Information processing device, information processing method, and program

Abstract

To provide a technique for balancing knowledge alignment and link prediction.SOLUTION: A graph relation acquisition unit 36 acquires inter-node relation information indicating relations for each of a plurality of viewpoints between a node configuring one of two different knowledge graphs and a node configuring the other knowledge graph. A positive example selection unit 33 selects the node included in equivalence relation information as a positive example. A vector updating unit 37 updates a vector based on a vector assigned to the positive example and an evaluation function regarding the inter-node relation information.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and a program, and in particular to a technique for expressing a knowledge graph using vectors. [Background technology]

[0002] Graphs that are configured by connecting different nodes with edges are known. Among graphs, knowledge graphs are graphs in which entities are assigned to each node and relationships between the nodes are assigned to edges. A technology is known that expresses entities and relationships based on one knowledge graph as low-dimensional vectors and evaluates the accuracy of unknown facts related to the graph based on vector operations to complement facts (Non-Patent Document 1).

[0003] Also, a technique is known in which multiple knowledge graphs are used to input equivalence relations between some entities and to complement facts related to one knowledge graph from another knowledge graph (Non-Patent Document 2). Furthermore, a technique is also known in which multiple knowledge graphs are used to input equivalence relations between some entities and to derive equivalence or similarity relations between other entities (Non-Patent Document 3). [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Bordes, A., Usunier, N., Garcia-Duran, A., Weston, J., and Yakhnenko, O, “Translating embeddings for modeling multi-relational data,” In Advances in neural information processing systems, 2013. [Non-Patent Document 2] Zhu, H., Xie, R., Liu, Z., and Sun, M., “Iterative Entity Alignment via Joint Knowledge Embeddings,” In Proceedings of the 26th International Joint Conference on Artificial Intelligence (IJCAI), 2017. [Non-Patent Document 3] Sun, Z., Hu, W., Zhang, Q., and Qu, Y., “Bootstrapping Entity Alignment with Knowledge Graph Embedding,” In Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI), 2018. Summary of the Invention [Problem to be solved by the invention]

[0005] Assigning vectors to entities and relations that make up a knowledge graph is equivalent to embedding these relations in a space equivalent to the dimension of the vector. The above techniques improve the accuracy of knowledge alignment by complementing the knowledge contained in each knowledge graph, but they distort the embedding space, and the accuracy of link prediction may not necessarily improve.

[0006] The present invention has been made in consideration of these points, and has an object to provide a technique for balancing knowledge alignment and link prediction. [Means for solving the problem]

[0007] A first aspect of the present invention is an information processing device. The device includes: a knowledge graph acquisition unit that acquires a knowledge graph, which is a graph formed by connecting different nodes with edges, and in which an entity is assigned to each node and a relationship between each node is assigned to an edge; a vector assignment unit that assigns a vector to each node and each edge constituting each of two different knowledge graphs acquired by the knowledge graph acquisition unit; a set extraction unit that extracts, for each of the two different knowledge graphs acquired by the knowledge graph acquisition unit, a set composed of three elements, a first node, an edge starting from the first node and ending at a second node, and the second node; a positive example selection unit that selects each of the multiple sets extracted by the set extraction unit as one positive example; a graph relationship acquisition unit that acquires inter-node relationship information indicating relationships from multiple perspectives between nodes constituting one knowledge graph acquired by the knowledge graph acquisition unit and nodes constituting the other knowledge graph; and a vector update unit that updates the vector assigned by the vector assignment unit based on the vector assigned to the positive example and an evaluation function related to the inter-node relationship information.

[0008] The evaluation function may be set as the norm of a difference vector between a vector assigned to the second node and a vector obtained by adding a vector assigned to the edge to a vector assigned to the first node for the pair that is a positive example selected by the positive example selection unit.

