JP7475590B2 - Information processing device and program - Google Patents

Description

An embodiment of the present invention relates to an information processing device and a program.

Recently, it has become difficult to continue selling products in a store while sales staff are serving customers (physical store sales). The proportion of physical store sales in total sales is decreasing, and the proportion of sales using electronic commerce (EC sales) via the Internet is increasing.

Furthermore, the use of SNS (Social Networking Service) to attract customers to EC sites is attracting attention as a sales model in which specific SNS users (image contributors), such as sales staff at physical stores, post photos (images) of products on Instagram (registered trademark), for example, and then direct SNS users from those posts to the EC site, thereby leading to product sales.

This sales model that utilizes SNS has shifted from so-called "passive sales," in which SNS users wait for an opportunity to visit the poster's posting page, to active sales activities in which advertisements are sent to SNS users individually to guide them to the posting page. There are various types of advertisements, such as display advertisements and email advertisements. All of these are delivered to a narrow range of target users, but the success rates, such as click rates and conversion rates, cannot be said to be very high.

One of the reasons for this is thought to be that the precision with which the target users for ad delivery can be narrowed down is not very high.

The purpose is to improve the accuracy of targeting users for ad delivery and to create personas that include interests along with profiles in order to achieve effective ad delivery.

In this embodiment, an information processing device is connected via an Internet line to an SNS server that provides SNS services to a plurality of SNS users, and includes a storage unit that stores a program, a processor that executes the program, and a communication unit that communicates with the SNS server via the Internet line. The processor executes the program to request and download from the SNS server via the communication unit the profiles of the SNS users, the images posted by the SNS users, the hashtags attached to the images by the SNS users, and the comments posted by the SNS users, a means for performing language analysis processing on the hashtags and the comments to extract nouns and to identify the classification of the interests of the SNS users based on the extracted nouns, and a means for creating persona data for the SNS users from the profile and the classification of the interests.

FIG. 1 is a diagram showing an example of a screen configuration of an SNS service. FIG. 2 is a configuration diagram of an entire system including an information processing apparatus according to this embodiment. FIG. 3 is a diagram showing the configuration of the information processing apparatus shown in FIG. FIG. 4 is a diagram illustrating an example of a persona. FIG. 5 is a diagram illustrating an example of a classification system of objects of interest identified by the processor of FIG. FIG. 6 is a flowchart showing the procedure of the persona creation process implemented by the processor of FIG. FIG. 7 is a conceptual diagram of the persona creation process of FIG. FIG. 8 is a flowchart showing a typical procedure for creating a persona of an SNS user who has a strong relationship with user A, which is realized by the processor of FIG. FIG. 9 is a conceptual diagram of the exemplary persona creation process of FIG. FIG. 10 is a flowchart showing the procedure of the purchase candidate extraction process (human axis) realized by the processor of FIG. FIG. 11 is a conceptual diagram of the extraction of potential purchasers in FIG. FIG. 12 is a flowchart showing the procedure of the candidate purchaser extraction process (product axis 1) realized by the processor of FIG. FIG. 13 is a conceptual diagram of the extraction of potential purchasers shown in FIG. FIG. 14 is a flowchart showing the procedure of the candidate purchaser extraction process (product axis 2) realized by the processor of FIG. FIG. 15 is a conceptual diagram of the extraction of potential purchasers in FIG.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
This embodiment relates to a process of extracting product purchase candidates by utilizing a posting and viewing service of images (still images (photos), videos) such as "Facebook (registered trademark)", "Twitter (registered trademark)", and "Instagram (registered trademark)" in a social networking service (hereinafter referred to as "SNS") as a community-type service that promotes and supports connections between people. As shown in FIG. 1 as an example of a screen of an image posting and viewing service, an SNS user who receives the SNS service is provided with a function of registering a profile and posting photos and videos (hereinafter referred to as images) related to hobbies, products for sale, etc., together with his/her own comments. On the other hand, an SNS user who views posted images, etc. is provided with a function of reacting to the user who posted the image (image poster) and the post and performing some kind of action. For example, various functions are provided, such as a function of following the image poster to grasp the update status of the image poster, a function of sending a comment on the post, a function of expressing sympathy such as "like" for the post, a function of retweeting and sharing to recommend and spread the post, etc. to his/her followers, etc., and a function of saving (downloading) the posted image.

