User relationship estimation device, user relationship estimation method, and program

Abstract

This user relationship estimation device comprises: a user relationship information acquisition unit that acquires user relationship information indicating a relationship between some users among a plurality of users; a user action information acquisition unit that acquires user action information indicating the objects of respective actions by the plurality of users and time periods of said actions; a knowledge data acquisition unit that acquires knowledge data for expressing a relationship of an entity with the objects of the respective actions by the plurality of users; and an estimation unit that estimates a relationship between remaining users among the plurality of users on the basis of the acquired user relationship information, the user action information, and the knowledge data.

Description

User relationship estimation device, user relationship estimation method, and program

[0001] The present disclosure relates to a user relationship estimation device, a user relationship estimation method, and a program.

[0002] 2. Description of the Related Art Conventionally, there has been known a technique for determining human relationships (for example, family relationships) between users by comparing the movement histories of wireless terminals carried by each user.

[0003] JP 2016-122289 A

[0004] However, in the conventional technology described above, relationships are determined using a rule-based method that sets rules for the areas and frequency in which the two wireless terminals being compared exist together, so there are limits to the number of relationships that can be determined and the accuracy with which they can be determined.

[0005] One aspect of the present disclosure provides a new method for estimating relationships between users.

[0006] A user relationship estimation device according to one aspect of the present disclosure includes a user relationship information acquisition unit that acquires user relationship information indicating relationships between some of a plurality of users; a user behavior information acquisition unit that acquires user behavior information indicating targets of behavior of each of the plurality of users and time periods of the behavior; a knowledge data acquisition unit that acquires knowledge data for expressing entity relationships between targets of behavior of each of the plurality of users; and an estimation unit that estimates relationships between the remaining users of the plurality of users based on the acquired user relationship information, the user behavior information, and the knowledge data.

[0007] 1 is a block diagram showing an example of a hardware configuration of a user relationship estimation device of this embodiment; FIG. 2 is a block diagram showing an example of a functional configuration of the user relationship estimation device of this embodiment; FIG. 3 is a diagram showing an example of user behavior information stored in a user behavior information storage unit of this embodiment; FIG. 4 is a diagram showing an example of a database showing the correspondence between mesh codes and area attributes; FIG. 5 is a diagram showing an example of data showing the correspondence between facility information and industry; FIG. 6 is a diagram showing an example of data showing the correspondence between event information and event type; FIG. 7 is a diagram showing an example of a graph constructed as a GNN model by the estimation unit of this embodiment; FIG. 8 is a diagram showing an example of a graph constructed as a GNN model by the estimation unit of this embodiment; FIG. 9 is a diagram showing an example of a GNN model learning method by the estimation unit of this embodiment; FIG. 10 is a diagram showing an example of an estimation method using a GNN model by the estimation unit of this embodiment; FIG. 11 is a flowchart showing an example of a user relationship estimation process performed by the user relationship estimation device of this embodiment;

[0008] Hereinafter, an embodiment of the present disclosure (hereinafter simply referred to as "the present embodiment") will be described in detail with reference to the drawings. Note that the present disclosure is not limited to the following embodiment.

[0009] The user relationship estimation device of this embodiment estimates relationships between users. Examples of relationships between users include, but are not limited to, relationships between family members, parents and children, siblings, friends, and lovers. Below, the user relationship estimation device of this embodiment will be described using an example of estimating relationships between users who own wireless communication terminals such as smartphones that enable wireless communication by signing a contract with a telecommunications carrier (carrier). However, the relationships between users estimated by the user relationship estimation device are not limited to this.

[0010] The user relationship estimation device of this embodiment may be realized by a single computer or may be realized as a system by multiple computers. When the user relationship estimation device of this embodiment is realized as a system, the computers may be directly connected via a communication cable or may be connected via a network such as a LAN (Local Area Network) or a WAN (Wide Area Network).

[0011] 1 is a block diagram showing an example of the hardware configuration of a user relationship estimation device 10 of this embodiment. As shown in FIG. 1, the user relationship estimation device 10 includes a control device 11, a main memory device 13, an auxiliary memory device 15, a display device 17, an input device 19, a communication device 21, and various buses 23. The control device 11, the main memory device 13, the auxiliary memory device 15, the display device 17, the input device 19, and the communication device 21 are connected via the various buses 23. As such, the user relationship estimation device 10 of this embodiment has a general hardware configuration using a typical computer.

[0012] The control device 11 controls the overall operation of the user relationship estimation device 10. The control device 11 may be, for example, at least one of a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit), but is not limited to these. There may be any number of CPUs or GPUs as long as they are one or more, and they may be single-core or multi-core.

