JP7416904B1 - Information processing device, information processing method, and information processing program - Google Patents

Abstract

The present invention determines content to be recommended to one or more users forming a family.
[Solution] An information processing device creates a relationship graph indicating social relationships among the plurality of users based on user characteristics of the plurality of users, and uses the relationship graph to Among them, family users, including registered users who are registered with the content distribution service and cohabiting users who live with the registered user, are predicted, and multiple contents provided to the registered user by the content distribution service. Each of them identifies which of the family users has viewed the content, and based on the identification results, recommended content to be recommended to the family user is determined using a learning model.
[Selection diagram] Figure 2

Description

The present invention relates to an information processing device, an information processing method, and an information processing program, and particularly relates to a technique for recommending content to a user.

In recent years, content distribution services that provide video content on the Internet have become widespread. In content distribution services, it is important to recommend media content, including video content such as movies and anime, and advertising content, in order to further improve the use of the service. Patent Document 1 discloses a system for displaying recommendations of selected content based on knowledge of a user's interests and activities (e.g., the user's search queries, search results, viewing history, purchase history, user's physical activity). A method is disclosed.

Special Publication No. 2022-527229

Generally, a content distribution service provides content to users who have registered with the service (registered users). By using the technology disclosed in Patent Document 1, it becomes possible to recommend content suitable for the registered user to the registered user.

On the other hand, if a family consisting of multiple users, including the registered user, shares a digital device (for example, a smart TV or tablet terminal) for viewing content, the viewed content may be shared with users other than the registered user. It may be. In other words, it is basically unknown which users in the family viewed content viewed on a shared digital device. Conventionally, in such cases, there has been no mechanism for determining content to be recommended to the entire family or to individual users, taking into consideration the characteristics of users other than the registered user in the family.

The present invention has been made in view of the above problems, and an object of the present invention is to provide a mechanism for determining content to be recommended to one or more users who constitute a family.

In order to solve the above problems, one aspect of the information processing apparatus according to the present invention provides a graph creation means for creating a relationship graph showing social relationships among a plurality of users based on user characteristics of the plurality of users. and a prediction unit that uses the relationship graph to predict family users among the plurality of users, including registered users who are registered in a content distribution service and cohabiting users who live together with the registered users; an identification means for identifying by which of the family users each of the plurality of contents provided to the registered user by the content distribution service; and a learning model based on the result of the identification. , determining means for determining recommended content to be recommended to the family user.

In order to solve the above problems, one aspect of the information processing method according to the present invention includes a graph creation step of creating a relationship graph showing social relationships among the plurality of users based on the user characteristics of the plurality of users. and a prediction step of predicting family users, including registered users who are registered in a content distribution service and cohabiting users who live together with the registered users, among the plurality of users, using the relationship graph; an identification step of identifying by which of the family users each of the plurality of contents provided to the registered user by the content distribution service; and a learning model based on the result of the identification. , a determining step of determining recommended content to be recommended to the family user.

In order to solve the above problems, one aspect of an information processing program according to the present invention is an information processing program for causing a computer to perform information processing, the program causing the computer to perform information processing based on user characteristics of a plurality of users. Graph creation processing that creates a relationship graph showing social relationships among the plurality of users based on the information, and a graph creation process that uses the relationship graph to determine which of the plurality of users is registered with the content distribution service. Prediction processing for predicting family users including the user and cohabiting users who live together with the registered user, and each of a plurality of contents provided to the registered user by the content distribution service, and a determination process for determining recommended content to be recommended to the family user using a learning model based on the result of the identification. belongs to.

According to the present invention, a mechanism for determining content to be recommended to one or more users forming a family is provided.
The objects, aspects, and effects of the present invention described above as well as the objects, aspects, and effects of the present invention not described above can be understood by those skilled in the art by referring to the accompanying drawings and claims. It can be understood from the form of

FIG. 1 shows a configuration example of an information processing system according to an embodiment. FIG. 2 is a schematic diagram illustrating a use case assumed in the embodiment. FIG. 3 shows an example of the functional configuration of the information processing device according to the embodiment. FIG. 4 shows a flowchart of social graph creation processing. FIG. 5A is a diagram for explaining explicit links in a social graph. FIG. 5B is a diagram for explaining implicit links in a social graph. FIG. 6A is a diagram for explaining inference processing of relationships between links. FIG. 6B shows a flowchart of an example process for grouping pairs into clusters. FIG. 7A shows a conceptual diagram of an intimacy score based on the intimacy of a relationship for a user pair. FIG. 7B shows the schematic architecture of the familiarity score prediction model. FIG. 7C shows a conceptual diagram of a social graph. FIG. 8A is a diagram for explaining the learning procedure of the viewing score prediction model. FIG. 8B is a diagram for explaining the learning procedure of the viewing score prediction model. FIG. 8C is a diagram for explaining the learning procedure of the viewing score prediction model. FIG. 9A is a diagram for explaining a procedure for determining recommended content. FIG. 9B is a diagram for explaining the procedure for determining recommended content. FIG. 10 shows a flowchart of overall processing executed by the information processing apparatus according to the embodiment. FIG. 11 shows an example of the hardware configuration of an information processing device and a communication device according to the embodiment.

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the accompanying drawings. Among the constituent elements disclosed below, those having the same functions are given the same reference numerals, and their explanations will be omitted. The embodiments disclosed below are examples of means for realizing the present invention, and should be modified or changed as appropriate depending on the configuration of the device to which the present invention is applied and various conditions. The present invention is not limited to this embodiment. Furthermore, not all combinations of features described in this embodiment are essential to the solution of the present invention.

[Configuration example of information processing system]
FIG. 1 shows a configuration example of an information processing system according to this embodiment. For example, as shown in FIG. 1, this information processing system includes an information processing device 10 and a plurality of communication devices 11-1 to 11-N (N>1) used by arbitrary plurality of users 1 to N. ). In the following description, unless otherwise specified, the communication devices 11-1 to 11-N may be collectively referred to as the communication device 11.

The communication device 11 is, for example, a device having a display unit (display surface) such as a smartphone, a tablet terminal, or a smart television. The communication device 11 is configured to be able to communicate with the information processing device 10 via a public network such as LTE (Long Term Evolution) or a wireless communication network such as a wireless LAN (Local Area Network). One user may use a plurality of communication devices 11, and in this case, each communication device 11 is configured to be able to communicate with the information processing device 10. If the communication device 11 is a device such as a smartphone or a tablet terminal that has a GUI (Graphic User Interface) installed in a display unit such as a liquid crystal display, the user can perform various operations using the GUI. The operation includes various operations on content such as an image displayed on the screen, such as a tap operation, a slide operation, and a scroll operation using a finger, a stylus, or the like.
Note that the communication device 11 is not limited to the device shown in FIG. 1, but may be a device such as a desktop PC (Personal Computer) or a notebook PC. In that case, operations by each user may be performed using an input device such as a mouse or a keyboard. Further, the communication device 11 may be provided with a separate display surface.