[0009] The evaluation function may be set to become smaller as the similarity between a distribution of points corresponding to vectors assigned to nodes included in one knowledge graph and a distribution of points corresponding to vectors assigned to nodes included in the other knowledge graph increases for nodes appearing in each of the multiple perspectives included in the inter-node relationship information, and the vector update unit may update the vectors assigned by the vector assignment unit so as to reduce the evaluation function.

[0010] The vector update unit may update the vector by iterative processing based on a gradient of an evaluation function derived from the pair, and after the update by the evaluation function derived from the pair converges, the vector may be updated by iterative processing based on the gradient of the evaluation function derived from the inter-node relationship information.

[0011] A second aspect of the present invention is an information processing method, in which a processor executes the steps of: acquiring a knowledge graph, which is a graph formed by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; assigning vectors to each node and each edge constituting each of the two acquired different knowledge graphs; extracting a set of three elements, a first node, an edge starting from the first node and ending at a second node, and the second node, for each of the two acquired different knowledge graphs; selecting each of the extracted sets as one positive example; acquiring node-to-node relationship information indicating relationships between nodes constituting one of the acquired knowledge graphs and nodes constituting the other knowledge graph from multiple viewpoints; and updating the assigned vectors based on an evaluation function related to the vectors assigned to the positive examples and the node-to-node relationship information.

[0012] A third aspect of the present invention is a program, which causes a computer to realize the following functions: acquiring a knowledge graph, which is a graph formed by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; assigning vectors to each node and each edge constituting each of the two acquired different knowledge graphs; extracting, for each of the two acquired different knowledge graphs, a set consisting of three elements: a first node, an edge starting from the first node and ending at a second node, and the second node; selecting each of the extracted sets as one positive example; acquiring inter-node relationship information indicating relationships between nodes constituting one of the acquired knowledge graphs and nodes constituting the other knowledge graph from multiple viewpoints; and updating the assigned vectors based on an evaluation function related to the vectors assigned to the positive examples and the inter-node relationship information.

[0013] In order to provide this program or to update a part of the program, a computer-readable recording medium having this program recorded thereon may be provided, or this program may be transmitted over a communication line.

[0014] Any combination of the above components, and any transformation of the present invention into a method, device, system, computer program, data structure, recording medium, etc., are also effective as aspects of the present invention. Effect of the Invention

[0015] According to the present invention, a technique for balancing knowledge alignment and link prediction can be provided. [Brief description of the drawings]

[0016] [Figure 1] 1 is a diagram illustrating an example of a knowledge graph to be processed by an information processing device according to an embodiment; [Diagram 2]FIG. 1 illustrates, in tabular form, entities in a knowledge graph and relationships between entities in the knowledge graph. [Diagram 3] FIG. 2 is a diagram illustrating a functional configuration of an information processing device according to an embodiment. [Figure 4] FIG. 4 is a diagram for explaining an evaluation function according to the embodiment. [Diagram 5] 4 is a flowchart illustrating a flow of information processing executed by an information processing device according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] <Outline of the embodiment> Hereinafter, an overview of an information processing device according to an embodiment will be described with reference to FIGS.

[0018] 1(a)-(c) are diagrams for explaining an overview of the embodiment, and are diagrams that diagrammatically show an example of a knowledge graph to be processed by an information processing device according to the embodiment. Specifically, FIG. 1(a) is a diagram showing the structure of knowledge graph R, the structure of knowledge graph S different from knowledge graph R, and the relationship shown by foresight information between the entities of knowledge graph R and the entities of knowledge graph S, FIG. 1(b) is a diagram showing the nodes and relationships of knowledge graph R in a table format, and FIG. 1(c) is a diagram showing the nodes and relationships of knowledge graph S in a table format.