[0013] Examples of the main storage device 13 include, but are not limited to, a read-only memory (ROM) and a random access memory (RAM). The ROM stores various programs, such as a program for controlling the user relationship estimation device 10 and a program for user relationship estimation according to this embodiment. The RAM is used as a work area when the control device 11 performs various controls based on the programs stored in the ROM.

[0014] The auxiliary storage device 15 stores the various programs described above and data for user relationship estimation in this embodiment. The various programs described above may be stored in at least one of the main storage device 13 and the auxiliary storage device 15. Examples of the auxiliary storage device 15 include, but are not limited to, at least one of existing storage devices capable of magnetic, electrical, or optical storage, such as a hard disk drive (HDD), a solid state drive (SSD), and a digital versatile disc (DVD). The auxiliary storage device 15 may be built into the user relationship estimation device 10 or may be externally attached to the user relationship estimation device 10 via an interface such as a universal serial bus (USB). The auxiliary storage device 15 may also be a network-attached storage (NAS) connected via a network such as a LAN or WAN.

[0015] The display device 17 displays various screens used when the user relationship estimation device 10 estimates a user relationship, and serves as a user interface with the user (operator). Examples of the display device 17 include, but are not limited to, various displays such as a liquid crystal display, an organic electroluminescence (EL) display, and a touch panel display. The display device 17 may be an internal display built into the user relationship estimation device 10, or an external display connected to the user relationship estimation device 10 via a display interface such as HDMI (registered trademark).

[0016] The input device 19 is used for various inputs used when the user relationship estimation device 10 estimates a user relationship, and serves as a user interface with the user (operator). Examples of the input device 19 include, but are not limited to, a keyboard, a mouse, and a touch panel. The input device 19 may be built into the user relationship estimation device 10, or may be externally attached to the user relationship estimation device 10 via an interface such as a USB.

[0017] Examples of the communication device 21 include, but are not limited to, a communication device for a wired LAN and a wireless communication device for a wireless LAN. The communication device 21 may be used to externally acquire the program or data for user relationship estimation of this embodiment, or may be used to externally output the user relationship estimated by the user relationship estimation device 10.

[0018] In addition to the above configuration, the user relationship estimation device 10 may further include hardwired circuits such as an IC (Integrated Circuit), an ASIC (Application Specific Integrated Circuit), and an FPGA (Field-Programmable Gate Array) that are specific to the user relationship estimation device 10 in order to realize the user relationship estimation function.

[0019] 2 is a block diagram showing an example of the functional configuration of the user relationship estimation device 10 according to this embodiment. As shown in FIG. 2 , the user relationship estimation device 10 includes a household information storage unit 101, a user behavior information storage unit 103, a user relationship information acquisition unit 111, a user behavior information acquisition unit 113, a knowledge data acquisition unit 115, an estimation unit 117, and an output unit 119.

[0020] The household information storage unit 101 and the user behavior information storage unit 103 can be realized by, for example, at least one of the main storage unit 13 and the auxiliary storage unit 15 described with reference to FIG.

[0021] The user relationship information acquisition unit 111, the user behavior information acquisition unit 113, the knowledge data acquisition unit 115, the estimation unit 117, and the output unit 119 can be realized by, for example, the control unit 11 and the main memory unit 13 described in FIG.

[0022] For example, the control device 11 reads the program for user relationship estimation of this embodiment stored in the main storage device 13 (ROM) or the auxiliary storage device 15, or obtained from the outside via the network from the communication device 21, and loads it into the main storage device 13 (RAM). The control device 11 executes various processes in accordance with the loaded program, thereby realizing each of the above-mentioned functional units. Here, the description has been given taking the example of realizing each of the above-mentioned functional units as software, but at least a portion of each of the above-mentioned functional units may also be realized as hardware. In this case, the functional units realized as hardware may be realized, for example, by the above-mentioned hardwired circuit. Furthermore, any of the above-mentioned functional units may be realized by a combination of software and hardware.

[0023] The household information storage unit 101 stores household information about users. The household information is, for example, information indicating a user's household that is provided to a telecommunications carrier when a user enters into a contract for use of a wireless communication terminal with a telecommunications carrier on a household basis, such as a family. According to the household information, the relationships between family members of a user who has entered into a contract for use of a wireless communication terminal on a household basis (e.g., parent-child relationships, husband-wife relationships, sibling relationships, etc.) are known, but relationships other than those between family members, such as friendships and romantic relationships, remain unknown. Furthermore, for a user who has entered into a contract for use of a wireless communication terminal alone, the relationships remain unknown.

