Information processing device, information processing method, and program

Abstract

An information processing device, wherein a partial graph data generation means generates a plurality of partial graph data corresponding respectively to a plurality of mutually associated data being analyzed. An integration means integrates the plurality of partial graph data to thereby generate graph data. A training means trains a learning model using the graph data and relationship data that indicates a known relationship between nodes that are linked in the graph data. An analysis means performs, using the trained learning model, an analysis that corresponds to the purpose of analysis associated with at least one of the plurality of data being analyzed.

Description

Information processing device, information processing method, and recording medium

[0001] The present disclosure relates to a technique for performing processing using graph data.

[0002] Graph data such as knowledge graphs have been known as data capable of expressing relationships between multiple entities. Graph data has also been utilized in recent years for purposes such as recommending products to customers.

[0003] Specifically, for example, Patent Document 1 discloses a technology for recommending various products to customers based on a knowledge graph generated to include multiple entities, attributes of the multiple entities, and relationships between the multiple entities.

[0004] Patent No. 6829240

[0005] Here, the graph data can also be used for purposes such as developing new products with concepts suited to each customer.

[0006] However, the knowledge graph disclosed in Patent Document 1 is intended for use with a specific purpose, namely, product recommendation, and is not intended for use with other purposes such as new product development, etc. Therefore, the technology disclosed in Patent Document 1 has a problem in that, in order to obtain analysis results corresponding to each of multiple purposes that are assumed to be related to each other, such as product recommendation and new product development, separate processing must be performed using multiple knowledge graphs generated for each of the multiple purposes.

[0007] In other words, the technology disclosed in Patent Document 1 presents a problem corresponding to the aforementioned problem of complicated procedures for obtaining analysis results corresponding to multiple objectives that are assumed to be related to each other.

[0008] An object of the present disclosure is to provide an information processing device that can easily obtain analysis results according to each of a plurality of purposes that are presumed to be related to one another.

[0009] In one aspect of the present disclosure, an information processing device has a subgraph data generation means for generating a plurality of subgraph data corresponding to each of a plurality of pieces of analysis target data that are related to each other, an integration means for generating graph data by integrating the plurality of subgraph data, a learning means for training a learning model using the graph data and relationship data that is data indicating known relationships between linked nodes in the graph data, and an analysis means for using the trained learning model to perform analysis according to an analysis purpose related to at least one of the plurality of pieces of analysis target data.

[0010] In another aspect of the present disclosure, an information processing method generates a plurality of subgraph data corresponding to each of a plurality of mutually related analysis target data, generates graph data by integrating the plurality of subgraph data, trains a learning model using the graph data and relationship data that is data indicating known relationships between linked nodes in the graph data, and uses the trained learning model to perform analysis according to an analysis objective related to at least one of the plurality of analysis target data.

[0011] In yet another aspect of the present disclosure, a recording medium records a program that causes a computer to execute a process of generating a plurality of partial graph data corresponding to each of a plurality of pieces of analysis target data that are related to each other, generating graph data by integrating the plurality of partial graph data, training a learning model using the graph data and relationship data that is data indicating known relationships between linked nodes in the graph data, and using the trained learning model to perform analysis according to an analysis objective related to at least one of the plurality of pieces of analysis target data.

[0012] According to the present disclosure, analysis results corresponding to each of a plurality of objectives that are presumed to be related to one another can be easily obtained.

[0013] 1 is a diagram showing a schematic configuration of an information processing system including a server device according to the first embodiment. FIG. 2 is a block diagram showing the hardware configuration of a server device according to the first embodiment. FIG. 3 is a block diagram showing the functional configuration of a server device according to the first embodiment. FIG. 4 is a diagram showing an example of a record obtained by converting input data using dictionary data. FIG. 5 is a diagram showing graph data generated based on each record in FIG. 4. FIG. 6 is a diagram showing an example of subgraph data obtained by processing of a server device according to the first embodiment. FIG. 7 is a diagram showing an example of subgraph data obtained by processing of a server device according to the first embodiment. FIG. 8 is a diagram showing an example of graph data obtained by processing of a server device according to the first embodiment. FIG. 9 is a diagram for explaining analysis results obtained by processing of a server device according to the first embodiment. A flowchart showing an example of processing performed in a server device according to the first embodiment. A block diagram showing the functional configuration of an information processing device according to a second embodiment. A flowchart for explaining processing performed in an information processing device according to the second embodiment.