The communication device 11 can use a web service (Internet-related service) provided from the information processing device 10 or from another device (not shown) via the information processing device 10.
The web service can include a content distribution service (streaming service). The content distribution service is a service that provides video content such as movies, anime, and dramas, and may also provide advertising content (videos or still images). In this disclosure, in order to distinguish between video content such as movies, anime, and dramas and advertising content, the video content is also referred to as cultural content.
Further, the web services can include online malls, online supermarkets, or services related to communications, finance, real estate, sports, and travel.

These web services can be used by a user by, for example, registering for use with an API (Application Programming Interface) provided by the information processing device 10 or another device. Further, the web service is not limited in its type as long as it includes some kind of membership service and allows the interaction (action) between entities to be observed or managed. When registering to use a web service, each user is associated with a user ID that identifies the user.

A user transmits information regarding the user and the communication device 11 used by the user to the information processing device 10 through the use of a web service. For example, the communication device 11 can transmit information such as the IP (Internet Protocol) address of the communication device 11, the user's address, and the user's name (first and last name) to the information processing device 10. Furthermore, by viewing content using the communication device 11, a user who uses the content distribution service transmits information on the history of the viewed content (viewing history) to the information processing device 10.

Here, one use case assumed in this embodiment will be explained with reference to FIG. 2. FIG. 2 is a schematic diagram illustrating a use case assumed in this embodiment. As described above, in this embodiment, the web service provided from the information processing apparatus 10 or from another device (not shown) via the information processing apparatus 10 can include a content distribution service. In this disclosure, a user who has registered to use a content distribution service in order to use the service is referred to as a "registered user."

The registered user 20 is a user who has registered to use the content distribution service. FIG. 2 shows that family users (family members) 23 including registered user 20 use communication device 11 to view a series of contents 26 provided to registered user 20 by a content distribution service. . The communication device 11 may be, for example, a smart television 251 or a tablet terminal 252. Family users 23 are members of a family (household) that includes registered user 20 and lives together with registered user 20 . The cohabiting user 24 is a family user other than the registered user 20 among the family users 23 . The cohabiting users 24 include, for example, a spouse user 21 who is the spouse of the registered user 20 and a child user 22 who is the child of the registered user 20.

The series of content 26 is content viewed by the communication device 11 used by the family user 23, and includes content 26-1 to 26-m (m>0). Each content is given a content ID that identifies the content. Each content can be cultural content or advertising content. Each content may be content other than cultural content or advertising content. As will be described later, metadata, which is information representing the characteristics of the content, is attached to each content as supplementary information. Each family user 23 can view a content selected from a plurality of contents provided to the communication device 11 by performing a predetermined operation. Information on the viewing history of the viewed content is transmitted from the communication device 11 to the information processing device 10.

In FIG. 2, family users 23 (registered user 20, spouse user 21, and child user 22) are any of users 1 to N shown in FIG. Therefore, the information processing device 10 is configured to allow the registered user 20, the spouse user 21, and the child user 22 to acquire information through the use of web services. On the other hand, the registered user 20 is registered in the content distribution service, but the spouse user 21 and child user 22 are not registered. Therefore, the information processing device 10 recognizes the content viewing history (viewing history by the family user 23) acquired from the communication device 11 used by the family user 23 to view content as the viewing history of the registered user 20. That is, the information processing device 10 cannot explicitly distinguish and recognize the viewing history of the family user 23 from the viewing history of the registered user 20, the spouse user 21, and the child user 22.

The information processing device 10 according to the present embodiment predicts the cohabiting users 24 (in the example of FIG. 2, the spouse user 21 and the child user 22) who live together with the registered user 20 (that is, predicts the family users 23). The information processing device 10 then predicts which of the family users 23 has viewed each of the contents provided by the content distribution service to the registered users 20. Further, the information processing device 10 determines content (hereinafter referred to as recommended content) to be recommended to all family users 23 and/or each family user 23 based on the prediction result.

Specifically, the information processing device 10 first acquires various information from the communication devices 11-1 to 11-N used by the users 1 to N, and creates a social graph (relationship graph) based on the information. Create. A social graph is a graph network that shows social relationships among multiple users. Second, the information processing device 10 predicts and detects cohabiting users 24 based on the social graph. Thirdly, the information processing device 10 predicts (identifies) which of the family users 23 has viewed each of the contents provided to the registered users 20 by the content distribution service. The information processing device 10 then learns the viewing score prediction model 113 for each family user 23 based on the prediction result (identification result), and generates the viewing score prediction model 113 for the entire family user 23 . The information processing device 10 uses the viewing score prediction model 113 to determine recommended content for each family user 23 and/or the entire family user 23 .

[Functional configuration of information processing device 10]
FIG. 3 shows an example of the functional configuration of the information processing device 10 according to this embodiment.
The information processing device 10 shown in FIG. 3 includes a user feature acquisition unit 101, a viewing history acquisition unit 102, a social graph creation unit 103, a family user prediction unit 104, a viewing user prediction unit 105, a learning unit 106, and a recommended content determination unit 107. , an output unit 108, a learning model storage unit 110, and a data storage unit 120. The learning model storage unit 110 stores a user feature prediction model 111, an intimacy score prediction model 112, and a viewing score prediction model 113. The various learning models will be described later. Furthermore, the data storage unit 120 stores user characteristics 121 and viewing history 122. In explaining FIG. 3, reference will be made to FIGS. 1 and 2.

The user characteristic acquisition unit 101 acquires factual characteristics (factual information) about the communication devices and users from each of the communication devices 11-1 to 11-N as user characteristics. User characteristics are fact-based characteristics (information) that are actually or objectively obtained from a communication device or user. The user characteristic acquisition unit 101 can acquire user characteristics directly from the communication device 11, for example. Further, the user characteristic acquisition unit 101 can acquire user characteristics as information registered in a predetermined web service by the user of the communication device 11. A user ID that identifies the user is linked to the user characteristics.

User characteristics include the IP address of the communication device 11, the user's address and name, the credit card number held by the user, and the user's demographic information (demographic information such as gender, age, area of residence, occupation, family structure, etc.) user attributes), etc.
Further, the user characteristics may include a registration number and a registered name when using the web service. In addition, user characteristics can be accumulated based on call history, product delivery address other than the user's address when using a specified web service, usage status when using the web service, purchase history, usage history, search history, and service usage. It may also include information about points where it is possible.
Further, the user characteristics may include characteristics related to home time, such as days of the week and times when the user is home. The characteristics related to the home time can be obtained, for example, from the delivery date and time specified by the user or the actual delivery date and time when the user purchases an item using a web service.
Further, the user characteristics may include characteristics related to preferences, such as smoking cigarettes and drinking alcohol. The characteristics related to the preferences may be obtained, for example, based on items purchased by the user using the web service.
In this way, the user characteristics can include any information, including information related to the communication device 11 or the user himself, and information related to the use of a predetermined service via communication.

Additionally, the user feature acquisition unit 101 applies the acquired user features to the trained user feature prediction model 111 to also obtain user features estimated for the user features (estimated user features (attributes)). It is configured as follows. The user feature prediction model 111 inputs user features (i.e. factual features) of a target user and outputs the probability that each of a plurality of user features corresponds to (suits) the target user (probability of relevance). configured. The user characteristic acquisition unit 101 finally determines the estimated user characteristics of the target user based on the applicable probability.