[0024] The user behavior information storage unit 103 stores user behavior information. The user behavior information is information indicating the target of each user's behavior and the time period of the behavior. The user behavior information is, for example, information that can be acquired from a wireless communication terminal carried by the user, and includes, but is not limited to, the locations where the user stayed, the facilities visited by the user, and the events attended by the user. The targets of the behavior indicated by the user behavior information include, but are not limited to, the above-mentioned locations where the user stayed, the facilities visited, and the events attended. The time periods of the behavior indicated by the user behavior information include, but are not limited to, the time period when the user stayed at a location where the user stayed, the time period when the user visited the facilities visited, and the time period when the user participated in an event.

[0025] The user behavior information is information that can be acquired based on location information detected by a GPS (Global Positioning System) sensor or the like of the wireless communication terminal. However, the user behavior information is not limited to this, and may be information that can be acquired from, for example, purchase information such as cashless payment using the wireless communication terminal or points.

[0026] User behavior information may be collected directly by the user relationship estimation device 10 of this embodiment from the wireless communication terminals carried by each user, or may be collected from each wireless communication terminal by a collection device different from the user relationship estimation device 10 of this embodiment.

[0027] 3 is a diagram showing an example of user behavior information stored in the user behavior information storage unit 103 of this embodiment. In the example shown in Fig. 3, the user behavior information is information in which user identification information, time information, location information, mesh code, facility information, and event information are associated with each other, but is not limited to this.

[0028] In the example shown in Figure 3, for convenience of explanation, the time information is described in units of time periods of the target behavior, but this is not limited to this. For example, if user behavior information is collected from wireless communication terminals in fixed time intervals, the user behavior information at the time of collection may be stored in the user behavior information storage unit 103. Even in this case, the user behavior information stored in the user behavior information storage unit 103 can be aggregated to generate user behavior information in the format shown in Figure 3. In addition, in the example shown in Figure 3, for convenience of explanation, each record describes one of mesh code, facility information, and event information, but this is not limited to this, and two or more pieces of information may be described.

[0029] The user identification information may be any information that can identify a user who owns a wireless communication terminal. Examples of user identification information include, but are not limited to, the MAC (Media Access Control) address of the wireless communication terminal. As described above, the time information indicates the time period of the target activity, such as staying at a staying location, visiting a visiting facility, and participating in an event. The location information is information that indicates the location of the user.

[0030] The mesh code is information indicating a regional mesh with a predetermined square size defined as a longitude-latitude grid on a map, and indicates the regional mesh of the location indicated by the location information. The size of the predetermined square can be defined, for example, as any of 1st to 6th order sizes, and can be selected depending on the purpose of the model being used, the estimation accuracy, etc. In this embodiment, an example is described in which the size of the predetermined square is a 500m square of a 4th order mesh, but the present disclosure is not limited to this. The mesh code is used as a location where the user stays. Facility information is information indicating facilities visited by the user, and indicates facilities located at the location indicated by the location information. Event information is information indicating events in which the user participated, and indicates events held at the location indicated by the location information.

[0031] The user relationship information acquisition unit 111 acquires user relationship information indicating relationships between some of the users. The relationships between some of the users include at least one of known relationships among the relationships between the users and unknown relationships among the relationships between the users that are estimated based on a rule. In this embodiment, the relationships between some of the users include both the known relationships and the unknown relationships that are estimated based on a rule, but the present invention is not limited to this.

[0032] Among the relationships between multiple users, a "known relationship" refers to a relationship between users that is already defined in the system, such as a relationship between family members of users indicated by the household information stored in the household information storage unit 101 described above. For example, if the relationship between user A and user B corresponds to any of the relationships defined in the system, such as parent-child, lovers, friends, colleagues, or acquaintances, the relationship between users A and B corresponds to a "known relationship." In this embodiment, the user relationship information acquisition unit 111 acquires household information from the household information storage unit 101, thereby acquiring user relationship information that indicates known relationships among the relationships between multiple users as relationships between some of the multiple users.

[0033] Among the relationships between multiple users, an "unknown relationship" refers to a relationship between users that does not fall under the category of a "known relationship" and has not yet been defined in the system. For example, if the relationship between user A and user C does not fall under any of the relationships defined in the system, the relationship between users A and C falls under the category of an "unknown relationship." Examples of rule-based estimation of unknown relationships among relationships between multiple users include, but are not limited to, rule-based estimation of unknown relationships between users from the user behavior information stored in the user behavior information storage unit 103 described above. Note that rule-based estimation of unknown relationships between users from user behavior information may be performed by the user relationship information acquisition unit 111 (the user relationship estimation device 10 of this embodiment) or by an estimation device different from the user relationship estimation device 10.