[0019] In FIG. 1(a), node e 1 , Node A, Node B, Node C, Node f, and Node X represent entities in the knowledge graph R. Also, the arrows connecting different nodes in the knowledge graph R represent relationships between entities in the knowledge graph R. As shown in Fig. 1(b), node e 1 indicates a TV program (Program 1), and nodes A, B, and C indicate the cast of the program (Actor A, Actor B, and Actor C). Node f indicates that the program belongs to the category of movies, and node X indicates the director X who produced the movie. Therefore, in Fig. 1(a), node e 1The edges connecting from node A to node B and node C indicate the relationship of “appearance”, and the edges connecting from node e to node 1 The edge connecting from node X to node f indicates a "category" relationship. 1 The edges connecting towards indicate a "production" relationship.

[0020] Similarly, in FIG. 1(a), node e 2 , node α, node β, node γ, node D, and node f represent entities in the knowledge graph S. As shown in FIG. 1(c), node e 2 1(a), node e indicates a certain TV program (Program 2), and nodes α, β, γ, and D indicate the actors of the program (Actor α, Actor β, Actor γ, and Actor D). Node f indicates that the program belongs to the movie category. Therefore, in Fig. 1(a), node e indicates that the program belongs to the movie category. 2 The edges connecting from node α to node β to node γ and node D indicate the relationship of “appearance”, and the edges connecting from node e to node 2 The edge connecting from to node f indicates a “category” relationship.

[0021] Although knowledge graph R and knowledge graph S are different knowledge graphs, the relationships between entities in knowledge graph R and entities in knowledge graph S from several different perspectives are obtained as prior information, which are shown in Figure 1 by dashed lines, dashed lines, and double-dashed lines, respectively.

[0022] FIG. 2 is a diagram showing in table form the relationship between the entities in the knowledge graph R shown in FIG. 1(a) and the entities in the knowledge graph S. FIG. 2 shows an example in which relationships in three different perspectives, from perspective 1 to perspective 3, between the entities in the knowledge graph R and the entities in the knowledge graph S are obtained as foresight information. Specifically, perspective 1 is foresight information regarding the attribute data of actors, and actor A and actor α are related in that they are both males of Japanese nationality. Furthermore, perspective 2 is foresight information regarding the co-starring relationships between actors, and actor A and actor β, actor B and actor γ, and actor C and actor α are related in that they have co-starred together. Furthermore, perspective 3 is foresight information regarding the equivalence relationship between actors (the relationship that the actors are the same person), and indicates that actor A who appeared in program 1 and actor α who appeared in program 2 are the same person, and actor B who appeared in program 1 and actor D who appeared in program 2 are the same person.

[0023] As a variation of viewpoints other than those described above, there is an external knowledge graph E (not shown), and an equivalence relationship between entities is defined between knowledge graph E and knowledge graph R, and between knowledge graph E and knowledge graph S, and further, it may be considered as foresight information of the relationship that the entities of knowledge graph R and the entities of knowledge graph S are connected via a chain of relationships on knowledge graph E. In this case, each chain pattern of relationships on knowledge graph E may be considered as a separate viewpoint.

[0024] The information processing device according to the embodiment first assigns an initial vector to each node and each edge constituting the knowledge graph R. Similarly, the information processing device according to the embodiment also assigns an initial vector to each node and each edge constituting the knowledge graph S. As a non-limiting example, the dimension of the vector that the information processing device assigns to each knowledge graph is about 200 dimensions.

[0025] Next, the information processing device according to the embodiment sets a value function related to the connection relationship between the nodes included in each knowledge graph and the similarity between the distributions of the vectors for each viewpoint to which the foresight information is given. The details will be described later, but the evaluation function set by the information processing device according to the embodiment includes a term such that when a certain node H and a certain node T are connected by an edge R, the sum of the vector assigned to the node H and the vector assigned to the edge R becomes smaller as the sum approaches the vector assigned to the node T. This corresponds to an evaluation item related to link prediction.

[0026] Furthermore, any vector can be represented as a point in a space of the same dimension as the elements of the vector. The information processing device according to the embodiment obtains, for each knowledge graph, a distribution of points corresponding to vectors assigned to a node to which a certain viewpoint is given. The evaluation function set by the information processing device according to the embodiment includes a term that becomes smaller as the similarity between these distributions increases (i.e., the more similar the distributions are). By making the distributions similar, it is expected that knowledge will be more easily shared between knowledge graphs. Therefore, this corresponds to an evaluation item related to knowledge alignment.