[0014] Hereinafter, preferred embodiments of the present disclosure will be described with reference to the drawings.

[0015] <First embodiment> [System configuration] Fig. 1 is a diagram showing a schematic configuration of an information processing system including a server device according to First embodiment. As shown in Fig. 1, the information processing system 1 has a server device 100, a customer terminal device 200, and a developer terminal device 300.

[0016] The server device 100 is configured to be able to communicate with the customer terminal device 200 and the developer terminal device 300. The server device 100 also performs analysis using a learning model described below to obtain, for example, analysis results relating to products that may be purchased by customers, and outputs the analysis results to the outside of the server device 100. The server device 100 also performs analysis using a learning model described below to obtain, for example, analysis results relating to insights that are useful in developing new products, and outputs the analysis results to the outside of the server device 100.

[0017] The customer terminal device 200 has a function of communicating with the server device 100, a function of inputting information to be sent to the server device 100, and a function of displaying information received from the server device 100. The customer terminal device 200 also has a function of displaying information according to operations performed by the customer. Specifically, the customer terminal device 200 may be configured by a device such as a personal computer, a smartphone, or a tablet computer.

[0018] The developer terminal device 300 has functions such as communicating with the server device 100, inputting information to be sent to the server device 100, and displaying information received from the server device 100. The developer terminal device 300 also has a function of displaying information in response to operations by the developer. Specifically, the developer terminal device 300 may be configured with devices such as a personal computer, a smartphone, or a tablet computer, for example.

[0019] [Hardware Configuration] Fig. 2 is a block diagram showing the hardware configuration of the server device according to the first embodiment. As shown in Fig. 2, the server device 100 has an interface (IF) 111, a processor 112, a memory 113, a recording medium 114, and a database (DB) 115.

[0020] The IF 111 inputs and outputs data to and from external devices. For example, information transmitted from the customer terminal device 200 and information transmitted from the developer terminal device 300 are input to the server device 100 via the IF 111.

[0021] The processor 112 is a computer such as a CPU (Central Processing Unit), and executes a prepared program to control the entire server device 100. Specifically, the processor 112 performs analysis using, for example, a learning model described below.

[0022] The memory 113 is configured by a ROM (Read Only Memory), a RAM (Random Access Memory), etc. The memory 113 is also used as a working memory while the processor 112 is executing various processes.

[0023] The recording medium 114 is a non-volatile, non-transitory recording medium such as a disk-shaped recording medium or a semiconductor memory, and is configured to be detachable from the server device 100. The recording medium 114 records various programs to be executed by the processor 112. When the server device 100 executes various processes, the programs recorded on the recording medium 114 are loaded into the memory 113 and executed by the processor 112.

[0024] The DB 115 stores, for example, information input via the IF 111 and processing results obtained by processing by the processor 112 .

[0025] [Functional Configuration] Fig. 3 is a block diagram showing the functional configuration of the server device according to the first embodiment. As shown in Fig. 3, the server device 100 has an analysis data storage unit 11, a graph data generation unit 12, a graph data storage unit 13, a learning processing unit 14, a processing result storage unit 15, a query generation unit 16, and an analysis processing unit 17.

[0026] The analytical data storage unit 11 stores analysis target data ADA, analysis target data ADB, and dictionary data JD.

[0027] The analysis target data ADA includes, for example, customer data indicating customer attributes, purchase history, etc. The analysis target data ADB includes, for example, idea data indicating similar products, prices, targets, etc. of a new product under development. Note that in this embodiment, the analysis target data ADA and ADB may be data that are related to each other but are used for different purposes.

[0028] The dictionary data JD includes data indicating rules to be referenced when the graph data generating unit 12 generates graph data and when the query generating unit 16 generates queries.

[0029] The graph data generation unit 12 generates graph data GD based on the analysis target data ADA, analysis target data ADB, and dictionary data JD read from the analysis data storage unit 11, and stores the generated graph data GD in the graph data storage unit 13. The graph data generation unit 12 also has a data conversion unit 12A and an integration processing unit 12B.