[0034] For example, in the case of the user behavior information shown in FIG. 3 , the locations where User A and User C stayed during the late-night period of "2024 / 4 / 5 22:00 to 2024 / 4 / 6 7:00" are the same, with mesh code "M001." Also, in the case of the user behavior information shown in FIG. 3 , the facilities visited by User A and User C during the daytime period of "2024 / 4 / 6 16:00 to 2024 / 4 / 6 17:00" are the same, with facility information "FG001." Note that facility information "FG001" may indicate, for example, Grocery Store A. In this way, since User A and User C stayed in the same location during the late-night period and visited Grocery Store A together during the daytime on a holiday, it may be rule-based to infer that they are, for example, family members.

[0035] 3, for example, the events attended by users C and E during the daytime period of "2024 / 4 / 7 10:00 to 2024 / 4 / 7 17:00" are the same in event information "I001." Also, for example, in the user behavior information shown in FIG. 3, the events attended by users C and E during the daytime period of "2024 / 4 / 14 10:00 to 2024 / 4 / 14 14:00" are the same in event information "I002." In this way, since users C and E have attended multiple identical events during the daytime, it may be rule-based to infer that they have a friendship relationship, for example.

[0036] However, these estimations are merely examples, and in order to further improve estimation accuracy, it is preferable to set rules that refer to other user actions or the frequency of identical user actions. Knowledge data, which will be described later, may also be used for rule-based estimation. Note that, since rule-based estimation of relationships between users has limitations in number and accuracy, in this embodiment, estimation accuracy is guaranteed, but is not limited to this.

[0037] In this embodiment, the user relationship information acquiring unit 111 acquires the above-described rule-based estimation result using the user behavior information stored in the user behavior information storage unit 103. As a result, the user relationship information acquiring unit 111 acquires user relationship information indicating the rule-based estimation of unknown relationships among the relationships among the plurality of users as relationships among some of the plurality of users.

[0038] The user behavior information acquisition unit 113 acquires user behavior information indicating the target of each of the actions of multiple users and the time period of the action. Specifically, the user behavior information acquisition unit 113 acquires the user behavior information from the user behavior information storage unit 103.

[0039] The knowledge data acquisition unit 115 acquires knowledge data for expressing entity relationships between targets of the actions of each of the users. For example, the knowledge data acquisition unit 115 acquires knowledge data for expressing entity relationships between targets of the actions of each of the users and the staying places, visited facilities, and attended events.

[0040] A knowledge graph is a representation of the relationship between entities using a triplet of <subject, predicate, object>. The knowledge graph for the above-mentioned location of stay can be expressed, for example, as follows: "The attribute (predicate) of mesh code "M001" (subject) is a residential area (object)." The knowledge graph for the above-mentioned visited facility can be expressed, for example, as follows: "The industry (predicate) of facility information "FG001" (subject) is a grocery store (object)." The knowledge graph for the above-mentioned event participation can be expressed, for example, as follows: "The type (predicate) of event information "I002" (subject) is sports (object)."

[0041] For this reason, the knowledge data acquisition unit 115 of this embodiment acquires data corresponding to objects as knowledge data for each of the above-mentioned staying places, visited facilities, and participated events. Note that since the relationship between entities is a static relationship rather than a dynamic relationship, the knowledge data acquisition unit 115 can acquire the knowledge data via, for example, the Internet 2.

[0042] For example, in the case of mesh codes, a database showing the correspondence between mesh codes and the attributes (uses) of the areas indicated by the mesh codes, as shown in Fig. 4, is made publicly available on the Internet 2. Therefore, the knowledge data acquisition unit 115 can acquire knowledge data (area attributes) for expressing the relationship between entities and mesh codes (stay locations) by referring to the database.

[0043] For example, in the case of facility information, data showing the correspondence between facility information (facility name) and the business type of the facility, as shown in Fig. 5, is published on the Internet 2. Note that the data does not need to be compiled in a database format as shown in Fig. 5, and the correspondence may be published in a dispersed manner on the Internet 2. Therefore, by referring to such data, the knowledge data acquisition unit 115 can acquire knowledge data (business type) for expressing the entity relationship with facility information (facility name).

[0044] For example, in the case of event information, data showing the correspondence between the event information (event name) and the type of the event, as shown in Fig. 6, is published on the Internet 2. Note that the data does not need to be compiled in a database format as shown in Fig. 6, and the correspondence may be published in a dispersed manner on the Internet 2. Therefore, by referring to such data, the knowledge data acquisition unit 115 can acquire knowledge data (event type) for expressing the relationship between the entity and the event information (event name).

[0045] The estimation unit 117 estimates the relationships between the remaining users among the multiple users based on the user relationship information acquired by the user relationship information acquisition unit 111, the user behavior information acquired by the user behavior information acquisition unit 113, and the knowledge data acquired by the knowledge data acquisition unit 115. Specifically, the estimation unit 117 estimates the relationships between the remaining users by learning the user behavior information and the knowledge data linked to the user behavior information, using the user relationship information as correct answer data.