[0027] The information processing device according to the embodiment can evaluate knowledge alignment and link prediction using the same evaluation function, and therefore can achieve a balance between knowledge alignment and link prediction by optimizing the evaluation function.

[0028] <Functional configuration of information processing device 1 according to the embodiment> FIG. 3 is a diagram showing a schematic functional configuration of an information processing device 1 according to an embodiment. The information processing device 1 includes a storage unit 2 and a control unit 3. In FIG. 3, arrows indicate main data flows, and there may be data flows not shown in FIG. 3. In FIG. 3, each functional block indicates a functional unit configuration, not a hardware (device) unit configuration. Therefore, the functional blocks shown in FIG. 3 may be implemented in a single device, or may be implemented separately in multiple devices. Data may be exchanged between functional blocks via any means, such as a data bus, a network, or a portable storage medium.

[0029] The memory unit 2 is a large-capacity storage device such as a ROM (Read Only Memory) that stores the BIOS (Basic Input Output System) of the computer that realizes the information processing device 1, a RAM (Random Access Memory) that serves as the working area of ​​the information processing device 1, an HDD (Hard Disk Drive) or an SSD (Solid State Drive) that stores the OS (Operating System), application programs, and various information referenced when the application programs are executed.

[0030] The control unit 3 is a processor such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) of the information processing device 1, and functions as a knowledge graph acquisition unit 30, a vector assignment unit 31, a set extraction unit 32, a positive example selection unit 33, a graph relationship acquisition unit 36, and a vector update unit 37 by executing programs stored in the memory unit 2.

[0031] 3 shows an example in which the information processing device 1 is configured as a single device. However, the information processing device 1 may be realized by a plurality of processors, memories, and other computing resources, such as a cloud computing system. In this case, each unit constituting the control unit 3 is realized by at least one of a plurality of different processors executing a program.

[0032] The knowledge graph acquisition unit 30 acquires a knowledge graph. Here, the knowledge graph acquired by the knowledge graph acquisition unit 30 is a graph configured by connecting different nodes with edges, and is a graph in which an entity is assigned to each node and a relationship between each node is assigned to an edge. The knowledge graph acquisition unit 30 acquires two different knowledge graphs.

[0033] The vector assignment unit 31 assigns vectors to each node and each edge constituting each knowledge graph for each of the two different knowledge graphs acquired by the knowledge graph acquisition unit 30. Although not limited thereto, the vector assignment unit 31 assigns initial vectors by using random numbers to each node and each edge constituting the knowledge graph acquired by the knowledge graph acquisition unit 30. The dimension of the vector is, for example, 200 dimensions.

[0034] For each of the two different knowledge graphs acquired by the knowledge graph acquisition unit 30, the pair extraction unit 32 extracts a pair consisting of three elements: a first node, an edge starting from the first node and ending at the second node, and the second node.

[0035] 3 is a schematic diagram showing, in a table format, an example of a pair extracted by the pair extraction unit 32 according to the embodiment. Specifically, the diagram shows an example of a pair extracted by the pair extraction unit 32 when the knowledge graph acquisition unit 30 acquires the two knowledge graphs (knowledge graph R and knowledge graph S) shown in FIG.

[0036] For example, in the knowledge graph R shown in Figure 1, node e 1 A relation R1 is defined that starts from node e and goes to node B. In this case, the set extraction unit 32 selects node e as the first node. 1 , B as the second node, e as the first node 1 The relationship R1 is extracted as one pair of edges starting from node B and ending at node B.

[0037] In addition, in the knowledge graph S shown in Fig. 1, node e 2A relation R3 is defined that starts from node e and goes to node f. In this case, the set extraction unit 32 selects node e as the first node. 2 , f as the second node, e as the first node 2 Then, the relation R3 is extracted as one pair, with the edge starting from and ending at the second node f.