Note that SNS users who respond to an image poster or its post by executing actions such as the follow function, comment posting function, function of expressing sympathy such as "like," spread function such as retweeting or sharing, and save function, and perform some action, are considered to be SNS users who find the target image poster or its post attractive and show a strong interest. Here, SNS users who execute at least one of the functions of the follow function, comment posting function, function of expressing sympathy such as "like," spread function such as retweeting or sharing, and save function are said to be SNS users who show an interest in the image poster or its post.

As shown in FIG. 2, an SNS server 2 that provides SNS services is connected to an information processing device 1 according to this embodiment via a public communication line network (Internet line) 8. In addition, terminals (user terminals) 3-7, such as PCs, smartphones, PHS, PDAs, tablet terminals, etc. owned by SNS users such as image contributors and viewers, are connected to the SNS server 2 via the Internet line 8.

For example, SNS users acting as image posters can post images, comments, etc. to the SNS server 2 via user terminal 3. On the other hand, SNS users acting as viewers can send various reactions to posts via user terminals 4-7, such as comments on posts, sympathy for posts such as "likes," antipathy for posts such as "no," spreading posts by retweeting or sharing, saving (downloading) posted images, and following the poster.

As shown in FIG. 3, in the information processing device 1, a RAM 12, a ROM 13, a storage unit 14, an input controller 15, a video controller 16, and a communication unit 17 are connected to a processor 11 via a data/control bus 10. An input device 18 such as a keyboard or a mouse is connected to the input controller 15. A display 19 such as an LCD (Liquid Crystal Display) is connected to the video controller 16.

The processor 11 is composed of a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 11 executes various programs loaded from the storage unit 14 or ROM 13 to the RAM 12 to execute the persona creation process, typical persona creation process, and three types of purchase candidate extraction process described below. The RAM 12 functions as the main memory, work area, etc. of the processor 11. The ROM 13 or the storage unit 14 stores the BIOS (Basic Input Output System), operating system program (OS), the persona creation program according to this embodiment, the typical persona creation program, the three types of purchase candidate extraction program, other programs for realizing various functions such as preprocessing, and various data required for these processes, which are executed by the processor 11.

Next, the persona creation process will be described. As shown in Figs. 4(a) to 4(c), a persona is defined as a profile of an SNS user, and is set here from various items related to product purchasing behavior, such as age, gender, place of residence, interests, and trends that indicate recent interests. As shown in Fig. 5, interests, which are the items most related to product purchasing behavior, are more specifically composed of major categories such as fashion and automobiles, as well as lower-level medium and small categories.

The various means related to the persona creation process are realized by the processor 11 executing a persona creation program loaded from the memory unit 14 or the ROM 13 to the RAM 12.

As shown in Figures 6 and 7, first in step S11, a request is made to the SNS server 2 via the communication unit 17 for the profiles registered by each SNS user, the images posted by each SNS user, the hashtags added to the posted images, and the comments sent by each SNS user on their own posts, and these are downloaded. The profile data separated by the separation means 302 is passed to the persona creation means 307, the posted image data is passed to the object extraction means 303, and the hashtags and comments are passed to the language analysis means 305.

In step S12, the language analysis means 305 performs language analysis processing (text mining) on the downloaded hashtags and comments to extract nouns, and the interest identification means 306 identifies the major categories of the SNS user's interests based on the extracted nouns. This identification processing may involve building a correspondence table of many nouns and multiple major categories in advance, consulting the correspondence table individually for each noun, and identifying the major category of interests showing the highest frequency as the major category of the SNS user's interests, or it may involve inputting multiple nouns using a trained neural network and outputting the major categories of interests.

In step S13, the object extraction means 303 uses the identified major category of interests to extract partial images from the posted image that contain objects that correspond to the major category of interests, for example, jackets that correspond to fashion. This extraction process typically uses a convolutional neural network. Since the scope of the image analysis process described below can be limited to partial images that contain objects that correspond to the major category of interests, it is expected that the accuracy of image analysis will improve.

The image analysis means 304 is divided into a plurality of classification means 304-1, 304-2, ... 304-n that are optimized for each major classification of objects of interest. The plurality of classification means 304-1, 304-2, ... 304-n are typically configured with a convolutional neural network system. As is well known, a convolutional neural network extracts features from a two-dimensional image by convolving the image with a filter. A convolutional neural network is a multi-layer network that repeats convolution and pooling. The coefficients (weights) of the filters that are effective for classification and that constitute the convolutional layer in the convolutional neural network are trained using a large amount of data such as a large amount of training images. The coefficients (weights) are obtained by learning to acquire invariance against various deformations by repeating convolution with a filter and pooling that summarizes responses in a certain area using a large amount of data. The classification performance of the convolutional neural network depends on the filters that constitute the convolutional layer. Filters are prepared for each major classification of objects of interest.