[0046] For example, the estimation unit 117 estimates relationships between the remaining users based on a graph in which nodes represent knowledge data for expressing each of the multiple users, relationships between some of the users, targets of the actions of each of the multiple users, and entity relationships between the targets of the actions of each of the multiple users, and edges represent time periods of the actions of each of the multiple users. For example, the estimation unit 117 performs machine learning on a GNN (Graph Neural Networks) model, which is the graph, and estimates relationships between the remaining users based on the machine-learned GNN model. For example, the estimation unit 117 estimates relationships between the remaining users by link prediction or edge classification using the machine-learned GNN model. Note that estimating relationships between the remaining users specifically means estimating unknown relationships between users. As described above, relationships between users to be estimated include, but are not limited to, relationships between family members, parents and children, siblings, friends, and lovers.

[0047] 7 to 9 are diagrams showing examples of graphs constructed as a GNN model by the estimation unit 117 of this embodiment. For convenience of explanation, the graphs shown in Fig. 7 to 9 are shown as separate graphs, but in reality, they are constructed on a single graph, and after machine learning, it is assumed that unknown relationships between users are compositely estimated.

[0048] 7 shows a portion of the graph constructed by the estimation unit 117, specifically, a portion of the graph where the target of user behavior is a location (mesh code). In the graph shown in FIG. 7, knowledge data for expressing relationships between users, each of the multiple users, the target of each of the multiple users' behavior (mesh code), and the entity relationships between each of the multiple users and the target of each of the multiple users' behavior are represented as nodes. In addition, in the graph shown in FIG. 7, the time periods of each of the multiple users' behavior are represented as edge labels.

[0049] Note that each node is represented as a multidimensional feature vector, but is not limited to this. Furthermore, the feature vector may be any type that expresses the feature of each node. Since each node is represented as a multidimensional feature vector, the graph constructed as a GNN model including the graph shown in FIG. 7 is constructed in a multidimensional feature space. Note that the arrangement of the nodes in the graph shown in FIG. 7 is shown schematically and does not represent the actual arrangement in the feature space.

[0050] Regarding the relationships between users, "family," "siblings," "friends," "parents and children," and "lovers" are represented as nodes. Regarding users, "User A" to "User F" are represented as nodes. Regarding mesh codes, "M001," "M003" to "M005" are represented as nodes. Regarding knowledge data, "residential areas" and "downtown areas" are represented as nodes.

[0051] 7, edges connecting nodes are represented by solid and dotted lines, indicating that data for nodes connected by solid edges can be obtained, and that dotted edges are estimated (link prediction) by machine learning, which will be described later. In other words, before machine learning, dotted edges do not actually connect nodes.

[0052] For example, data on nodes of user relationships connected to user nodes by solid edges is obtained from user relationship information, such as the fact that user A and user C are family members, or that user C and user E are friends. Furthermore, for example, the fact that user D and user E are parent-child is obtained from user relationship information (household information). Furthermore, for example, data on mesh code nodes connected to user nodes by solid edges is obtained from user behavior information. Furthermore, for example, data on knowledge data nodes connected to mesh code nodes by solid edges is obtained from knowledge data.

[0053] Furthermore, the label of the edge connecting the user node and the mesh code node indicates the time period during which the user stayed at the mesh code stay location. For example, the label of the edge connecting the user node "User C" and the mesh code node "M001" indicates "late night", and the label of the edge connecting the user node "User E" and the mesh code node "M003" indicates "daytime". The data on the label (time period) of the edge connecting the user node and the mesh code node is obtained from user behavior information.

[0054] In the graph shown in FIG. 7, being connected to a node of the same mesh code by an edge means that the places where users stayed in the past are close to each other. Therefore, as in the graph shown in FIG. 7, it is expected that some kind of close relationship will be inferred between users of nodes connected to a node of the same mesh code by an edge, as in the graph shown in FIG. 7, by inference after machine learning, which will be described later. Furthermore, in the graph shown in FIG. 7, knowledge data nodes are associated with mesh code nodes, and time period information is assigned to the labels of the edges of the mesh code nodes. Therefore, it is expected that some kind of close relationship will be inferred between users of nodes connected to a node of the same mesh code by an edge, as well as that the relationship will be classified in more detail by inference after machine learning, which will be described later.

[0055] Fig. 8 shows a portion of the graph constructed by the estimation unit 117, specifically a portion of the graph where the target of the user's behavior is the visited facilities. Fig. 9 shows a portion of the graph constructed by the estimation unit 117, specifically a portion of the graph where the target of the user's behavior is the attended events. Note that the graphs shown in Figs. 8 and 9 are similar to the graph shown in Fig. 7 except that the nodes that are the target of the user's behavior are represented as nodes for visited facilities and nodes for attended events, respectively, and therefore detailed description thereof will be omitted.