[0030] The data conversion unit 12A functions as a subgraph data generation means. The data conversion unit 12A also converts the analysis target data ADA into subgraph data GDA based on the rules indicated by the dictionary data JD. The data conversion unit 12A also converts the analysis target data ADB into subgraph data GDB based on the rules indicated by the dictionary data JD.

[0031] The integration processing unit 12B functions as an integration unit. The integration processing unit 12B also generates graph data GD by integrating the partial graph data GDA and GDB, and stores the generated graph data GD in the graph data storage unit 13.

[0032] The graph data storage unit 13 stores the graph data GD generated by the graph data generation unit 12 .

[0033] The learning processing unit 14 functions as a learning means. The learning processing unit 14 acquires, as relationship data KD, data indicating known relationships between linked nodes in the graph data GD read from the graph data storage unit 13. The relationship data KD may include, for example, data indicating relationships between multiple nodes as universal rules by replacing each of multiple nodes having the same relationship with a variable node. The learning processing unit 14 uses the graph data GD and the relationship data KD to train a learning model GMD so as to derive unknown relationships between unlinked nodes in the graph data GD. The learning processing unit 14 stores the relationship data KD and the trained learning model GMD in the processing result storage unit 15.

[0034] The processing result storage unit 15 stores the relationship data KD and the learned learning model GMD as data obtained by the processing of the learning processing unit 14.

[0035] The query generation unit 16 functions as a query generation means. The query generation unit 16 generates a query according to an analysis purpose input from outside the server device 100, based on the rules indicated by the dictionary data JD. The analysis purpose may be set as being related to at least one of the analysis target data ADA and ADB.

[0036] The analysis processing unit 17 functions as an analysis unit. The analysis processing unit 17 uses the trained learning model GMD read from the processing result storage unit 15 to analyze the query generated by the query generation unit 16. The analysis processing unit 17 also outputs the analysis results obtained by the above-mentioned analysis to the outside of the server device 100.

[0037] [Specific Example] Next, a specific example of the processing performed in each unit of the server device 100 will be described.

[0038] The dictionary data JD contains rules for converting input data in various formats, such as a table format, into records in a predetermined format. According to these rules, for example, input data can be converted into records RA to RD as shown in Figure 4. Figure 4 shows examples of records obtained by converting input data using the dictionary data.

[0039] For each of records RA to RD, the leftmost element (the character strings "Jacket," "T-Shirt," and "Tom") represents the link source node in the graph, and the rightmost element (the character strings "clothes" and "Male") represents the link destination node in the graph. Furthermore, for each of records RA to RD, the central element (the character strings "is_category," "is_gender," and "interested_for") represents the relationship between the linked nodes in the graph. Furthermore, for each of records RA to RD, the order of each element (each character string) represents the link direction (the direction of the arrow in the graph) when linking nodes in the graph. Therefore, records RA to RD in FIG. 4 can be treated as the data that forms the basis of graph data GDX as shown in FIG. 5. Note that the arrows linking nodes in the graph may be interpreted as edges. FIG. 5 is a diagram showing graph data generated based on each record in FIG.

[0040] The data conversion unit 12A converts the analysis target data ADA based on the rules described in the specific examples above while referring to the dictionary data JD, thereby generating subgraph data GDA including multiple records corresponding to the analysis target data ADA. The data conversion unit 12A also converts the analysis target data ADB based on the rules described in the specific examples above while referring to the dictionary data JD, thereby generating subgraph data GDB including multiple records corresponding to the analysis target data ADB. That is, the data conversion unit 12A generates multiple subgraph data corresponding to each of the multiple analysis target data that are related to each other. The data conversion unit 12A also generates multiple subgraph data by converting each of the multiple analysis target data into data in a predetermined format that represents four elements: a link source node, a link destination node, the relationship between the nodes, and the link direction when linking the nodes.

