belong to the same product cluster by a simple process, it is possible to avoid performing a useless process, thereby reducing the processing load.
Moreover, when the same evaluation axis is given to the product images a1 and b1 by the first threshold value or more, that is, when the product images a1 and b1 are determined to have the same evaluation axis, the respective evaluations set as the same evaluation axis It is determined whether the axis trends are the same. Specifically, it is confirmed whether the tendency of the evaluation axis equal to or higher than the second threshold (for example, 80%) is the same. When the tendency of the evaluation axis equal to or greater than the second threshold is the same, the product images a1 and b1 are determined as similar evaluation images.
Here, since it was found that the product images a1 and b1 are similar, it is conceivable that the product image b1 is presented when an image about the product B is presented to a user who likes the product image a1. However, according to this configuration, a further determination process is performed to determine whether or not to present the product image b.
Specifically, it is determined how many users among the users belonging to the same user cluster prefer the product images a1 and b1 with respect to the product images a1 and b1 that are set as similar evaluation images. For example, among users belonging to the same user cluster, a plurality of images including a product image a1 for the product A are presented to obtain user selection information, and a plurality of images including the product image b1 for the product B are displayed. Assuming that there are 30 presented users, if the third threshold value (for example, 80%) or more, that is, 24 or more users select both product images a1 and b1, products A and B are assigned to the same product cluster. Classify into:
If this is simply explained, when considering whether or not the photograph shows the model, the product image a1 includes the model, and the product image b1 also includes the model. Users who prefer a1 tend to think that they should also like the product image b1. However, when the product A is clothing and the product B is food, the product image about the clothing is preferred to the product image worn by the model, and the product image about the food is held by the model. There may be a user who does not like the product image in the state.
Therefore, by determining whether or not users belonging to the same user cluster prefer the product images a1 and b1 that are similar evaluation images, it is easy to perform optimal image selection.

As described in the above examples, the product image selection device 7 receives presentation image information about a plurality of types of images presented to the user as a product image for one product and the presentation of the plurality of types of images. A selection information management unit 7a for managing the user information of the user and the selection result information based on the user's preference for the plurality of types of images, and the selection result information of the plurality of users for a plurality of products. A user cluster management unit 7b for managing a user cluster into which each user is classified based on;
The product cluster management unit 7c that manages the product cluster in which each product is classified based on the user's answer, and the user based on the relationship between each cluster managed by the user cluster management unit and the product cluster management unit for a certain product A presentation image selection unit 7d that selects a product image to be presented to any target user belonging to any of the clusters as a presentation image from a plurality of presentation candidate images.
In other words, presentation image information about a plurality of types of images presented to the user as a product image for one product, user information of a user who has received the presentation of the plurality of types of images, and a user for the plurality of types of images A selection information management unit 7a that manages the selection result information based on the user's preference, and a user cluster in which the information of each user is classified based on the selection result information of a plurality of users for a plurality of products. The user cluster management unit 7b to manage, the product cluster management unit 7c to manage the product cluster in which each product is classified based on the user's answer, and a certain product managed by the user cluster management unit and the product cluster management unit A product image to be presented to an arbitrary target user belonging to the user cluster based on the relationship between the clusters is displayed as a plurality of presentation images. It includes a presentation image selector 7d for selecting the candidate image.
Thus, an appropriate product image is presented to the user based on the user's selection tendency (preference for preference) and the selection tendency of other users similar to the user.
For example, according to the embodiment described in the modification example 2 of the product clustering process, the users are clustered based on the selection tendency in response to the user selecting the product image. Further, product clustering is also performed based on user selection information. That is, since both the user cluster UC and the product cluster SC are clustered based on the user selection information, it is not necessary to use a complicated method and an excessive processing load is not imposed.

As described in the example of the presented image selection process (particularly steps S408 and S409) and the modified example of the presented image selection process (particularly step S420, S422, step S443, S444, step S448, and S449), the presented image selection unit 7d. In the process of selecting a presentation image to be presented to the target user from a plurality of presentation candidate images for a certain product, the target user has no history of selecting any of the plurality of presentation candidate images for the certain product. You may select goods based on the selection tendency in the user cluster to which the target user belongs.
In other words, the presentation image selection unit 7d has a history in which the target user has selected any type of image of the certain product in the process of selecting a presentation image to be presented to the target user from a plurality of presentation candidate images for the certain product. If there is not, the presentation candidate image may be selected as the presentation image to the target user based on the selection tendency in the user cluster to which the target user belongs with respect to the image of the certain product.
Thereby, even if there is no user selection information, an appropriate product image is selected based on the selection information of other users who have similar selection tendencies to the user and presented to the user.