[0038] In this manner, the set extraction unit 32 extracts sets individually for each of the knowledge graphs R and S acquired by the knowledge graph acquisition unit 30, based on the relationships between the nodes.

[0039] The positive example selection unit 33 selects, as a positive example, each of the multiple pairs extracted by the pair extraction unit 32. Here, a "positive example" selected by the positive example selection unit 33 means that the relationship between the nodes and edges constituting the pair is correct.

[0040] The graph relationship acquisition unit 36 ​​acquires node relationship information indicating relationships from multiple perspectives between nodes constituting one knowledge graph acquired by the knowledge graph acquisition unit 30 and nodes constituting the other knowledge graph.

[0041] As described above, Fig. 2 is a diagram showing, in a tabular form, an example of inter-node relationship information showing the relationship between an entity in the knowledge graph R shown in Fig. 1(a) and an entity in the knowledge graph S. The inter-node relationship information shown in Fig. 2 is an example, and any inter-node relationship information can be obtained if there is foresight information showing the relationship between a node constituting one knowledge graph and a node constituting the other knowledge graph.

[0042] The vector update unit 37 updates the vectors assigned by the vector assignment unit 31 based on the vectors assigned to the positive examples and an evaluation function related to the inter-node relation information.

[0043] The evaluation function L referred to by the vector update unit 37 according to the embodiment is expressed by the following formula (1).

[0044]

number

[0045] In formula (1), the first term on the right-hand side is a term relating to the structure of one of the two knowledge graphs, i.e., the link prediction described above. The second term on the right-hand side is a term relating to the link prediction of the other of the two knowledge graphs. The third term on the right-hand side is a term relating to inter-node relationship information, and is a term relating to the knowledge alignment described above. p, which is the index of Σ in the third term on the right-hand side, indicates the type of relationship between nodes. For example, in the example of inter-node relationship information shown in FIG. 2, there are three types of inter-node relationship information, viewpoint 1 to viewpoint 3, so p varies from 1 to 3.

[0046] The first term on the right side of the evaluation function L shown in equation (1) is expressed by the following equation (2).

[0047]

number

[0048] In the formula (2), for the set of positive examples selected by the positive example selection unit 33, h 1 is the vector assigned to the first node, r 1 is a vector assigned to edge relations, t 1 is the vector assigned to the second node, and |h 1 +r 1 -t 1 | is the norm of the difference vector between the vector assigned to the second node and the vector assigned to the edge added to the vector assigned to the first node. 1 may be evaluated taking into consideration not only positive examples but also negative examples.

[0049] The second term on the right side of the evaluation function L shown in equation (1) is expressed by the following equation (3).

[0050]

number

[0051] On the right side of equation (3), h 2 , r 2 and t 2 The subscript "2" in indicates that it is the other of the two knowledge graphs. The rest is the same as in equation (2). The first term L on the right side of the evaluation function L shown in equation (1) 1 And the second term on the right hand side, L 2 In both cases, the smaller the value, the better the structure of the knowledge graph can be expressed by vectors.

[0052] The third term on the right side of the evaluation function L shown in equation (1) is expressed by the following equation (4).

[0053]

number

[0054] On the right side of equation (4), Q p 1 represents the distribution of points corresponding to nodes with viewpoint P in one of the two knowledge graphs, and Q p 2 represents the distribution of points corresponding to nodes with viewpoint P in the other knowledge graph of the two knowledge graphs. Sim(A,B) is a function that reflects the similarity between distribution A in one of the two spaces and distribution B in the other space, such as the well-known Kullback-Leibler divergence, transportation distance, canonical correlation analysis, and the average value of pairwise Euclidean distance.

[0055] Figures 4(a)-(b) are diagrams for explaining the third term L3 on the right-hand side of the evaluation function L. Specifically, Figure 4(a) is a diagram showing, in a tabular form, vectors assigned to the nodes of the knowledge graph R and the nodes of the knowledge graph S shown in Figure 1, and Figure 4(b) is a schematic diagram showing a set of points corresponding to a vector when the vector is associated with one point in space.