[0041] When generating a record from the analysis target data, the data conversion unit 12A may use category names or item names included in the analysis target data, such as "gender" and "name." Furthermore, for example, when the analysis target data does not contain data necessary for generating a record, the data conversion unit 12A may generate a record using data that can be obtained based on the analysis target data. Specifically, for example, when the analysis target data includes a shopping list recording multiple items purchased by a single customer, the data conversion unit 12A may calculate the average price of the multiple items as the average purchase amount of the single customer and generate a record using the calculated average purchase amount. Furthermore, the data conversion unit 12A may generate a record from multiple data included in the analysis target data.

[0042] The integration processing unit 12B generates graph data GD by integrating the subgraph data GDA and GDB and stores the generated graph data GD in the graph data storage unit 13. Specifically, the integration processing unit 12B integrates, for example, subgraph data GDA as shown in FIG. 6 with subgraph data GDB as shown in FIG. 7 to generate graph data GD as shown in FIG. 8 and stores the generated graph data GD in the graph data storage unit 13. That is, the integration processing unit 12B generates graph data by integrating multiple subgraph data. Furthermore, the integration processing unit 12B generates graph data by integrating multiple subgraph data based on common elements contained in the multiple subgraph data. FIGS. 6 and 7 are diagrams showing examples of subgraph data obtained by processing by the server device according to the first embodiment. FIG. 8 is a diagram showing an example of graph data obtained by processing by the server device according to the first embodiment.

[0043] The partial graph data GDA in Fig. 6 can be generated based on a plurality of records obtained by converting the analysis target data ADA including customer data related to customer P. Specifically, the partial graph data GDA in Fig. 6 can be generated based on records indicating, for example, that customer P is a woman, that customer P frequently purchases product L, that the average purchase amount for customer P is in the 1,000 yen range, that a similar product to product L is product M, and that the characteristics of product M include "strong sourness" and "strong aroma."

[0044] The partial graph data GDB in Fig. 7 can be generated based on a plurality of records obtained by converting the analysis target data ADB that includes idea data related to new product N under development. Specifically, the partial graph data GDB in Fig. 7 can be generated based on records that indicate, for example, that the target audience of new product N is women, that the price of new product N is in the 1,000 yen range, that a similar product to new product N is product M, and that the characteristics of product M include "strong sourness."

[0045] Here, the hatched nodes in the graph data GD of Fig. 8 indicate nodes that are included in both the partial graph data GDA of Fig. 6 and the partial graph data GDB of Fig. 7. Note that "other new product ideas" in the graph data GD of Fig. 8 represent products under development that are different from new product N and that are included in the graph data before the partial graph data GDA and GDB are integrated. Also, "other customers" in the graph data GD of Fig. 8 represent customers that are different from customer P and that are included in the graph data before the partial graph data GDA and GDB are integrated.

[0046] The learning processing unit 14 acquires relationship data KD indicating known relationships between linked nodes in the graph data GD read from the graph data storage unit 13 .

[0047] The relationship data KD includes data indicating relationships corresponding to each of the multiple arrows in the graph data GD of Fig. 8. Specifically, the relationship data KD includes, for example, data indicating that product M has the characteristic of being "strongly sour" and data indicating that product M has the characteristic of being "strongly scented." The relationship data KD also includes, for example, data indicating that product L is a similar product to product M and data indicating that new product N is a similar product to product M.

[0048] The learning processing unit 14 learns the learning model GMD constructed based on, for example, "KBLRN" by inputting feature values ​​corresponding to each node included in the graph data GD and feature values ​​corresponding to each relationship included in the relationship data KD to the learning model GMD. The learning processing unit 14 also stores the relationship data KD and the learned learning model GMD in the processing result storage unit 15.

[0049] The aforementioned "KBLRN" is disclosed, for example, in "KBLRN: End-to-End Learning of Knowledge Base Representations with Latent, Relational, and Numerical Features" by Alberto Garcia-Duran et al. Furthermore, the learning model GMD may be constructed based on a model other than "KBLRN" as long as it has a configuration capable of performing link prediction for graph data. Furthermore, when the learning processing unit 14 learns the learning model GMD using the graph data GD and the relationship data KD, it is desirable to perform zero-shot learning, such as that disclosed in Japanese Patent Application Laid-Open No. 2019-125364. Furthermore, according to this specific example, for example, each time at least one of the analysis target data ADA and ADB is updated, the graph data GD may be updated and the learning model GMD may be re-learned using the updated graph data GD.