For example, the user characteristic acquisition unit 101 inputs data indicating the target user's demographic information and past purchase history and preferences in web services to the user characteristic prediction model 111 as the user characteristics of the target user. The user characteristic prediction model 111 outputs, as corresponding probabilities, probabilities for a plurality of items that the target user is estimated to purchase, a plurality of preferences that the target user may have, and each of the behavioral trends of the target user. . Then, the user characteristic acquisition unit 101 can acquire items, tastes, and behavioral tendencies that have a probability of a predetermined value or more as the estimated user characteristics of the target user. Behavioral tendencies include, for example, characteristics related to a life stage such as having a child, and characteristics related to a lifestyle such as working from home.

The user characteristic acquisition unit 101 stores the acquired and estimated user characteristics in the data storage unit 120 as the user characteristics 121. The user characteristics are stored in association with the user ID.

The viewing history acquisition unit 102 acquires content viewing history information from the communication devices 11 of registered users who are registered with the content distribution service among the users 1 to N. The viewing history information acquired by the viewing history acquisition unit 102 corresponds to the viewing history information of the registered user. Referring to FIG. 2, the viewing history acquisition unit 102 acquires the history of content viewed by any of the family users 23 as the viewing history of the registered user 20. The viewing history information includes the content ID, metadata, and viewing date and time (viewing time) of each content viewed.

Metadata (supplementary information) is information that is given to each content and represents the characteristics of the content. For example, in the case of cultural content such as movies, anime, and dramas, metadata includes information such as genre, target age group of the content, title, cast (actors, etc.), director, content time, series number, and synopsis. sell. Furthermore, in the case of advertising content, the metadata may include information such as the advertiser, target item (product or service), genre, and the like. The viewing history acquisition unit 102 causes the data storage unit 120 to store the acquired viewing history information as the viewing history 122. The viewing history 122 is stored in association with the user ID. Referring to FIG. 2, the viewing history acquired from the registered user 20 is stored in association with the user ID of the registered user 20.

The social graph creation unit 103 creates a social graph (relationship graph) using the various user characteristics acquired by the user characteristic acquisition unit 101. The procedure for creating a social graph will be described later.

The family user prediction unit 104 predicts (searches for) family users including registered users of the content distribution service. First, the family user prediction unit 104 obtains the user ID of a registered user who is one of users 1 to N, and identifies the registered user. The user ID of the registered user may be input by the operator via the input unit (input unit 1105 in FIG. 11) or may be input via the communication unit (communication I/F 1107 in FIG. 11). . Alternatively, the user ID of the registered user may be set by any program stored in the storage unit (ROM 1102 or RAM 1103 in FIG. 11). The family user prediction unit 104 acquires the user characteristics of the registered user from the user characteristics 121 of the data storage unit 120 based on the user ID of the registered user.

Then, the family user prediction unit 104 predicts family users including the registered user based on the acquired user characteristics of the registered user and the social graph created by the social graph creation unit 103. Referring to FIG. 2, the family user prediction unit 104 predicts cohabiting users 24 based on the user characteristics and social graph of the registered users 20. Note that since there are cases where the registered user is single and there are no cohabiting users (number of family users=1), the family user prediction unit 104 is configured to be able to determine whether there are cohabiting users with respect to the registered user. The family user prediction procedure based on the social graph will be described later.

The viewing user prediction unit 105 determines whether each of the one or more contents included in the viewing history has been viewed based on the viewing history of the registered user included in the viewing history 122 and the family user predicted by the family user prediction unit 104. Predict (identify) one or more users (hereinafter referred to as viewing users). That is, the viewing user prediction unit 105 predicts which one or more users among the family users have viewed one or more contents included in the viewing history of the registered user. The prediction corresponds to an alignment between each of the one or more contents and each of the family users. Referring to FIG. 2, the viewing user prediction unit 105 predicts each viewing user of the content 26. That is, the viewing user prediction unit 105 predicts (identifies) which of the registered user 20, spouse user 21, and child user 22 has viewed each of the content 26.

In the present embodiment, the viewing user prediction unit 105 refers to the metadata and/or viewing date and time of one or more contents included in the viewing history of the registered user, and the user characteristics of the family user 23 acquired from the user characteristics 121. , make the prediction. For example, the viewing user prediction unit 105 may perform the prediction using metadata and/or viewing date and time, and common features with user characteristics. Referring to FIG. 2, the metadata of the content 26-1 includes the genre=anime, the target age group of the content=3 to 10 years old, the viewing date and time is weekday evening, and the user characteristics of the child user 22. Assume that the age included in is 8 years old. In this case, since the target age of the content included in the metadata of the content 26-1 and the age characteristics of the child user 22 are common, the viewing user prediction unit 105 determines that the viewing user of the content 26-1 is the child user 22. It can be predicted that there will be.

Instead of this, or in addition to this, the viewing user prediction unit 105 may perform the prediction based on a rule. The rule is a rule that associates content metadata and/or viewing date and time with user characteristics. An example of this rule is that a user who views content whose metadata includes genre=anime and whose viewing date and time is on a weekday evening has user characteristics such as being between 6 and 12 years old and/or being at home on a weekday evening. This is a rule that indicates that the user is a user. When the metadata of the content 26-2 includes genre=anime and the viewing date and time is weekday evening, the viewing user prediction unit 105 determines that the viewing user of the content 26-2 is based on the exemplary rule. , child user 22 (8 years old). Furthermore, when the user characteristics 121 of the spouse user 21 include being at home on weekday evenings, the viewing user prediction unit 105 determines that the viewing user of the content 26-2 is based on the exemplary rule. , furthermore, it can be predicted that the spouse user 21 will be included.

Further, the rule may be a rule that associates metadata and/or viewing date and time of the content with a behavioral tendency included in the user characteristics. For example, a rule is set in which content having metadata indicating a long content duration (for example, 2 hours or more) is associated with a user whose behavior tendency includes working from home. The example rule is based on the prediction (e.g., a rule of thumb) that users who work from home are more likely to watch content with longer content durations because they save time commuting and spend more time at home. If the metadata of the content 26-3 includes a long content time and the behavioral tendency of the registered user 20 includes working from home, the viewing user prediction unit 105 determines that the viewing user of the content 26-3 is the registered user 20. It can be predicted that Thereby, the viewing user prediction unit 105 can predict the viewing users of each content according to the behavioral tendencies of each family user.

The rules can be created by, for example, empirical rules or machine learning. The rule may be input by the operator via the input unit (input unit 1105 in FIG. 11) or via the communication unit (communication I/F 1107 in FIG. 11). Alternatively, the rule may be set in advance in the system, or may be set by any program stored in the storage unit (ROM 1102 or RAM 1103 in FIG. 11).

Through such prediction processing, the viewing user prediction unit 105 associates each content included in the viewing history 122 with one or more viewing users. Furthermore, the viewing user prediction unit 105 generates (prepares) combination data of each content included in the viewing history 122 and one or more viewing users based on the association. The combination data includes the content ID and metadata of the content, and the user characteristics of the viewing data associated with the content. Further, the combination data may include information on the viewing date and time of the content.