[0056] In the graph shown in FIG. 8 , being connected by an edge to the same visited facility node means that the facilities visited in the past are close. Therefore, as in the graph shown in FIG. 8 , it is expected that some kind of close relationship will be inferred between users of nodes connected by an edge to the same visited facility node through inference after machine learning, which will be described later. Furthermore, in the graph shown in FIG. 8 , nodes of knowledge data are associated with the visited facility node, and time zone information is assigned to the label of the edge of the visited facility node. Therefore, it is expected that some kind of close relationship will be inferred between users of nodes connected by an edge to the same visited facility node through inference after machine learning, which will be described later, and that the relationship will be classified in more detail.

[0057] Furthermore, in the graph shown in FIG. 9 , being connected by an edge to a node of the same attended event means that the events attended in the past are close. Therefore, as in the graph shown in FIG. 9 , it is expected that some kind of close relationship will be inferred between users of nodes connected by an edge to the node of the same attended event through inference after machine learning, which will be described later. Furthermore, in the graph shown in FIG. 9 , nodes of knowledge data are associated with the attended event nodes, and time period information is assigned to the labels of the edges of the attended event nodes. Therefore, it is expected that some kind of close relationship will be inferred between users of nodes connected by an edge to the node of the same attended event through inference after machine learning, which will be described later, and that the relationship will be classified in more detail.

[0058] The estimation unit 117, as a GNN encoder, learns the GNN model, which is the graph described above, and updates each node to its optimal coordinates (new features) in the feature space. For example, the estimation unit 117 acquires information (e.g., features) of one or more nodes connected by edges for each node, and updates the node's coordinates (features) by summing or averaging the acquired information. The estimation unit 117 repeats this process until the loss function converges, thereby updating each node to its optimal coordinates (new features) in the feature space. As a result, nodes with a close relationship become closer to each other in the feature space and are placed closer to each other (positive example). On the other hand, nodes with a distant relationship become farther apart in the feature space and are placed farther apart from each other (negative example). Note that the distance in the feature space may be measured in any manner. For example, it may be Euclidean distance or Mahalanobis distance, which takes data variance into account. However, the GNN model learning method used by the estimation unit 117 is not limited to this.

[0059] Fig. 10 is a diagram showing an example of a learning method of a GNN model by the estimation unit 117 of this embodiment, and shows the arrangement of each node in the feature space of a part of the graph shown in Fig. 7. The example shown in Fig. 10 shows that, as a result of learning of the GNN model by the estimation unit 117, the position (feature) of user B's node in the feature space has been updated from the position of the node indicated by the solid line to the position of the node indicated by the dotted line (new feature). As a result, user B's node is arranged close to user A's node, user C's node, and family nodes, and has a close relationship with these nodes.

[0060] The estimation unit 117 uses the trained GNN model to estimate unknown relationships between user nodes. For example, in the example shown in Fig. 10, the node of user B is close to both the node of user A and the node of user C, so the estimation unit 117 estimates that user B has a close relationship with both user A and user C. In addition, in the example shown in Fig. 10, the nodes of user A and user C are connected to family nodes, and the node of user B is also close to the family nodes. Therefore, the estimation unit 117 estimates that user B also has a family relationship with user A and user C, and connects the node of user B to the family nodes with edges.

[0061] 11 is a diagram showing an example of an estimation method using a GNN model by the estimation unit 117 of this embodiment. As described above, the estimation unit 117 estimates that user B is also in a family relationship with users A and C, and therefore connects the node of user B with the family nodes with edges. However, since the edge connecting the node of user B with the family nodes is estimated by machine learning, it is shown by a dashed line in FIG. 11.

[0062] In this embodiment, the relationships between users are also represented by nodes, but the relationships between users may be represented as labels of edges connecting the nodes between users. In this case, the estimation unit 117, as a GNN decoder, uses the trained GNN model to calculate the distances between nodes between users and perform link prediction, and calculates labels of edges connecting the nodes in the link prediction and performs edge classification, thereby specifying unknown relationships between users.

[0063] For example, in FIG. 10 , assume that the node for user A and the node for user C are connected by an edge, the label of the edge is "family," and the position of the node for user B in the feature space has been updated to the position of the dotted line node. In this case, the node for user B is close to both the node for user A and the node for user C. Therefore, the estimation unit 117 estimates that user B has a close relationship with both user A and user C, and connects the node for user A and the node for user C to the node for user B with edges (link prediction). Furthermore, the label of the edge connecting the node for user A and the node for user C is "family." Therefore, the estimation unit 117 estimates that user B is also in a family relationship with user A and user C, and labels both the edge connecting the node for user A and the node for user B and the edge connecting the node for user B and the node for user C as "family" (edge ​​classification).