In step S14, one of the multiple classification means 304-1, 304-2, ... 304-n that matches the identified major classification of the object of interest performs image analysis processing on the partial image to estimate the lower-level medium and small classifications of the object of interest. If the object of interest (major classification) is "fashion", a classification means constructed to match the major classification "fashion" is selected, and the selected classifier analyzes the partial image and outputs "cool" as the medium classification and "high-end brand" as the minor classification (see FIG. 4(a)).

When there are multiple images posted by the SNS user, steps S12, S13, and S14 are repeated for each posted image. Since a major category, medium category, and small category of interests are identified for each posted image, there may be multiple major categories, medium categories, and small categories. In this case, the major category, medium category, and small category with the highest frequency are identified as the major category, medium category, and small category of representative interests for the SNS user.

In step S15, the trend analysis means 308 identifies trends that indicate topics in which the SNS user has recently shown interest, based on the results of language analysis of the hashtags and comments attached to the latest or a predetermined number of recent posted images among multiple posted images posted by the SNS user. For example, language analysis is used to extract multiple nouns from the hashtags attached to the latest or a predetermined number of recent posted images among multiple posted images posted by the SNS user, and the user's own comments, and the frequency of appearance of each noun is tallied, and the noun that appears most frequently is identified as the trend. For example, when the frequency of the noun "Korean fashion" is highest, "Korean fashion" is identified as the trend for that user (see FIG. 4(a)).

In step S16, the persona creation means 307 creates persona data by associating specific items in the profile, such as age, gender, place of residence, interests (major categories, medium categories, minor categories), and trends, with the SNS user's account. The persona data is stored in the storage unit 14 in step S17.

Steps S11-S17 are repeated through step S18, and persona data is created and stored for all SNS users who receive the SNS service or for SNS users who are included in a specific predefined group.

Next, with reference to Figures 8 and 9, a process for creating a typical persona of an SNS user (D) who has a strong interest in a particular user A and also in the products posted by user A will be described. Note that it is assumed here that user A is a sales clerk working at an apparel shop, an SNS user, and an image contributor who posts images of various products sold at the apparel shop.

In step S21, the account of user A or user B who shows a strong interest in the post is requested from the SNS server 2 via the communication unit 17 and downloaded. Examples of user A or user B who shows a strong interest in the post include followers of user A, users who have expressed sympathy for user A's post by "liking" or the like, users who have spread user A's post by retweeting or sharing, users who have saved (downloaded) images posted by user A, users who have posted comments on user A's post, etc.

In step S22, the hashtag attached to the image posted by user A is requested from the SNS server 2 via the communication unit 17 and downloaded.

In step S23, the name (product name) of the product posted by user A is identified based on the results of language analysis processing of the hashtag attached to the image posted by user A. Note that the name (product name) of the product posted by user A may be input by the operator via the input device 18.

In step S24, the account of user C, who posted an image with a hashtag including the name of the product (product name) posted by user A, is requested from the SNS server 2 via the communication unit 17, and downloaded. Since user C posted an image of the same product as the product posted by user A, it is presumed that user C is showing a strong interest in the product posted by user A.

In step S25, multiple users D are extracted that overlap with user B who shows a strong interest in user A or his/her posts and user C who shows a strong interest in the products posted by user A. It can be said that user D is a user who shows a strong interest in both user A or his/her posts and the products posted by user A.

In step S26, the persona data of user D is read from the storage unit 14, and in step S27, the persona of user D is integrated to create a typical persona (PA) of user D. The items constituting the persona are tallied by age group, sex, place of residence, interest (major category), interest (medium category), interest (minor category), and trend, and the specific contents of the age group, sex, place of residence, interest (major category), interest (medium category), interest (minor category), and trend that show the highest frequency are identified. A typical persona can be said to be a group of elements that characterize the personality shared by multiple users D who show a strong interest in user A or his posts, and who also show a strong interest in the products posted by user A.

In step S28, the typical persona data is associated with user A's account and stored in the memory unit 14.