The learning unit 106 uses the combination data generated by the viewing user prediction unit 105 to train the viewing score prediction model 113. The viewing score prediction model 113 is a learning model for machine learning that predicts the viewing score of the content by inputting the combined data of the metadata of the content and the user characteristics of the user. The viewing score indicates the probability (possibility) that the input content will be viewed by a user having the input user characteristics, and is expressed, for example, as a numerical value between 0 and 1 (maximum probability=1). The viewing score indicates not only the probability of viewing but also the probability that the content will be selected by any action, such as the probability that the content will be selected by clicking on the screen on which the content is displayed. The learning procedure of the viewing score prediction model 113 by the learning unit 106 will be described later.

The recommended content determining unit 107 determines recommended content using the viewing score prediction model 113 learned by the learning unit 106. Referring to FIG. 2, the recommended content determining unit 107 determines recommended content for the family user 23. The procedure for determining recommended content will be described later.

The output unit 108 outputs the recommended content determined by the recommended content determining unit 107. For example, the output unit 108 generates information regarding recommended content, and outputs (displays) the information to the display unit of the communication device 11 used by the registered user. The output unit 108 may generate and output information regarding the recommended content. The output may include any form of output, such as visual output or audio output.

[Social graph creation procedure]
Next, a procedure for creating a social graph according to this embodiment will be explained. Note that in the following description, users A to E are users referred to for the purpose of description, and may be users of the communication device 11. Further, the social graph is composed of connections between multiple entities (corresponding to the circled user nodes in the examples of FIGS. 5A and 5B). In the following description, the user node will simply be referred to as a user. FIG. 4 shows a flowchart of the social graph creation process executed by the social graph creation unit 103 according to this embodiment. Each step of the process shown in FIG. 4 will be explained below.

<S41: Creating explicit and implicit links in the social graph>
In S41, the social graph creation unit 103 predicts and creates explicit links and implicit links between multiple users.
Link creation processing in the social graph will be described with reference to FIGS. 5A and 5B. FIG. 5A is a diagram for explaining explicit links in a social graph, and FIG. 5B is a diagram for explaining implicit links in a social graph. An explicit link in a social graph is a link created by explicit common user characteristics between two users (user pair). An implicit link in a social graph is a link that is created as an indirect relationship by using an already created explicit link, although the existence of explicit common user characteristics between the user pair is unclear. It is.

FIG. 5A shows an example of creating explicit links in a social graph using the IP address of a user's communication device as a common user feature. FIG. 5A shows an example in which an online mall 51, a golf course reservation service 52, a travel-related reservation service 53, and a card management system 54 exist as web services available to users A to C. Although these four web services are shown in FIG. 5A, the number of web services is not limited to a particular number.

The online mall 51 is a shopping mall that is available online (using the Internet). The online mall 51 can provide a wide variety of products and services, such as fashion, books, food, concert tickets, and real estate, for example.
The golf course reservation service 52 is operated by a website that provides services related to golf courses online, and can provide, for example, golf course searches, reservations, and lesson information.
The travel-related reservation service 53 is operated by a website that provides various travel services available online. The travel-related reservation service 53 can provide, for example, reservations for hotels and travel tours, reservations for airline tickets and rental cars, tourist information, hotels, and information around the hotels.
The card management system 54 is operated by a website that provides services related to credit cards issued and managed by a predetermined card management company. The card management system 54 may provide services in association with at least one of the online mall 51, the golf course reservation service 52, and the travel-related reservation service 53.

In the example of FIG. 5A, users A to C use the same IP address (=198.45.66.xx) to use the online mall 51, the golf course reservation service 52, and the travel-related reservation service 53. . Information on the IP address can be acquired by the user characteristic acquisition unit 101.
In such a case, the social graph creation unit 103 creates explicit links (solid line links) between users A to C in the social graph using the same IP address characteristics, as shown in link status 55.

In addition to FIG. 5A, explicit links in the social graph can be created using common user characteristics such as the characteristics of the user's address and the characteristics of the credit card number used by the user.

FIG. 5B shows an example of creating implicit links between users in a social graph. In the example of FIG. 5B, user C, user D, and user E are connected to user A by explicit links, and user C, user D, and user E are connected to user B by explicit links. There is. Such link features (features that indicate relationships between links) are embedded in a common feature space, and links that are inferred to implicitly establish relationships between each user (each node) are called implicit links. Created (established) as In the example of FIG. 5B, although users A and B are not connected by an explicit link, an implicit link (dotted line link) is created as a result of inferring that they have a relationship in a common feature space. There is. Note that the social graph creation unit 103 performs learning (representation learning, relational learning, embedding learning, knowledge graph embedding) of an inter-user relationship graph composed of nodes (users) connected by explicit links. Predict and create implicit links between users. At this time, the social graph creation unit 103 may perform the learning based on a known embedded model or its extension as appropriate.

<S42: Inference of relationships between links>
In S42, the social graph creation unit 103 infers the relationship between the links predicted and created in S41. Processing for inferring relationships between links will be described with reference to FIGS. 6A and 6B. FIG. 6A is a diagram for explaining the process of inferring the relationship between links, and shows an example of inferring the relationship between the links between user A and user B who are connected by an explicit link.

The social graph creation unit 103 treats the pair of users connected by the link created in S41 as a data point, and uses the various information acquired by the user feature acquisition unit 101 to define the pair (data point) as a common group into clusters representing types of The various information can be information such as IP address, address, credit card, age, gender, and friends. Further, each cluster can be a cluster having relationships such as spouses, parents and children, neighbors, same households, colleagues, friends, same-sex siblings, and different-sex siblings. In the example of FIG. 6A, pairs of users are indicated by cross marks, and the clusters in which the pairs can be grouped include a parent-child cluster 61, a spouse cluster 62, a same-sex sibling cluster 63, a friend cluster 64, and a same-household cluster 65. show. Note that although five clusters are shown in FIG. 6A, the number of clusters is not limited to a specific number.

For example, if user A and user B have (share) characteristics 60 such as the same last name, an age difference of less than 10 years, opposite genders, and the same address, the social graph creation unit 103 , the pair of user A and user B can be grouped into a cluster (spouse cluster 62) representing a relationship between husband and wife (spouse).

FIG. 6B shows a flowchart of an example of a process of grouping pairs into clusters, which is executed by the social graph creation unit 103.
At the start of S61, it is assumed that the pairs to be grouped have the same address and the same last name. In S62, the social graph creation unit 103 determines whether or not the subject pair has the same gender characteristics. If the pair of subjects have the same gender characteristics (Yes in S62), in S63 the social graph creation unit 103 determines whether the age difference between the pair of subjects is equal to or less than a predetermined threshold (=X value). judge. If the age difference between the pair of subjects is greater than the X value (No in S63), the social graph creation unit 103 groups the pair of subjects into a parent-child cluster 61. If the age difference is less than or equal to the X value (Yes in S63), the social graph creation unit 103 groups the subject pair into a same-sex sibling cluster 63. Furthermore, if the pair of targets does not have the same gender characteristics (No in S62), the social graph creation unit 103 determines whether the age difference between the pair of targets is equal to or less than a predetermined threshold (=Y value) in S64. Determine whether or not. If the age difference is greater than the Y value (No in S64), the social graph creation unit 103 groups the target pair into a parent-child cluster 61. If the age difference is less than or equal to the Y value (Yes in S64), the social graph creation unit 103 groups the target pair into a spouse cluster 62.