[0064] The output unit 119 outputs the relationships between the plurality of users including the relationships between the remaining users estimated by the estimation unit 117 .

[0065] FIG. 12 is a flowchart showing an example of a user relationship estimation process performed by the user relationship estimation device 10 of this embodiment.

[0066] First, the user relationship information acquisition unit 111 acquires user relationship information indicating relationships between some of a plurality of users (step S101).

[0067] Next, the user behavior information acquisition unit 113 acquires user behavior information indicating the target of each of the behaviors of the multiple users and the time period during which the behavior occurred (step S103).

[0068] Next, the knowledge data acquisition unit 115 acquires knowledge data for expressing the relationship between entities and targets of the actions of each of the multiple users (step S105).

[0069] Next, the estimation unit 117 uses the user relationship information, user behavior information, and knowledge data to construct a GNN model, which is a graph in which nodes are knowledge data for expressing each of the multiple users, the relationships between some of the users, the targets of each of the multiple users' behaviors, and the entity relationships between the targets of each of the multiple users' behaviors, and edges are the time periods of each of the multiple users' behaviors (step S107).

[0070] Next, the estimation unit 117 learns the constructed GNN model and optimizes the arrangement of each node constituting the GNN model (step S109).

[0071] Next, the estimation unit 117 estimates unknown relationships between users using the trained GNN model (step S111).

[0072] As described above, in this embodiment, knowledge data is linked to the target of behavior indicated by user behavior information and treated as a knowledge graph, and the time period of the target of behavior indicated by the user behavior information is also used to construct a GNN model and estimate unknown relationships between users, so that even a huge number of unknown relationships between users can be accurately estimated.

[0073] Furthermore, in this embodiment, not only are known relationships between users (household information) used as correct answer data for the GNN model, but rule-based estimates of unknown relationships between users are also used. This allows the quantity of correct answer data to be increased while ensuring the quality of the data, thereby improving the accuracy of estimation of unknown relationships between users.

[0074] As described above, by using the relationships between multiple users, including unknown relationships between users, estimated by this embodiment in, for example, advertising businesses, a greater advertising effect than before can be expected. In general advertising businesses, while there is an abundance of data (such as browsing history and purchase history) representing the relationships between users and content, data representing the relationships between users is often scarce. However, according to this embodiment, data representing the relationships between users can be estimated, which makes it possible to implement targeting measures that anticipate ripple effects on family and friends, for example, and a greater advertising effect than before can be expected.

[0075] (Program) The program executed by the user relationship estimation device 10 of the above embodiment is provided by being stored in a computer-readable storage medium such as a CD-ROM, CD-R, memory card, DVD, or flexible disk (FD) in the form of a file in an installable or executable format.

[0076] The program executed by the user relationship estimation device 10 of the above embodiment may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. The program executed by the user relationship estimation device 10 of the above embodiment may be provided or distributed via a network such as the Internet. The program executed by the user relationship estimation device 10 of the above embodiment may be provided by being pre-installed in a ROM or the like.

[0077] The program executed by the user relationship estimation device 10 of the above embodiment has a modular configuration for implementing the above-mentioned units on a computer. In terms of actual hardware, for example, the CPU reads the learning program from the HDD onto the RAM and executes it, thereby implementing the above-mentioned units on the computer.

[0078] As described above, according to the above embodiment, it is possible to accurately estimate a huge number of unknown relationships between users.

[0079] The above-described embodiment and each of the modifications merely illustrate examples of implementations of the present disclosure, and the technical scope of the present disclosure should not be construed as being limited by these. Therefore, the present disclosure can be implemented in various forms without departing from the spirit or main features thereof. For example, the above-described embodiment and each of the modifications may be appropriately combined in their respective constituent units. Furthermore, for example, some components may be deleted from all components in the above-described embodiment and each of the modifications.

[0080] The present disclosure includes the following aspects.

[0081] (1) A user relationship estimation device comprising: a user relationship information acquisition unit that acquires user relationship information indicating relationships between some of a plurality of users; a user behavior information acquisition unit that acquires user behavior information indicating targets of behavior of each of the plurality of users and time periods of the behavior; a knowledge data acquisition unit that acquires knowledge data for expressing entity relationships between targets of behavior of each of the plurality of users; and an estimation unit that estimates relationships between remaining users of the plurality of users based on the acquired user relationship information, user behavior information, and knowledge data.

[0082] (2) The user relationship estimation device according to (1), wherein the relationships between the some users include at least unknown relationships among relationships between a plurality of users that are estimated based on a rule.