Next, the process of extracting potential purchasers will be described. It is assumed that the process by which a person uses an SNS service to purchase a product is that the person becomes very interested in User A or his/her posts, and while viewing User A's posts many times, becomes very interested in the product that User A has posted, and then moves to an EC site via User A's post and purchases the product being sold on the EC site.

In this embodiment, two methods are provided to extract users who are likely to show a strong interest in user A or his/her posts in the future as purchase candidates: a method for extracting purchase candidates mainly based on the persona of a specific user A (called the person-based method), and a method for extracting purchase candidates based on products posted by user A (called the product-based method). Two types of the latter product-based method are also provided. In this embodiment, the person-based method and the two types of product-based methods may be equipped and applied selectively, or the purchase candidates extracted by all or any two of the person-based method and the two types of product-based methods may be further narrowed down by the logical product of the two methods.

10 and 11 show the processing procedure of the mandrel method.
In step S31, the persona data of the SNS user created in advance from the storage unit 14, that is, the persona data of the specific user A and each of the other users, is read out, and in step S32, a user E having the same or similar persona as the user A is extracted from the other users as a user with a similar personality to the user A. The user E has the same age, sex, place of residence, interests (major categories), interests (medium categories), interests (minor categories), and trends as the user A. Alternatively, the user E refers to a case where some of the items of age, sex, place of residence, interests (major categories), interests (medium categories), interests (minor categories), and trends are different from the user A, and the other items are the same. Here, the different items are items other than the items considered to be important for showing a strong interest in a person, and for example, the place of residence and interests (minor categories) are set. In other words, users whose place of residence and interests (minor categories) are different from those of user A, but whose other categories such as age, gender, interests (major categories), interests (medium categories), and trends are the same as those of user A, are extracted.

In step S33, the account of user F who shows a strong interest in user E or his/her posts is requested from the SNS server 2 via the communication unit 17 and downloaded. In other words, it is assumed that user F who shows a strong interest in user E who has a similar personality to specific user A is a user who is likely to show a strong interest in specific user A as well. As described above, users F who show a strong interest in user E or his/her posts include followers of user E, users who have expressed sympathy for user E's posts by "liking" or the like, users who have spread user E's posts by retweeting or sharing, users who have saved (downloaded) images posted by user E, users who have posted comments on user E's posts, etc.

In step S34, user G is extracted from user F by excluding user D, who has already shown a strong interest in user A or his posts and has also shown a strong interest in the products posted by user A. User G is assumed to be a user who has shown a strong interest in user E, whose personality is similar to user A, but who has not had any contact with user A up until now.

In step S35, data of a typical persona (PA) of user D, who shows a strong interest in user A or the posts made by user A and also in the products posted by user A, created in advance is read from the storage unit 14, and user H, who has a persona identical or similar to the typical persona (PA), is extracted from user G. Here, "similar" refers to a case where some items among age, sex, place of residence, interests (major categories), interests (medium categories), interests (minor categories), and trends are different and other items are the same, as described above, and items other than those considered important for showing a strong interest in a person are set in advance as different items, for example, place of residence and interests (minor categories) are set.

It is assumed that User H is a user who is highly likely to show a strong interest in User A and purchase products posted by User A in the future. In other words, User H is highly likely to become a follower of User A in the future, or to respond to User A's posts by expressing sympathy, such as by clicking "like," and is highly likely to move to an EC site via User A's post and ultimately purchase a product.

12 and 13 show the processing procedure of the first product axis method.
First, in step S41, a hashtag attached to an image posted by a specific user A is requested from the SNS server 2 via the communication unit 17 and downloaded. In step S42, the names (product names) of the products a, b, and c posted by the user A are identified based on the results of language analysis processing of the hashtag. The names (product names) of the products posted by the user A may be input by the operator via the input device 18. In step S43, the names of similar products d and e that are similar to the products a, b, and c posted by the user A are input by the operator via the input device 18. Typically, the names of products competing with the products a, b, and c posted by the user A are input as the similar products d and e.

In step S44, the account of user J who posted an image with a hashtag including the names of products a, b, and c posted by user A is requested from the SNS server 2 via the communication unit 17, and downloaded. Similarly, in step S45, the account of user K who posted an image with a hashtag including the names of similar products d and e similar to the product posted by user A is requested from the SNS server 2 via the communication unit 17, and downloaded. Since users J and K have posted products a, b, and c or similar products d and e posted by user A, it is assumed that they have already purchased products a, b, c, d, and e, or have a strong interest in them even if they have not purchased them, and that the products handled by the apparel shop where user A works are relatively consistent with the user's preferences.