<S43: Assignment of intimacy score based on intimacy of relationship>
In S43, the social graph creation unit 103 predicts a score based on the intimacy of the relationship for the pair inferred in S42 (a score representing the height of the social relationship), and assigns the score to the pair. In this embodiment, this score is referred to as an intimacy score. In this embodiment, the familiarity score is a numerical value between 0 and 1, but there is no particular limitation on the numerical value that the familiarity score can take. FIG. 7A shows a conceptual diagram of the intimacy score based on the intimacy of the relationship between user pairs.

In the example of FIG. 7A, the intimacy of the relationship between the user pair changes depending on the characteristics that users A and B, who are connected by an explicit link, have (share). In the upper part of FIG. 7A, if user A and user B have the characteristics 70 of being siblings of the same gender, same address, 1200 call history, and 50 gift exchanges, the parent of the relationship between the user pair The density (ie, familiarity score) will be high. On the other hand, in the lower part of FIG. 7A, if user A and user B have the characteristics 71 of being siblings of the same gender, different addresses, 30 call history, and 2 gift exchanges, the relationship between the user pair is The familiarity (i.e., intimacy score) of will be low. In this way, even if users A and B are siblings of the same gender, as in the example of FIG. It will be different. Pairs with high relationship intimacy are observed to have close social distance and high influence. On the other hand, pairs with low relationship intimacy are observed to have a long social distance from each other and are not close to each other.

In this embodiment, the familiarity score for a user pair is predicted using the familiarity score prediction model 112. FIG. 7B shows a schematic architecture of the familiarity score prediction model 112. The intimacy score prediction model 112 is a learning model that predicts the intimacy score 73 for the feature 72 by inputting the feature 72 of the user pair (feature shared by the pair).

The intimacy score prediction model 112 is, for example, a learning model that performs weak supervised learning. In this embodiment, the familiarity score prediction model 112 is a learning model trained using familiarity scores (0 to 1) assigned to multiple features for user pairs as training data, as shown in FIG. 6A. shall be. For example, in the learning stage, combined data such as a familiarity score close to 1 set to the feature 70 in FIG. 7A and a closeness score close to 0 set to the feature 71 in FIG. 7A is used as the teacher data. Note that the intimacy score prediction model 112 may be different for each type of relationship between user pairs, and may be a learning model trained according to one type of relationship.

Note that in this embodiment, the intimacy score for a user pair is predicted using the intimacy score prediction model 112, but the social graph creation unit 103 is configured to predict the score using another method. Good too. In this way, the social graph creation unit 103 assigns a familiarity score to a pair of users based on the characteristics shared by the pair.

Through the above processing, explicit or implicit links are formed between a plurality of users, a familiarity score is assigned between each link, and a social graph is created. A conceptual diagram of the social graph is shown in FIG. 7C. In the social graph 700 shown in FIG. 7C, each of the users 701 to 703 has a plurality of characteristics, and the predicted intimacy score is assigned to the pair of users as described above.

[Family user prediction procedure]
Next, a family user prediction procedure according to the present embodiment will be described. In this embodiment, the family user prediction unit 104 predicts family users including registered users based on the social graph created by the social graph creation unit 103. In this embodiment, family users are predicted in the social graph based not only on intimacy scores (see FIG. 7C) but also on shared (common) user characteristics between each pair (see FIG. 6A).

The family user prediction unit 104 first determines whether there is a cohabiting user who lives with the registered user. For example, the family user prediction unit 104 determines whether a user has a high intimacy score in relation to the registered user (for example, the intimacy score is equal to or higher than a predetermined threshold) in the social graph, and who has a common address and the same last name as the registered user. A search is made for a user who satisfies the first condition of having a user. If there is a user who satisfies the first condition, it can be determined that there is a cohabiting user, and if not, it can be determined that there is no cohabiting user. Furthermore, the family user prediction unit 104 may search for a user who satisfies the second condition of having a common address with the registered user in the social graph. In this case, if there is a user who satisfies the second condition, it can be determined that there is a cohabiting user, and if not, it can be determined that there is no cohabiting user.

If it is determined that there are no users living together, the family user prediction unit 104 predicts that the only family users are registered users. On the other hand, if it is determined that there is a cohabiting user, a user who satisfies the conditions used for determining the presence or absence of a cohabiting user (for example, the above-mentioned first condition or second condition) is predicted to be a cohabiting user. Next, the family user prediction unit 104 extracts the relationship (family composition) between the registered user and the cohabiting user based on the social graph. In the example of FIG. 2, the family user prediction unit 104 recognizes that the spouse user 21 is the spouse of the registered user 20, and that the child user 22 is the child of the registered user 20.

[Learning procedure for viewing score prediction model]
Next, the learning procedure of the viewing score prediction model 113 will be explained with reference to FIG. 2 and FIGS. 8A to 8C. FIGS. 8A and 8B are diagrams for explaining the learning procedure of the viewing score prediction model 113. The viewing score prediction model 113 according to the present embodiment is a tree (decision tree)-based model using boosting, and is, for example, a learning model based on XGBoost. Note that the viewing score prediction model 113 may be learned or integrated individually for each time period.

The relationship between input data and output data of the viewing score prediction model 113 will be described with reference to FIG. 8A. In the example of FIG. 8A, combination data of the content 26-1 and the user characteristics 80 is input to the viewing score prediction model 113. Here, the input content includes the content ID and metadata of the content. Further, the user characteristics 80 include demographic information, purchase history, and service usage status on the web service. User characteristics 80 may be configured to include other types of user characteristics. The viewing score 81 output from the viewing score prediction model 113 for the input combination data indicates that the input content 26-1 is selected (including viewing) by a user having the input user characteristics 80. Indicates probability (possibility).

During learning, the viewing score prediction model 113 is trained using the combination data generated by the viewing user prediction unit 105. Since the combination data indicates a combination of the viewed content and the viewing user, the viewing score output from the viewing score prediction model 113 is learned as 1 (maximum probability). For example, in FIG. 8A, combination data of content 26-1 and user characteristics 80 is input, and learning is performed to output a viewing score 81 (=1).

With reference to FIG. 8B, a learning process using the combination data generated by the viewing user prediction unit 105 for the family user 23 shown in FIG. 2 will be described. Referring to FIG. 2, assume that family users 23 including registered user 20 view content 26. In this case, the viewing user prediction unit 105 generates combination data of the user characteristics of at least one of the registered user 20, spouse user 21, and child user 22 and each of the contents 26-1 to 26-m. In the example of FIG. 8B, the content 26 includes contents 26-1 to 26-9 (ie, m=9). The generated combination data includes combination data of each of the contents 26-1 to 26-3 and the user characteristics 80 of the registered user 20, and combination data of each of the contents 26-4 to 26-6 and the user characteristics 83 of the spouse user 21. , and combination data of each of the contents 26-6 to 26-9 and the user characteristics 84 of the child user 22. Content 26-6 is combined with user characteristics of spouse user 21 and child user 22. This means that the viewing user prediction unit 105 predicts that the viewing users of the content 26-6 are the spouse user 21 and the child user 22, and that the content 26-6 is associated with both users. .