[0083] (3) The user relationship estimation device according to (2), wherein the relationships between the some users include at least known relationships among relationships between a plurality of users.

[0084] (4) The user relationship estimation device according to (1), wherein the estimation unit estimates the relationships between the remaining users based on a graph in which nodes are knowledge data representing each of the plurality of users, relationships between some of the users, targets of the actions of each of the plurality of users, and entity relationships between targets of the actions of each of the plurality of users, and edges are time periods of the actions of each of the plurality of users.

[0085] (5) The user relationship estimation device according to (2), wherein the graph is a GNN (Graph Neural Networks) model, and the estimation unit performs machine learning on the GNN model and estimates the relationships between the remaining users based on the machine-learned GNN model.

[0086] (6) A user relationship estimation method including: a user relationship information acquisition step in which a user relationship information acquisition unit acquires user relationship information indicating relationships between some of a plurality of users; a user behavior information acquisition step in which a user behavior information acquisition unit acquires user behavior information indicating targets of behavior of each of the plurality of users and time periods of the behavior; a knowledge data acquisition step in which a knowledge data acquisition unit acquires knowledge data for expressing entity relationships between targets of behavior of each of the plurality of users; and an estimation step in which an estimation unit estimates relationships between remaining users of the plurality of users based on the acquired user relationship information, the user behavior information, and the knowledge data.

[0087] (7) A program for causing a computer to execute the following steps: a user relationship information acquisition step for acquiring user relationship information indicating relationships between some of a plurality of users; a user behavior information acquisition step for acquiring user behavior information indicating targets of the behavior of each of the plurality of users and time periods of the behavior; a knowledge data acquisition step for acquiring knowledge data for expressing entity relationships between targets of the behavior of each of the plurality of users; and an estimation step for estimating relationships between the remaining users of the plurality of users based on the acquired user relationship information, user behavior information, and knowledge data.

[0088] REFERENCE SIGNS LIST 10 User relationship estimation device 11 Control device 13 Main memory device 15 Auxiliary memory device 17 Display device 19 Input device 21 Communication device 23 Various buses 101 Household information storage unit 103 User behavior information storage unit 111 User relationship information acquisition unit 113 User behavior information acquisition unit 115 Knowledge data acquisition unit 117 Estimation unit 119 Output unit

Claims

1. A user relationship estimation device comprising: a user relationship information acquisition unit that acquires user relationship information indicating relationships between some of a plurality of users; a user behavior information acquisition unit that acquires user behavior information indicating targets of behavior of each of the plurality of users and time periods of the behavior; a knowledge data acquisition unit that acquires knowledge data for expressing entity relationships between targets of behavior of each of the plurality of users; and an estimation unit that estimates relationships between the remaining users of the plurality of users based on the acquired user relationship information, user behavior information, and knowledge data.

2. The user relationship estimation device according to claim 1, wherein the relationships between the selected users include at least unknown relationships among relationships between a plurality of users that are estimated based on a rule.

3. The user relationship estimation device according to claim 2, wherein the relationships between the selected users include at least known relationships among relationships between a plurality of users.

4. The user relationship estimation device according to claim 1, wherein the estimation unit estimates the relationships between the remaining users based on a graph in which nodes are knowledge data representing each of the plurality of users, relationships between some of the users, targets of the actions of each of the plurality of users, and entity relationships between targets of the actions of each of the plurality of users, and edges are time periods of the actions of each of the plurality of users.

5. The user relationship estimation device according to claim 2, wherein the graph is a GNN (Graph Neural Networks) model, and the estimation unit performs machine learning on the GNN model and estimates the relationships between the remaining users based on the machine-learned GNN model.

6. A user relationship estimation method including: a user relationship information acquisition step in which a user relationship information acquisition unit acquires user relationship information indicating relationships between some of a plurality of users; a user behavior information acquisition step in which a user behavior information acquisition unit acquires user behavior information indicating targets of behavior of each of the plurality of users and time periods of the behavior; a knowledge data acquisition step in which a knowledge data acquisition unit acquires knowledge data for expressing entity relationships between targets of behavior of each of the plurality of users; and an estimation step in which an estimation unit estimates relationships between the remaining users of the plurality of users based on the acquired user relationship information, user behavior information, and knowledge data.

7. A program for causing a computer to execute the following steps: a user relationship information acquisition step for acquiring user relationship information indicating relationships between some of a plurality of users; a user behavior information acquisition step for acquiring user behavior information indicating the target of each of the plurality of users' behavior and the time period of said behavior; a knowledge data acquisition step for acquiring knowledge data for expressing entity relationships between the target of each of the plurality of users' behavior; and an estimation step for estimating relationships between the remaining users of the plurality of users based on the acquired user relationship information, user behavior information, and knowledge data.

Images