[0050] Based on the rules indicated by the dictionary data JD, the query generation unit 16 generates a query corresponding to an analysis objective input from outside the server device 100, for example, a record in which a link source node, a relationship between the nodes, and a link direction are represented, but the link destination node is unknown. For example, when an analysis objective such as "items predicted to be purchased by customer P" is input, the query generation unit 16 generates a query QA corresponding to the analysis objective. For example, the query QA is generated as a record in which a character string representing "customer P" corresponding to the link source node and a character string representing "items predicted to be purchased" corresponding to the relationship between the nodes are present, and the word order of "customer P" and "items predicted to be purchased" corresponding to the link direction is clear, but no character or character string representing the link destination node is present. Furthermore, when an analysis objective such as "features suitable for new product N under development" is input, the query generation unit 16 generates a query QB corresponding to the analysis objective. The query QB is generated as a record in which there is a character string representing "new product N" corresponding to the link source node and a character string representing a "feature" corresponding to the relationship between the nodes, and the word order of "new product N" and "feature" corresponding to the link direction is clear, but there is no character or character string representing the link destination node. Note that the query generation unit 16 is only required to generate a record in which one of the four elements of the link source node, the link destination node, the relationship between the nodes, and the link direction is unknown, as a query according to the analysis purpose input from outside the server device 100.

[0051] The analysis processing unit 17 uses the trained learning model GMD read from the processing result storage unit 15 to perform an analysis related to the query generated by the query generation unit 16. In other words, the analysis processing unit 17 uses the trained learning model GMD to perform an analysis according to an analysis purpose related to at least one of the multiple pieces of analysis target data. Furthermore, the analysis processing unit 17 uses the trained learning model GMD to perform an analysis related to the query generated by the query generation unit 16, thereby obtaining an analysis result corresponding to one unknown element in the query. Furthermore, the analysis processing unit 17 outputs the analysis result obtained by the above-mentioned analysis to an output destination corresponding to the query generated by the query generation unit 16.

[0052] For example, the analysis processing unit 17 uses the trained learning model GMD to perform an analysis to predict a link destination node in the record of the query QA. By performing the above-described analysis, the analysis processing unit 17 obtains an analysis result indicating that existing product M and new product N under development are predicted as products that customer P is likely to purchase. The analysis processing unit 17 also outputs the analysis result corresponding to the query QA to at least one of the customer terminal device 200 and the developer terminal device 300. The analysis result corresponding to the query QA can be represented, for example, as a dashed arrow pointing from the node for "customer P" to the node for "product M" in the graph data GD, and a dashed-dotted arrow pointing from the node for "customer P" to the node for "new product N" in the graph data GD, as shown in FIG. 9 . Furthermore, the analysis result corresponding to the query QA can recommend product M to customer P, who is female and whose average purchase amount is in the 1,000 yen range, and can also identify that new product N is a potential purchase target for customer P. Furthermore, according to the analysis result corresponding to the query QA, for example, information for recommending the product M can be displayed on the customer terminal device 200 used by the customer P. Fig. 9 is a diagram for explaining the analysis result obtained by the processing of the server device according to the first embodiment.

[0053] Furthermore, for example, the analysis processing unit 17 uses the trained learning model GMD to perform an analysis to predict a link destination node in the record of query QB. By performing the above-described analysis, the analysis processing unit 17 obtains an analysis result indicating that "strong sourness" and "strong aroma" are suitable features for new product N under development. The analysis processing unit 17 also outputs the analysis result corresponding to query QB to the developer terminal device 300. The analysis result corresponding to query QB can be represented, for example, as shown in FIG. 9 , by a dashed arrow pointing from the node for "new product N" to the node for "strong sourness" in the graph data GD, and a dashed arrow pointing from the node for "new product N" to the node for "strong aroma" in the graph data GD. Furthermore, the analysis result corresponding to query QB can provide a developer with an insight that, for example, when developing a new product N for women priced in the 1,000 yen range similar to product M, it would be beneficial to include the features "strong sourness" and "strong aroma" in the new product N. Furthermore, based on the analysis results corresponding to the query QB, for example, the developer terminal device 300 can display information to recommend "strong sourness" and "strong aroma" as characteristics of the new product N.