Using the combined data, a viewing score prediction model 113 (user-based viewing score prediction model) is learned for each family user 23. In FIG. 8B, the viewing score prediction models 113 of the registered user 20, spouse user 21, and child user 22 are shown as viewing score prediction models 113-1, 113-2, and 113-3, respectively. The viewing score prediction model 113-1 receives the combination data of each of the contents 26-1 to 26-3 and the user characteristics 82 of the registered user 20, and learns to output the viewing score 81-1 (=1). be done. The viewing score prediction model 113-2 receives the combination data of each of the contents 26-4 to 26-6 and the user characteristics 83 of the spouse user 21, and outputs the viewing score 81-2 (=1). be learned. The viewing score prediction model 113-3 receives the combination data of each of the contents 26-6 to 26-9 and the user characteristics 84 of the child user 22, and learns to output the viewing score 81-3 (=1). be done.

Similar to FIG. 8A, user characteristics 82-84 are configured to include, but are not limited to, demographic information, purchase history, and service usage on the web service. Further, the user characteristics 83 of the spouse user 21 and the user characteristics 84 of the child user 22 may include the relationship in the family structure with the registered user 20. For example, the user feature 83 may include a feature indicating "spouse" and the user feature 84 may include a feature indicating "child."

Furthermore, when content 26-6 to content 26-9 include viewing date and time information of the content, viewing score prediction models 113-1 to 113-3 are configured to weight the content according to the viewing date and time. May be learned. For example, the viewing score prediction models 113-1 to 113-3 may be trained such that the closer the viewing date and time is to the current or predetermined time period, the higher the viewing score becomes. As a result, for example, content having metadata similar to recently viewed content may be more likely to be determined as recommended content.

The learned viewing score prediction models 113-1 to 113-3 are integrated into the viewing score prediction model 113 for all family users. FIG. 8B shows a conceptual diagram of the viewing score prediction model 113 that integrates the learned viewing score prediction models 113-1 to 113-3. By using the integrated viewing score prediction model 113, it becomes possible to predict the viewing score of the entire family user 23 for the input content (including metadata and content ID). The learning unit 106 stores the viewing score prediction model 113 generated by the integration in the learning model storage unit 110 as the trained viewing score prediction model 113. The learning unit 106 may also store the viewing score prediction models 113 (for example, the viewing score prediction models 113-1 to 113-3 in FIG. 8B) learned for each user in the learning model storage unit 110.

When the family user prediction unit 104 determines that there are users living together, for example, as shown in FIG. 8B, viewing score prediction models 113-1 to 113-3 are used using combination data of each family user and each content. is learned. On the other hand, if it is predicted that there are no cohabiting users, all the viewed content will have been viewed by registered users, and learning can be performed using more learning data. Therefore, the viewing score prediction model 113 may be trained preferentially for registered users predicted by the family user prediction unit 104 to have no cohabiting users.

For example, referring to FIG. 2, assume that the registered user 20 has viewed all of the content 26. In this case, the viewing user prediction unit 105 generates combination data of the user characteristics of the registered user 20 and each of the contents 26-1 to 26-m. The viewing score prediction model 113 receives input of combination data of each of the contents 26-1 to 26-m and the user characteristics of the registered user 20, and is trained to output a viewing score (=1). Such learning processing for only registered users can improve the accuracy of viewing scores for users who have the user characteristics of registered users.

[Recommendation content prediction procedure]
Next, a procedure for determining recommended content by the recommended content determining unit 107 will be described with reference to FIG. 2, FIG. 9A, and FIG. 9B. FIGS. 9A and 9B are diagrams for explaining a procedure for determining recommended content. In this embodiment, the recommended content determining unit 107 is configured to determine recommended content using the viewing score prediction model 113 learned and generated by the learning unit 106.

In FIG. 9A, content 91 including content 91-1 to 91-n (n>0) is recommendation candidate content (candidate content) prepared in advance. The content 91 may be input by an operator via an input unit (input unit 1105 in FIG. 11) or via a communication unit (communication I/F 1107 in FIG. 11). Alternatively, the content 91 may be set in advance in the system, or may be set by any program stored in the storage unit (ROM 1102 or RAM 1103 in FIG. 11).

The recommended content determining unit 107 inputs the content 91 (including metadata and content ID) and the user characteristics of each family user 23 into the trained viewing score prediction model 113. Specifically, the recommended content determining unit 107 generates combination data of each of the contents 91 and the user characteristics 82 of the registered user 20, combination data of each of the contents 91 and the user characteristics 83 of the spouse user 21, and the combination data of the contents 91 and the user characteristics 83 of the spouse user 21. and the user characteristics 83 of the child user 22 are input into the viewing score prediction model 113. The viewing score prediction model 113 predicts and outputs the viewing score 92 of the entire family user 23 from the input data. The recommended content determining unit 107 can determine, among the contents 91-1 to 91-n, the content having the viewing score 92 equal to or higher than a predetermined threshold as the recommended content for the family user 23. The predetermined threshold is, for example, 0.7. Content with a higher viewing score 92 is likely to be viewed by at least one of the family users 23, and the recommendation is highly effective. Note that the content 91 may include at least a part of the content 26 used in the learning stage.

In this way, by using the viewing score prediction model 113, content to be recommended to all family users can be determined. In addition to this, or in place of this, the recommended content determining unit 107 may be configured to determine recommended content for each user making up the family users. With reference to FIG. 9B, a procedure for determining recommended content for each of the family users 23 in FIG. 2 will be described.

In FIG. 9B, the recommended content determination unit 107 inputs the content 91 (including metadata and content ID) and the user characteristics of each family user 23 into trained viewing score prediction models 113-1 to 113-3. do. Specifically, the recommended content determining unit 107 inputs the combination data of each of the contents 91 and the user characteristics 82 of the registered user 20 to the viewing score prediction model 113-1. Similarly, the recommended content determining unit 107 inputs the combination data of each of the contents 91 and the user characteristics 83 of the spouse user 21 into the viewing score prediction model 113-2, The combination data with the user characteristics 83 is input to the viewing score prediction model 113-3. The viewing score prediction models 113-1 to 113-3 each calculate the viewing score 92-1 of the registered user 20, the viewing score 92-2 of the spouse user 21, and the viewing score 92-3 of the child user 22 from the input data. Predict and output. The recommended content determining unit 107 can determine content having a viewing score equal to or higher than a predetermined threshold value (for example, 0.7) from among the contents 91-1 to 91-n to be recommended content. In the example of FIG. 8B, the recommended content determining unit 107 can determine content having a viewing score 92-1 equal to or higher than a predetermined threshold as the recommended content for the registered user 20. Similarly, the recommended content determining unit 107 determines the content having the viewing score 92-2 equal to or higher than a predetermined threshold as the recommended content for the spouse user 21, and the content having the viewed score 92-2 equal to or higher than the predetermined threshold. The content may be determined to be recommended content for the child user 22. Note that the content 91 may include at least a part of the content 26 used in the learning stage.