[0056] As described above, the number of dimensions of the vector assigned to each node is, for example, 200. In this case, the vector assigned to the node corresponds to one point in a 200-dimensional space, but for convenience of illustration, in FIG. 4(b), it is conveniently illustrated as a point in a two-dimensional space. Note that FIG. 2 shows an example in which node-to-node relationship information regarding two or three nodes is given for each viewpoint. In this case, the set of points shown in FIG. 4(b) is also two or three points, but generally, the number of nodes to which node-to-node relationship information is given is two or more, and a distribution is formed as a set of points. The third term L3 on the right side of the evaluation function L becomes smaller as the similarity between the distribution of points corresponding to vectors assigned to nodes included in one knowledge graph and the distribution of points corresponding to vectors assigned to nodes included in the other knowledge graph increases for nodes appearing in each of the multiple viewpoints included in the node-to-node relationship information.

[0057] The vector update unit 37 updates the vectors assigned by the vector assignment unit 31 so as to reduce the evaluation function L. This allows the vector update unit 37 to balance link prediction, which expresses a knowledge graph with vectors, and knowledge alignment based on inter-node relationship information.

[0058] Here, the vector update unit 37 may gradually update the vectors assigned by the vector assignment unit 31 by iterative processing based on the gradient of the evaluation function L shown in formula (1) so that the evaluation function L becomes smaller. In addition, the vector update unit 37 updates the vectors assigned by the vector assignment unit 31 by iterative processing based on the gradient of the evaluation function L shown in formula (1) so that the evaluation function L becomes smaller. 2 After the update by the evaluation function derived from the set converges, the vector is updated by iterative processing based on the gradient of L 3 The vector may be updated by an iterative process based on the gradient of

[0059] This allows the information processing device 1 to execute, on different computers, updates using evaluation functions derived from sets that can be calculated for each knowledge graph and updates using evaluation functions derived from inter-node relationship information. Vector updates can be executed without moving the information for each knowledge graph from the computer on which it is stored, and if the knowledge graph contains personal information, the information can be updated while protecting it. Furthermore, identifiers of entities related to inter-node relationship information can be de-identified using an irreversible conversion function such as a hash function and shared. In this way, it is possible to manage personally identifiable information and non-identifiable information separately, and execute vector updates on each computer while keeping them separate.

[0060] <Processing flow of information processing executed by information processing device 1> 5 is a flowchart for explaining the flow of information processing executed by the information processing device 1 according to the embodiment. The processing in this flowchart starts, for example, when the information processing device 1 is started.

[0061] The knowledge graph acquisition unit 30 acquires two different knowledge graphs (S2). The vector assignment unit 31 assigns vectors to each node and each edge constituting each of the two different knowledge graphs acquired by the knowledge graph acquisition unit 30 (S4).

[0062] For each of the two different knowledge graphs acquired by the knowledge graph acquisition unit 30, the pair extraction unit 32 extracts a pair consisting of three elements: a first node, an edge starting from the first node and ending at the second node, and the second node (S6).

[0063] The positive example selection unit 33 selects each of the multiple pairs extracted by the pair extraction unit 32 as one positive example (S8). The graph relation acquisition unit 36 ​​acquires graph relation information indicating the relation between the nodes constituting one knowledge graph acquired by the knowledge graph acquisition unit 30 and the nodes constituting the other knowledge graph, for each of multiple viewpoints (S10).

[0064] The vector update unit 37 iteratively updates the vectors assigned by the vector assignment unit 31 based on the vectors assigned to the positive examples and an evaluation function related to the inter-node relation information (S12).

[0065] When the iterative vector update process by the vector allocation unit 31 ends, the process in this flowchart ends.