As described in the modified example of the presented image selection process, the image processing apparatus further includes an image evaluation axis setting unit (evaluation axis setting unit 7e) for setting an evaluation axis representing an image tendency for each product image, and a selection tendency in the user cluster. May be determined using an evaluation axis.
In other words, it further includes an image evaluation axis setting unit (evaluation axis setting unit 7e) for setting an evaluation axis representing an image tendency for each product image, and the evaluation axis is selected in the user cluster in the presentation image selection unit 7d. It may be used to determine trends.
Accordingly, an evaluation axis that can be a factor for selecting each product image based on the selection tendency of each user with respect to the product image is appropriately set, and a product image to be presented to the user using the evaluation axis can be easily set. Can be determined.

As described in the modification of the presented image selection process, the product cluster management unit 7c has the evaluation value based on the same evaluation axis determined by the image evaluation axis setting unit (evaluation axis setting unit 7e) for each product image. Images that are equal to or greater than one threshold (first evaluation threshold Te1 in the above example) are determined to be images having the same evaluation axis, and the same evaluation axis of the product image that is the image having the same evaluation axis. Among them, images having an evaluation axis that is equal to or higher than a second threshold value (second evaluation threshold value Te2 in the above example) higher than the first threshold value are determined to be images having similar tendencies, and the presented image selection unit 7d May select a presentation image based on a similar tendency.
In other words, the product cluster management unit 7c is the same for each product image for images whose evaluation values based on the same evaluation axis determined by the image evaluation axis setting unit (evaluation axis setting unit 7e) are equal to or greater than the first threshold. It is determined as an image having an evaluation axis, and has an evaluation axis whose evaluation value is equal to or higher than a second threshold value that is higher than the first threshold value among the same evaluation axes of product images that are images having the same evaluation axis. The images may be determined as images having a similar tendency, and the presented image selection unit 7d may select the presented image as a presented image to the target user based on the similar tendency.
Thereby, a similar tendency image is appropriately determined for product images having similar user selection tendencies. Further, by appropriately using the similar tendency image, a product image suitable for presentation to the user is selected.

As described in the example of the user cluster information illustrated in FIG. 9, the user cluster management unit 7b may generate a user cluster based on a combination of selection information for a plurality of products.
In other words, the user cluster management unit 7b may generate a user cluster based on a combination of a plurality of selection results.
Accordingly, users having similar selection tendencies, in other words, preference tendencies are classified into the same user cluster.

As described in the other modifications, the user cluster management unit 7b may generate a new user cluster when the variance of the selection information of each user belonging to one user cluster is a certain value or more. .
In other words, the user cluster management unit and / or the product cluster management unit may generate a new cluster when it is determined that the relationship between the individual data in the generated cluster is more than a predetermined value.
Similarly, regarding the product cluster, when information on each product belonging to one product cluster is dispersed more than a certain amount, each corresponding product may be reclassified into a new product cluster.
Thereby, the unification of the tendency of the product cluster is achieved, and the product image to be presented to the user can be appropriately selected.

<8. Program and Storage Medium>

The program according to the embodiment is a program that causes an arithmetic processing device (CPU or the like) of the product image selection device 7 to execute various processes.

The program according to the embodiment includes presentation image information about a plurality of images presented to a user as a product image for a certain product, user information of a user who has received the presentation of the plurality of images, and user's information for the plurality of images. The arithmetic processing unit is caused to execute a selection information management function for managing the selection information in association with each other.
Further, the arithmetic processing unit is caused to execute a user cluster management function for managing user clusters in which users are classified based on selection information.
Further, the arithmetic processing unit is caused to execute a product cluster management function for managing a product cluster in which products are classified based on a user's answer.
In addition, the processing device is caused to execute a presentation image selection function for selecting a product image to be presented to the user as a presentation image based on the user cluster and the product cluster.

In addition, the program according to the embodiment includes presentation image information about a plurality of types of images presented to a user as a product image for one product, user information about a user who has received the presentation of the plurality of types of images, The arithmetic processing unit is caused to execute a selection information management function for managing information associated with selection results based on user preferences for types of images.
Further, the arithmetic processing unit is caused to execute a user cluster management function for managing a user cluster in which information of each user is classified based on information on the selection results of a plurality of users for a plurality of products.
Furthermore, the processing unit is caused to execute a product cluster management function for managing a product cluster in which each product is classified based on a user's answer.
Furthermore, for a certain product, a product image to be presented to any target user belonging to the user cluster based on the relationship between the clusters managed by the user cluster management function and the product cluster management function is used as a presentation image. A calculation processing apparatus is caused to execute a presentation image selection function for selecting from a plurality of presentation candidate images.