[0054] [Processing Flow] Next, a description will be given of the flow of processing performed in the server device according to the first embodiment. Fig. 10 is a flowchart showing an example of processing performed in the server device according to the first embodiment.

[0055] First, the server device 100 generates subgraph data corresponding to each of the plurality of pieces of analysis target data stored in the analysis data storage unit 11 (step S11).

[0056] Next, the server device 100 generates graph data by integrating the subgraph data generated in step S11 (step S12).

[0057] Next, the server device 100 performs learning of the learning model using the graph data generated in step S12 and relationship data indicating known relationships between linked nodes in the graph data (step S13).

[0058] Next, the server device 100 generates a query according to the analysis purpose input from the outside (step S14).

[0059] Next, the server device 100 uses the trained learning model obtained in step S13 to analyze the query generated in step S14 (step S15).

[0060] Next, the server device 100 outputs the analysis result obtained in step S15 (step S16). Specifically, the server device 100 outputs the analysis result obtained in step S16 to either the customer terminal device 200 or the developer terminal device 300, for example, based on the query generated in step S14.

[0061] As described above, according to this embodiment, a plurality of subgraph data corresponding to each of a plurality of pieces of analysis target data that are related to each other are generated, and the learning model is trained using graph data that integrates the plurality of subgraph data. Furthermore, according to this embodiment, the above-described trained learning model can be used to perform analyses according to various purposes. Therefore, according to this embodiment, analysis results according to each of a plurality of purposes that are presumed to be related to each other can be easily obtained. Furthermore, according to this embodiment, by performing analysis using the above-described learning model, highly accurate analysis results according to the analysis purpose can be obtained. Furthermore, according to this embodiment, for example, the learning model can be retrained using graph data updated according to the analysis results, and therefore analysis using the learning model obtained by the retraining can be performed smoothly.

[0062] This embodiment can be applied to various fields as long as it is possible to perform link prediction in graph data obtained by integrating multiple subgraph data. Specifically, this embodiment can be applied, for example, to presenting information for recommending reviews and / or recipes suitable for each customer registered on a specific community site. This embodiment can also be applied, for example, to presenting information related to insights useful for marketing a new product to be developed, such as attributes preferred by a specific user, unknown usage methods, and potential customers.

[0063] Second Embodiment FIG. 11 is a block diagram showing the functional configuration of an information processing apparatus according to a second embodiment.

[0064] The information processing device 500 according to this embodiment has the same hardware configuration as the server device 100. The information processing device 500 also has a subgraph data generation unit 511, an integration unit 512, a learning unit 513, and an analysis unit 514.

[0065] FIG. 12 is a flowchart for explaining the processing performed in the information processing apparatus according to the second embodiment.

[0066] The subgraph data generating means 511 generates a plurality of subgraph data corresponding to each of a plurality of pieces of analysis target data that are related to each other (step S51).

[0067] The integration means 512 integrates a plurality of pieces of subgraph data to generate graph data (step S52).

[0068] The learning means 513 learns the learning model 600 using the graph data and relationship data that indicates known relationships between linked nodes in the graph data (step S53).

[0069] The analysis means 514 uses the learned learning model 600 to perform an analysis according to the analysis purpose related to at least one of the plurality of data to be analyzed (step S54).

[0070] According to this embodiment, analysis results according to each of a plurality of objectives that are presumed to have a relationship with each other can be easily obtained.

[0071] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0072] (Supplementary Note 1) An information processing device having: a subgraph data generation means for generating a plurality of subgraph data corresponding to each of a plurality of mutually related analysis target data; an integration means for generating graph data by integrating the plurality of subgraph data; a learning means for learning a learning model using the graph data and relationship data that is data indicating known relationships between linked nodes in the graph data; and an analysis means for using the learned learning model to perform analysis according to an analysis purpose related to at least one of the plurality of analysis target data.