[0066] <Advantages of the Information Processing Device 1 According to the Embodiment> As described above, according to the information processing device 1 of the embodiment, unknown facts in knowledge graphs can be more effectively complemented by using foresight information on the relationships between nodes of different knowledge graphs while expressing the connection relationships of the nodes of each knowledge graph as vectors. This allows the information processing device 1 of the embodiment to achieve a balance between knowledge alignment and link prediction.

[0067] Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments, and various modifications and changes are possible within the scope of the gist of the present invention. For example, all or part of the device can be configured by distributing or integrating functionally or physically in any unit. In addition, new embodiments resulting from any combination of multiple embodiments are also included in the embodiments of the present invention. The effect of the new embodiment resulting from the combination also has the effect of the original embodiment.

[0068] <Modification> In the above, the evaluation function L is L 1 , L 2 , and L p Instead, the evaluation function L is expressed as the sum of L 1 , L 2 , and L p Specifically, the evaluation function L may be expressed by the following formula (5).

[0069]

number

[0070] This allows the information processing device 1 according to the modified example to change the degree of importance of the evaluation function related to link prediction and the evaluation function related to knowledge alignment in vector update. [Explanation of symbols]

[0071] 1. Information processing device 2...Storage section 3. Control section 30 Knowledge graph acquisition unit 31. Vector allocation unit 32...Pair extraction part 33 Positive example selection section 36. Graph relation acquisition part 37. Vector update section

Claims

1. a knowledge graph acquisition unit that acquires a knowledge graph, which is a graph configured by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; a vector assignment unit that assigns vectors to each node and each edge constituting each of the two different knowledge graphs acquired by the knowledge graph acquisition unit; a set extraction unit that extracts, for each of the two different knowledge graphs acquired by the knowledge graph acquisition unit, a set consisting of three elements: a first node, an edge starting from the first node and ending at a second node, and the second node; a positive example selection unit that selects each of the plurality of pairs extracted by the pair extraction unit as a positive example; a graph relationship acquisition unit that acquires node relationship information indicating relationships between nodes constituting one knowledge graph acquired by the knowledge graph acquisition unit and nodes constituting the other knowledge graph from a plurality of perspectives; a vector updating unit that updates the vector assigned by the vector assignment unit based on the vector assigned to the positive example and an evaluation function related to the inter-node relationship information, the evaluation function is set so as to become smaller as the degree of similarity between a distribution of points corresponding to vectors assigned to nodes included in one knowledge graph and a distribution of points corresponding to vectors assigned to nodes included in the other knowledge graph increases for nodes appearing in each of the plurality of viewpoints included in the node relationship information; the vector updating unit updates the vector assigned by the vector assigning unit so as to reduce the evaluation function. Information processing device.

2. a knowledge graph acquisition unit that acquires a knowledge graph, which is a graph configured by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; a vector assignment unit that assigns vectors to each node and each edge constituting each of the two different knowledge graphs acquired by the knowledge graph acquisition unit; a set extraction unit that extracts, for each of the two different knowledge graphs acquired by the knowledge graph acquisition unit, a set consisting of three elements: a first node, an edge starting from the first node and ending at a second node, and the second node; a positive example selection unit that selects each of the plurality of pairs extracted by the pair extraction unit as a positive example; a graph relationship acquisition unit that acquires node relationship information indicating relationships between nodes constituting one knowledge graph acquired by the knowledge graph acquisition unit and nodes constituting the other knowledge graph from a plurality of perspectives; a vector updating unit that updates the vector assigned by the vector assignment unit based on the vector assigned to the positive example and an evaluation function related to the inter-node relationship information, The vector update unit is updating the vector by an iterative process based on a gradient of a cost function derived from the set; updating the vector by an iterative process based on a gradient of the evaluation function derived from the inter-node relationship information after the update by the evaluation function derived from the set has converged; Information processing device.

3. the evaluation function is set as a norm of a difference vector between a vector obtained by adding a vector assigned to the edge to a vector assigned to the first node, and a vector assigned to the second node, for the pair that is a positive example selected by the positive example selection unit.