In step S46, user D, who has already shown a strong interest in user A or his posts, is excluded from users J and K, and it is assumed that the users are those who have shown a strong interest in the products posted by user A or similar products, but who have not had any contact with user A until now.

In step S47, data of a typical persona (PA) of user D, who shows a strong interest in user A or the user's post and in the posted product or a similar product, created in advance is read from the storage unit 14, and user M, who has a persona identical to or similar to the typical persona (PA), is extracted from user L. Here, "similar" is as described above.

It is assumed that User L is highly likely to become interested in products posted by specific User A in the future, and is a potential purchaser who is likely to show an interest in products posted by specific User A, move to the EC site via User A's post, and ultimately purchase the product.

14 and 15 show the processing procedure of the second product axis method.
First, in step S51, the hashtag attached to the image posted by a specific user A is requested and downloaded from the SNS server 2 via the communication unit 17. In step S52, the names (product names) of the products a, b, and c posted by the user A are identified based on the results of language analysis processing of the hashtag. Of course, the names (product names) of the products posted by the user A may be input by the operator via the input device 18.

In step S53, the account of user J, who posted the image with hashtags including the names of products a, b, and c posted by user A, is requested from the SNS server 2 via the communication unit 17, and is downloaded.

In step S54, user N is extracted from user J, excluding user A downloaded in step S21 or user B who shows a strong interest in the posts. User N is assumed to be a user who has a strong interest in the products posted by user A, but has not had any contact with user A up until now.

In step S55, data on a typical persona (PA) of user A, which was created in advance, or user D, who shows a strong interest in the user A's posts, is read from the storage unit 14, and user P, who has a persona identical to or similar to the typical persona (PA), is extracted from user N. Here, "similar" is as described above.

It is assumed that User P is highly likely to become interested in products posted by specific User A in the future, and is a potential buyer who is likely to show interest in products posted by specific User A, move to the EC site via User A's post, and ultimately purchase the product.

According to this embodiment, it is possible to extract potential buyers who are highly likely to show a strong interest in a specific user A in the future, move to an EC site via user A's posts, and ultimately purchase a product, as well as potential buyers who are highly likely to show a strong interest in products posted by user A in the future, move to an EC site via user A's posts, and ultimately purchase a product. Therefore, it is possible to realize effective advertisement delivery to these potential buyers.

Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are within the scope of the invention and its equivalents as set forth in the claims, as well as the scope and gist of the invention.

1... Information processing device, 2... SNS server, 3-7... User terminal.

Claims (5)

An information processing device connected via an Internet line to an SNS server that provides an SNS service to a plurality of SNS users,
A storage unit that stores a program;
A processor for executing the program;
a communication unit that communicates with the SNS server via the Internet line,
When the processor executes the program,
a means for requesting the SNS server via the communication unit, and downloading the profiles of the SNS users, the images posted by the SNS users, the hashtags attached to the images by the SNS users, and the comments posted by the SNS users;
A means for extracting nouns by performing a language analysis process on the hashtags and the comments, and identifying a category of an interest of each of the SNS users based on the extracted nouns;
An information processing device that realizes a means for creating persona data for each of the SNS users from the profile and the classification of the interests. When the processor executes the program,
and further realizing an image analysis means for performing an image analysis process on the posted image that matches the classification of the subject of interest and identifying a lower classification of the subject of interest;
The information processing apparatus according to claim 1 , wherein the persona data is generated from the profile and the categories of interests, as well as sub-categories of the interests. 3. The information processing apparatus according to claim 2 , wherein the image analysis means comprises a convolutional neural network. When the processor executes the program,
A means for identifying trends of the SNS users based on a language analysis process result of the hashtag and the comments attached to one or a predetermined number of the latest posted images among the posted images is further realized,
The information processing device according to claim 1 , wherein the persona data is created from the profile, the classification of interests, and trends of the SNS user. A computer connected via an Internet line to an SNS server that provides an SNS service to a plurality of SNS users,
a means for requesting the SNS server and downloading the profiles of the SNS users, the images posted by the SNS users, the hashtags attached to the images by the SNS users, and the comments posted by the SNS users;
A means for extracting nouns by performing a language analysis process on the hashtags and the comments, and identifying a category of an interest of each of the SNS users based on the extracted nouns;
A program for functioning as a means for creating persona data for each SNS user from the profile and the classification of interests .