Further, the recommended content determining unit 107 may determine the viewing score 93 of the entire family user 23 from the viewing scores 92-1 to 92-3. For example, the recommended content determining unit 107 may calculate the sum of the viewing scores 92-1 to 92-3 for each content, and determine the sum as the viewing score 93. Alternatively, the recommended content determining unit 107 may extract the maximum score of each of the viewing scores 92-1 to 92-3 for each content, and determine the extracted three maximum scores as the viewing score 93. Then, the recommended content determining unit 107 can determine the content having the viewing score 93 equal to or higher than a predetermined threshold as the recommended content for the entire family user 23 .

The recommended content for all family users 23 determined by the recommended content determining unit 107 is displayed on the communication device 11 of the registered user 20 (for example, the smart TV 251 or the tablet terminal 252 in FIG. 2) by the output unit 108. . Here, the learning unit 106 calculates the viewing score prediction model 113 (and/or the viewing score prediction models 113-1 to 113-3) using the conversion rate (CVR) based on whether the viewing was performed by any of the family users. You can make them learn more.

[Overall processing flow]
FIG. 10 shows a flowchart of the overall processing executed by the information processing device 10 according to this embodiment. For explanation of FIG. 10, reference will be made to the information processing system shown in FIG.

In S1001, the user characteristic acquisition unit 101 acquires the user characteristics of each user from the communication devices 11-1 to 11-N, and stores them as the user characteristics 121 in the data storage unit 120. The process of S1001 may be a process of acquiring (collecting) user characteristics and estimated user characteristics during a certain period in the past.

In S1002, the viewing history acquisition unit 102 acquires content viewing history information from the communication devices 11 of registered users who are registered with the content distribution service among the users 1 to N. The viewing history information includes the content ID, metadata, and viewing date and time (viewing time) of each content viewed. The viewing history acquisition unit 102 stores the acquired viewing history information in the data storage unit 120 as the viewing history 122.

In S1003, the social graph creation unit 103 creates social graphs for users 1 to N using the user characteristics 121 stored in the data storage unit 120 in S1001. The social graph creation unit 103 assigns a familiarity score to each user pair in the social graph based on characteristics common to the pair. The specific procedure for creating the social graph is as described above.

In S1004, the family user prediction unit 104 identifies registered users of the content distribution service and predicts family users including the registered users. The family user prediction unit 104 acquires the input (set) user ID of the registered user, and acquires the user characteristics (user characteristics of the registered user) corresponding to the user ID of the registered user from the user characteristics 121. Next, the family user prediction unit 104 predicts users who live together with the registered user based on the acquired user characteristics of the registered user and the social graph created in S1003, and predicts family users. The specific prediction procedure for family users is as described above.

In S1005, the viewing user prediction unit 105 determines the viewing history of the registered user included in the viewing history 122 stored in the data storage unit 120 in S1002 and the family user predicted in S1004, based on the viewing history of the registered user included in the viewing history 122 stored in the data storage unit 120 in S1002. Predict (identify) viewing users of one or more contents. The prediction corresponds to an alignment between each of the one or more contents and each of the family users. Furthermore, the viewing user prediction unit 105 generates combination data of each content included in the viewing history 122 and one or more viewing users based on the association. The combination data includes the content ID and metadata of the content, and the user characteristics of the viewing data associated with the content. Further, the combination data may include information on the viewing date and time of the content.

In S1006, the learning unit 106 trains the viewing score prediction model 113. Specifically, the learning unit 106 first trains the viewing score prediction model 113 corresponding to each viewing user based on the combination data of content and viewing users generated in S1005. In the example of FIG. 8B, the learning unit 106 trains the viewing score prediction models 113-1 to 113-3. Furthermore, the learning unit 106 generates the viewing score prediction model 113 for all family users by integrating the viewing score prediction models 113 corresponding to each viewing user. In the example of FIG. 8C, the learning unit 106 generates the viewing score prediction model 113 that integrates the viewing score prediction models 113-1 to 113-3. The specific learning and generation procedure of the viewing score prediction model 113 is as described above.

In S1007, the recommended content determining unit 107 determines recommended content for the family user. The recommended content determining unit 107 determines recommended content for family users using the viewing score prediction model 113 learned in S1006. Here, the recommended content determining unit 107 may determine recommended content for each family user. The specific procedure for determining recommended content is as described above.

In S1008, the output unit 108 presents the recommended content determined in S1007 to family users including the registered user. For example, referring to FIG. 2, the output unit 108 displays recommended content for the family user 23 on the communication device 11 of the registered user 20 (for example, the smart TV 251 or the tablet terminal 252 in FIG. 2). Here, the recommended content can be displayed in such a way that the content, such as a partial image of the content, the title and performers of the content, can be grasped.

[Hardware configuration of information processing device 10]
FIG. 11 is a block diagram showing an example of the hardware configuration of the information processing device 10 according to this embodiment.
The information processing apparatus 10 according to this embodiment can be implemented on any single or multiple computers, mobile devices, or any other processing platform.
Referring to FIG. 11, an example is shown in which the information processing device 10 is implemented in a single computer, but the information processing device 10 according to the present embodiment is implemented in a computer system including a plurality of computers. good. A plurality of computers may be connected to each other through a wired or wireless network so that they can communicate with each other.

As shown in FIG. 11, the information processing device 10 may include a CPU 1101, a ROM 1102, a RAM 1103, an HDD 1104, an input section 1105, a display section 1106, a communication I/F 1107, and a system bus 1108. The information processing device 10 may also include an external memory.
A CPU (Central Processing Unit) 1101 centrally controls operations in the information processing apparatus 10, and controls each component (1102 to 1107) via a system bus 1108, which is a data transmission path.

A ROM (Read Only Memory) 1102 is a nonvolatile memory that stores control programs and the like necessary for the CPU 1101 to execute processing. Note that the program may be stored in a nonvolatile memory such as an HDD (Hard Disk Drive) 1104 or an SSD (Solid State Drive), or an external memory such as a removable storage medium (not shown).
A RAM (Random Access Memory) 1103 is a volatile memory and functions as the main memory, work area, etc. of the CPU 1101. That is, the CPU 1101 loads necessary programs and the like from the ROM 1102 into the RAM 1103 when executing processing, and implements various functional operations by executing the programs and the like. The learning model storage unit 110 and data storage unit 120 shown in FIG. 2 may be configured with a RAM 1103.

The HDD 1104 stores, for example, various data and information necessary when the CPU 1101 performs processing using a program. Further, the HDD 1104 stores, for example, various data and various information obtained by the CPU 1101 performing processing using programs and the like.
The input unit 1105 includes a pointing device such as a keyboard and a mouse.
The display unit 1106 includes a monitor such as a liquid crystal display (LCD). The display section 1106 may function as a GUI (Graphical User Interface) by being configured in combination with the input section 1105.

The communication I/F 1107 is an interface that controls communication between the information processing device 10 and external devices.
Communication I/F 1107 provides an interface with a network and executes communication with an external device via the network. Via the communication I/F 1107, various data, various parameters, etc. are transmitted and received with an external device. In this embodiment, the communication I/F 1107 may perform communication via a wired LAN (Local Area Network) or a dedicated line that complies with communication standards such as Ethernet (registered trademark). However, the network that can be used in this embodiment is not limited to this, and may be configured as a wireless network. This wireless network includes a wireless PAN (Personal Area Network) such as Bluetooth (registered trademark), ZigBee (registered trademark), and UWB (Ultra Wide Band). It also includes wireless LAN (Local Area Network) such as Wi-Fi (Wireless Fidelity) (registered trademark) and wireless MAN (Metropolitan Area Network) such as WiMAX (registered trademark). Furthermore, it includes wireless WAN (Wide Area Network) such as LTE/3G, 4G, and 5G. Note that the network only needs to connect each device so that they can communicate with each other, and the communication standard, scale, and configuration are not limited to the above.