3. The information processing device according to claim 1 or 2.

4. The processor: A step of obtaining a knowledge graph, which is a graph configured by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; A step of assigning vectors to each node and each edge constituting each of the two different knowledge graphs obtained; A step of extracting, for each of the two different knowledge graphs obtained, a set of three elements including a first node, an edge starting from the first node and ending at a second node, and the second node; selecting each of the extracted sets as a positive example; acquiring inter-node relationship information indicating relationships between nodes constituting one of the acquired knowledge graphs and nodes constituting the other knowledge graph from a plurality of perspectives; updating the assigned vector based on an evaluation function related to the vector assigned to the positive example and the inter-node relationship information; the evaluation function is set so as to become smaller as the degree of similarity between a distribution of points corresponding to vectors assigned to nodes included in one knowledge graph and a distribution of points corresponding to vectors assigned to nodes included in the other knowledge graph increases for nodes appearing in each of the plurality of viewpoints included in the node relationship information; In the updating step, the assigned vector is updated so that the evaluation function becomes smaller. Information processing methods.

5. The processor: A step of obtaining a knowledge graph, which is a graph configured by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; A step of assigning vectors to each node and each edge constituting each of the two different knowledge graphs obtained; A step of extracting, for each of the two different knowledge graphs obtained, a set of three elements including a first node, an edge starting from the first node and ending at a second node, and the second node; selecting each of the extracted sets as a positive example; acquiring inter-node relationship information indicating relationships between nodes constituting one of the acquired knowledge graphs and nodes constituting the other knowledge graph from a plurality of perspectives; updating the assigned vector based on the vector assigned to the positive example and an evaluation function related to the node-to-node relationship information, the vector being updated by an iterative process based on a gradient of the evaluation function derived from the set, and after the update based on the evaluation function derived from the set has converged, updating the vector by an iterative process based on the gradient of the evaluation function derived from the node-to-node relationship information; Information processing methods.

6. On the computer, A function of acquiring a knowledge graph, which is a graph configured by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; A function of assigning vectors to each node and each edge constituting each of the two different knowledge graphs obtained; A function of extracting, for each of the two different acquired knowledge graphs, a set of three elements: a first node, an edge starting from the first node and ending at a second node, and the second node; and A function of selecting each of the extracted multiple pairs as one positive example; A function of acquiring node-to-node relationship information indicating relationships between nodes constituting one acquired knowledge graph and nodes constituting the other acquired knowledge graph from multiple perspectives; updating the assigned vector based on an evaluation function related to the vector assigned to the positive example and the inter-node relationship information; the evaluation function is set so as to become smaller as the degree of similarity between a distribution of points corresponding to vectors assigned to nodes included in one knowledge graph and a distribution of points corresponding to vectors assigned to nodes included in the other knowledge graph increases for nodes appearing in each of the plurality of viewpoints included in the node relationship information; The updating function updates the assigned vector so that the evaluation function becomes smaller. program.

7. On the computer, A function of acquiring a knowledge graph, which is a graph configured by connecting different nodes with edges, in which an entity is assigned to each node and a relationship between each node is assigned to an edge; A function of assigning vectors to each node and each edge constituting each of the two different knowledge graphs obtained; A function of extracting, for each of the two different acquired knowledge graphs, a set of three elements: a first node, an edge starting from the first node and ending at a second node, and the second node; and A function of selecting each of the extracted multiple pairs as one positive example; A function of acquiring node-to-node relationship information indicating relationships between nodes constituting one acquired knowledge graph and nodes constituting the other acquired knowledge graph from multiple perspectives; a function of updating the assigned vector based on the vector assigned to the positive example and an evaluation function related to the node relationship information, the function updating the vector by an iterative process based on a gradient of the evaluation function derived from the set, and updating the vector by an iterative process based on the gradient of the evaluation function derived from the node relationship information after the update by the evaluation function derived from the set has converged; program.

Images