  That is, these programs are programs that cause the arithmetic processing unit of the information processing apparatus to execute the processes shown in FIGS. 11, 12, 14, 17, 18, 19, 23, and 24.

  With such a program, it is possible to realize one or a plurality of information processing apparatuses as the product image selection apparatus 7 described above.

Such a program can be stored in advance in an HDD as a storage medium built in a device such as a computer device, a ROM in a microcomputer having a CPU, or the like. Alternatively, it can be stored (stored) temporarily or permanently in a removable storage medium such as a semiconductor memory, memory card, optical disk, magneto-optical disk, or magnetic disk. Such a removable storage medium can be provided as so-called package software.
Further, such a program can be installed from a removable storage medium to a personal computer or the like, or can be downloaded from a download site via a network such as a LAN or the Internet.

  1 network system 2 communication network 3 electronic commerce system 4 user terminal 5 store terminal 6 shopping server 7 product image selection device 7a selection information management unit 7b user cluster management unit 7c product cluster management unit 7d Presentation image selection unit, 7e evaluation axis setting unit, 50 user DB, 51 product DB, 52 store DB, 53 selection information DB, 54 cluster DB

Claims (8)

Based on presentation image information on a plurality of types of images presented to the user as a product image for one product, user information on a user who has received the presentation of the plurality of types of images, and user preferences for the plurality of types of images A selection information management unit that links and manages selection result information;
A user cluster management unit that manages a user cluster in which information of each user is classified based on information of the selection results of a plurality of users for a plurality of products;
A product cluster management unit that manages a product cluster in which each product is classified based on a user's answer;
For a certain product, a plurality of product images as presentation images to be presented to any target user belonging to the user cluster based on the relationship between the clusters managed by the user cluster management unit and the product cluster management unit An information processing apparatus comprising: a presentation image selection unit that selects from candidate images. The presented image selection unit
In the process of selecting a presentation image to be presented to the target user from a plurality of presentation candidate images for a certain product, if there is no history that the target user has selected any type of image of the certain product, the image of the certain product The information processing apparatus according to claim 1, wherein a presentation candidate image is selected as a presentation image to the target user based on a selection tendency in a user cluster to which the target user belongs. An image evaluation axis setting unit for setting an evaluation axis representing the tendency of the image for each product image;
The information processing apparatus according to claim 2, wherein the evaluation axis is used for determination of a selection tendency in the user cluster in a presentation image selection unit. The product cluster management unit
For each product image, an image having an evaluation value based on the same evaluation axis determined by the image evaluation axis setting unit is determined to be an image having the same evaluation axis for images having a first threshold value or more,
Of the same evaluation axes of product images that are images having the same evaluation axis, images having an evaluation axis whose evaluation value is equal to or higher than a second threshold higher than the first threshold tend to be similar to each other. Judge as an image,
The information processing apparatus according to claim 3, wherein the presentation image selection unit selects the presentation image as a presentation image to the target user based on the similarity tendency. The information processing apparatus according to claim 1, wherein the user cluster management unit generates the user cluster based on a combination of a plurality of the selection results. 6. The user cluster management unit and / or the product cluster management unit generates a new cluster when it is determined that the relationship between individual data in the generated cluster is more than a predetermined value. An information processing apparatus according to any one of the above. Based on presentation image information on a plurality of types of images presented to the user as a product image for one product, user information on a user who has received the presentation of the plurality of types of images, and user preferences for the plurality of types of images A selection information management step for linking and managing information of selection results;
A user cluster management step of managing a user cluster in which information of each user is classified based on information of the selection results of a plurality of users for a plurality of products;
A product cluster management step for managing a product cluster in which each product is classified based on a user's answer;
For a certain product, a plurality of product images to be presented to any target user belonging to the user cluster based on the relationship between the clusters managed in the user cluster management step and the product cluster management step as a presentation image An information processing method in which an information processing apparatus executes a presentation image selection step of selecting from candidate images. Based on presentation image information on a plurality of types of images presented to the user as a product image for one product, user information on a user who has received the presentation of the plurality of types of images, and user preferences for the plurality of types of images Selection information management function that links and manages selection result information,
A user cluster management function for managing a user cluster in which information of each user is classified based on information of the selection results of a plurality of users for a plurality of products;
A product cluster management function for managing a product cluster in which each product is classified based on a user's answer;
For a certain product, a plurality of product images to be presented to any target user belonging to the user cluster based on the relationship between the clusters managed by the user cluster management function and the product cluster management function as a presentation image A program for causing a computer to execute a presentation image selection function for selecting from candidate images.

Images