At least some of the functions of each element of the information processing apparatus 10 shown in FIG. 3 can be realized by the CPU 1101 executing a program. However, at least some of the functions of each element of the information processing device 10 shown in FIG. 2 may operate as dedicated hardware. In this case, the dedicated hardware operates under the control of the CPU 1101.

[Hardware configuration of communication device 11]
The hardware configuration of the communication device 11 shown in FIG. 1 may be the same as that in FIG. 11. That is, the communication device 11 can include a CPU 1101, a ROM 1102, a RAM 1103, an HDD 1104, an input section 1105, a display section 1106, a communication I/F 1107, and a system bus 1108. The communication device 11 displays various information provided by the information processing device 10 on the display unit 1106, and performs processing corresponding to input operations received from the user via the GUI (configured by the input unit 1105 and the display unit 1106). be able to.

In this manner, in this embodiment, family users including registered users are predicted from a social graph created based on the user characteristics of an unspecified number of users. Then, the viewer of each of the plurality of contents provided to the registered user (that is, viewed by the registered user) is predicted (identified). A user-based viewing score prediction model 113 (first learning model) for each family user is learned using the combined data obtained by the prediction (identification), and by integrating the plurality of learned first learning models. , a viewing score prediction model 113 (second learning model) for the entire family user 23 is generated. The generated viewing score prediction model 113 is used to determine recommended content for all family users. The determined recommended content is content with a higher viewing score (the probability that the content will be selected), which is the output of the viewing score prediction model 113. By presenting the recommended content to family users, it is expected that the effectiveness of the recommendation, such as improved CVR, will be improved.

Note that although specific embodiments have been described above, these embodiments are merely illustrative and are not intended to limit the scope of the present invention. The apparatus and methods described herein may be embodied in forms other than those described above. In addition, appropriate omissions, substitutions, and changes may be made to the embodiments described above without departing from the scope of the present invention. Such omissions, substitutions, and modifications are included within the scope of the claims and their equivalents, and fall within the technical scope of the present invention.

The disclosure of this embodiment includes the following configurations.
[1] Graph creation means for creating a relationship graph showing social relationships among the plurality of users based on user characteristics of the plurality of users; , a prediction means for predicting family users including registered users registered in a content distribution service and cohabiting users who live together with the registered user; an identification means for identifying by which of the family users each piece of content has been viewed; and a determination unit for determining recommended content to be recommended to the family user using a learning model based on the result of the identification; An information processing device comprising: means.

[2] The learning model is configured to input metadata representing characteristics of candidate content and user characteristics of the family user, and output a probability that the candidate content is selected by the family user; The information processing device according to [1], wherein the determining means determines the recommended content for the family user using the learning model.

[3] The user characteristics of each of the family users include behavioral tendencies of each of the family users, and the identification means identifies by which of the family users each of the plurality of contents was viewed. The information processing device according to [1] or [2], which identifies family users according to their behavioral tendencies.

[4] For each of the plurality of contents, the identification means generates combination data that is a combination of metadata representing characteristics of the content and the user characteristics of one or more users who are identified as having viewed the content. The information processing device according to any one of [1] to [3], which is generated as a result of.

[5] Further comprising learning means for generating the learning model, the learning means using the combined data to learn a first learning model for each of the family users, and learning the first learning model for each of the family users. The information processing device according to [4], which generates a second learning model that integrates learning models as the learning model.

[6] The information processing according to any one of [1] to [5], wherein the identification means identifies on a rule basis which of the family users has viewed each of the plurality of contents. Device.

[7] The information processing device according to any one of [1] to [6], wherein the plurality of contents include at least one of cultural content including movies, anime, or dramas, and advertising content.

1 to N: User, 10: Information processing device, 11-1 to 11-N: Communication device, 101: User feature acquisition unit, 102: Viewing history acquisition unit, 103: Social graph creation unit, 104: Family user prediction unit , 105: Viewing user prediction unit, 106: Learning unit, 107: Recommendation content determination unit, 108: Output unit, 110: Learning model storage unit, 111: User feature prediction model, 112: Familiarity score prediction model, 112: Viewing score prediction model, 120: Data storage unit, 121: User characteristics, 122: Viewing history

Claims (9)

Graph creation means for creating a relationship graph showing social relationships among the plurality of users based on user characteristics of the plurality of users;
A prediction unit that uses the relationship graph to predict family users among the plurality of users, including registered users who are registered in a content distribution service and cohabiting users who live together with the registered users;
Identification means for identifying by which of the family users each of the plurality of contents provided to the registered user by the content distribution service was viewed;
a determining means that uses a learning model to determine recommended content to be recommended to the family user based on the identification result;
An information processing device comprising: The learning model is configured to input metadata representing characteristics of candidate content and the user characteristics of the family user, and output a probability that the candidate content is selected by the family user,
The determining means determines the recommended content for the family user using the learning model.
The information processing device according to claim 1, characterized in that: The user characteristics of each of the family users include behavioral tendencies of each of the family users,
The identification means identifies which of the family users has viewed each of the plurality of contents according to the behavioral tendency of each of the family users.
The information processing device according to claim 1, characterized in that: The identification means, for each of the plurality of contents, combines metadata representing characteristics of the content with the user characteristics of one or more users who are identified as having viewed the content, and generates combination data as a result of the identification. generate,
The information processing device according to claim 1, characterized in that: further comprising learning means for generating the learning model,
The learning means uses the combined data to learn a first learning model for each of the family users,
Generating a second learning model that integrates the first learning models of all the family users as the learning model;
The information processing device according to claim 4, characterized in that: The identification means identifies, on a rule basis, by which of the family users each of the plurality of contents has been viewed.
The information processing device according to claim 1. The plurality of contents include at least one of cultural content including movies, anime, or dramas, and advertising content.
The information processing device according to claim 1, characterized in that: a graph creation step of creating a relationship graph showing social relationships among the plurality of users based on user characteristics of the plurality of users;
A prediction step of predicting family users, including registered users who are registered in a content distribution service and cohabiting users who live together with the registered users, among the plurality of users, using the relationship graph;
an identification step of identifying which of the family users has viewed each of the plurality of contents provided to the registered user by the content distribution service;
a determining step of determining recommended content to be recommended to the family user using a learning model based on the identification result;
An information processing method characterized by comprising: An information processing program for causing a computer to perform information processing, the program causing the computer to:
Graph creation processing that creates a relationship graph showing social relationships among the plurality of users based on user characteristics of the plurality of users;
Prediction processing that uses the relationship graph to predict family users among the plurality of users, including registered users who are registered in a content distribution service and cohabiting users who live together with the registered users;
Identification processing that identifies which of the family users has viewed each of the plurality of contents provided to the registered user by the content distribution service;
A determination process for determining recommended content to be recommended to the family user using a learning model based on the identification result,
Information processing program.

Images