[0073] (Supplementary Note 2) The information processing device of Supplementary Note 1, wherein the subgraph data generation means generates the plurality of subgraph data by converting each of the plurality of analysis target data into data in a predetermined format that represents four elements: a link source node, a link destination node, a relationship between the nodes, and a link direction when linking the nodes.

[0074] (Supplementary Note 3) The information processing device of Supplementary Note 2 further comprises a query generation means for generating data in which one of the four elements is unknown as a query according to the analysis purpose, wherein the analysis means uses the learning model that has undergone the learning to perform an analysis related to the query, thereby obtaining an analysis result corresponding to the one element.

[0075] (Supplementary Note 4) The information processing device of Supplementary Note 1, wherein the integration means generates the graph data by integrating the plurality of subgraph data based on a common element contained in the plurality of subgraph data.

[0076] (Supplementary Note 5) The information processing device of Supplementary Note 1, wherein the learning means learns the learning model so that unknown relationships between unlinked nodes in the graph data are derived.

[0077] (Supplementary Note 6) An information processing method comprising: generating a plurality of partial graph data corresponding to each of a plurality of pieces of analysis target data that are related to each other; generating graph data by integrating the plurality of partial graph data; training a learning model using the graph data and relationship data that is data indicating known relationships between linked nodes in the graph data; and using the trained learning model to perform analysis according to an analysis objective related to at least one of the plurality of pieces of analysis target data.

[0078] (Supplementary Note 7) A recording medium having recorded thereon a program that causes a computer to execute a process of generating a plurality of partial graph data corresponding to each of a plurality of pieces of analysis target data that are related to each other, generating graph data by integrating the plurality of partial graph data, training a learning model using the graph data and relationship data that is data that indicates known relationships between nodes linked in the graph data, and using the trained learning model to perform analysis according to an analysis purpose related to at least one of the plurality of pieces of analysis target data.

[0079] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments and examples. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

[0080] 12 Graph data generation unit 12A Data conversion unit 12B Integration processing unit 14 Learning processing unit 16 Query generation unit 17 Analysis processing unit 100 Server device

Claims

1. a subgraph data generating means for generating a plurality of subgraph data corresponding to each of a plurality of mutually related analysis target data; an integration means for generating graph data by integrating the plurality of subgraph data; a learning means for learning a learning model using the graph data and relationship data that indicates known relationships between linked nodes in the graph data; an analysis means for performing an analysis according to an analysis purpose related to at least one of the plurality of analysis target data using the learning model that has undergone the learning; An information processing device having the above.

2. 2. The information processing device according to claim 1, wherein the subgraph data generation means generates the plurality of subgraph data by converting each of the plurality of analysis target data into data in a predetermined format that represents four elements: a link source node, a link destination node, a relationship between the nodes, and a link direction when linking the nodes.

3. further comprising a query generation means for generating data in which one of the four elements is unknown as a query according to the analysis purpose; The information processing device according to claim 2 , wherein the analysis means acquires an analysis result corresponding to the one element by performing an analysis related to the query using the learning model that has undergone the learning.

4. The information processing apparatus according to claim 1 , wherein the integrating means generates the graph data by integrating the plurality of subgraph data based on a common element contained in the plurality of subgraph data.

5. The information processing apparatus according to claim 1 , wherein the learning means performs learning of the learning model so that unknown relationships between nodes that are not linked in the graph data are derived.

6. An information processing method executed by a computer, comprising: generating a plurality of subgraph data corresponding to each of a plurality of mutually related analysis target data; generating graph data by integrating the plurality of subgraph data; learning a learning model using the graph data and relationship data that is data indicating known relationships between linked nodes in the graph data; An information processing method that uses the learning model that has undergone the learning to perform analysis according to an analysis purpose related to at least one of the plurality of pieces of analysis target data.

7. generating a plurality of subgraph data corresponding to each of a plurality of mutually related analysis target data; generating graph data by integrating the plurality of subgraph data; learning a learning model using the graph data and relationship data that is data indicating known relationships between linked nodes in the graph data; A program that causes a computer to execute a process of using the learning model that has undergone the learning to perform an analysis according to an analysis purpose related to at least one of the plurality of data to be analyzed.