Device for integrating knowledge graphs, and recommendation device

Abstract

A device 10 for integrating knowledge graphs IG comprises: a first acquisition unit 113 that acquires a first knowledge graph IG1 having a plurality of nodes N1 and a second knowledge graph IG2 having a plurality of nodes N2; and an integration unit 117 that connects a first individual node IN1 included among the plurality of nodes N1 of the first knowledge graph IG1 and a second individual node IN2 included among the plurality of nodes N2 of the second knowledge graph IG2 via a connection edge CE, thereby integrating the first knowledge graph IG1 and the second knowledge graph IG2. The integration unit 117 determines the first individual node IN1 and the second individual node IN2 on the basis of a plurality of first degrees of similarity between each of the plurality of nodes N1 of the first knowledge graph IG1 and each of the plurality of nodes N2 of the second knowledge graph IG2, and a plurality of second degrees of similarity between the plurality of nodes N2 of the second knowledge graph IG2.

Description

Knowledge Graph Integration Device and Recommendation Device

[0001] The present invention relates to a knowledge graph integration device and a recommendation device.

[0002] Conventionally, as an example, in order to provide information related to products necessary for recommending products or services, a graph showing the relationship between nodes has been used.

[0003] For example, Patent Document 1 discloses a product name matching system that provides information related to products necessary for recommending products by additional information. In the technology according to Patent Document 1, the above additional information includes a product graph showing the relationship between products or graph vector information for generating a product graph. In the technology according to Patent Document 1, in order to more effectively recommend products, it is conceivable to use other graphs in addition to the above product graph.

[0004] In this regard, Patent Document 2 discloses a technique for generating an integrated graph that integrates a first knowledge graph and a second knowledge graph when using a plurality of graphs.

[0005] JP-A-2023-014409 JP-A-2024-005871

[0006] However, in the technology according to Patent Document 2, the node at the apex of another tree-shaped second knowledge graph is connected under the terminal node of the tree-shaped first knowledge graph. The connection was executed only based on the similarity between the nodes of the first knowledge graph and the nodes of the second knowledge graph.

[0007] An object of the present disclosure is to provide a knowledge graph integration device and a recommendation device that can more effectively recommend products or services when using a graph in which a plurality of knowledge graphs are integrated for recommending products or services as an example.

[0008] The knowledge graph integrator according to this disclosure comprises a first acquisition unit that acquires a first knowledge graph having a plurality of nodes and a second knowledge graph having a plurality of nodes, and an integration unit that integrates the first knowledge graph and the second knowledge graph by joining a first individual node included in the plurality of nodes of the first knowledge graph and a second individual node included in the plurality of nodes of the second knowledge graph with a joining edge, wherein the integration unit determines the first individual node and the second individual node based on a plurality of first similarities between each of the plurality of nodes of the first knowledge graph and each of the plurality of nodes of the second knowledge graph and a plurality of second similarities between the plurality of nodes of the second knowledge graph.

[0009] The recommendation device according to this disclosure comprises a first acquisition unit that acquires a first knowledge graph having a plurality of nodes and a second knowledge graph having a plurality of nodes, and an integration unit that integrates the first knowledge graph and the second knowledge graph by joining a first individual node included in the plurality of nodes of the first knowledge graph and a second individual node included in the plurality of nodes of the second knowledge graph with a joining edge, wherein the integration unit combines the first individual node and the second individual node based on a plurality of first similarities between each of the plurality of nodes of the first knowledge graph and each of the plurality of nodes of the second knowledge graph and a plurality of second similarities between the plurality of nodes of the second knowledge graph. The system further comprises: a determination unit that determines, based on information about the user's behavior, which nodes of the first knowledge graph are presumed to be of interest to the user, which are among a plurality of nodes in the first knowledge graph; a search unit that uses the first knowledge graph and the second knowledge graph, which have been integrated by the integration unit, to search the second knowledge graph starting from the node determined by the determination unit; and a recommendation unit that recommends products or services to the user based on the search results of the search unit.

[0010] According to this disclosure, for example, when using a graph that integrates multiple knowledge graphs to recommend goods or services, it becomes possible to recommend goods or services more effectively.

[0011] A block diagram showing an example of the overall configuration of recommendation system 1. A block diagram showing an example of the configuration of the integration device 10. A diagram showing an example of the configuration of the advertising database MDB. A diagram showing a first example of knowledge graph IG2. A diagram showing a second example of knowledge graph IG2. A diagram showing a first example of knowledge graph IG1. A diagram showing a second example of knowledge graph IG1. A diagram showing an example of knowledge graph IG1 and knowledge graph IG2 integrated by the integration unit 117. A diagram showing an example of knowledge graph IG1 and knowledge graph IG2 before integration by the integration unit 117. A diagram showing an example of j first similarity values ​​extracted by the integration unit 117. A diagram showing an example of j products determined by the integration unit 117. A diagram showing an example of knowledge graph IG1 and knowledge graph IG2 after integration by the integration unit 117. A block diagram showing an example of the configuration of recommendation device 20. A flowchart showing an example of the operation of integration device 10. A flowchart showing an example of the operation of integration device 10. A flowchart showing an example of the operation of recommendation device 20.

[0012] The recommendation system 1 according to this embodiment will be described below with reference to Figures 1 to 16.

[0013] 1: Configuration of the Embodiment 1-1: Overall Configuration Diagram 1 is a block diagram showing an example of the overall configuration of the recommendation system 1 according to this embodiment. As shown in Figure 1, the recommendation system 1 comprises an integration device 10, a recommendation device 20, and terminals 30[1] to 30[n]. The integration device 10, the recommendation device 20, and terminals 30[1] to 30[n] are connected to each other via a communication network NET so that they can communicate with one another. n is an integer of 1 or more.

[0014] In the following explanation, terminals 30[1] to 30[n] may be collectively referred to as "terminal 30". In Figure 1, user U uses terminal 30. Also, user U[1] uses terminal 30[1]. User U[2] uses terminal 30[2]. User U[k] uses terminal 30[k]. User U[n] uses terminal 30[n]. k is an integer between 1 and n, inclusive. In the following explanation, user U[k] may be used as a representative example of user U, and terminal 30[k] may be used as a representative example of terminal 30.

[0015] Note that in Figure 1, the number of integration devices 10 and recommendation devices 20 being one is merely an example. The recommendation system 1 comprises any number of integration devices 10 and any number of recommendation devices 20. The number of integration devices 10 and recommendation devices 20 may be the same or different.

[0016] The recommendation device 20 recommends products or services to user U[k] using terminal 30[k]. The integration device 10 integrates multiple knowledge graphs IG that the recommendation device 20 uses when recommending products or services to user U[k]. Based on the multiple knowledge graphs IG integrated by the integration device 10, the recommendation device 20 searches for advertisements to recommend products or services to user U[k]. The recommendation device 20 also provides the advertisements as a result of the search to terminal 30[k] used by user U[k].

[0017] In Figure 1, the integration device 10 and the recommendation device 20 are depicted as separate entities. However, the integration device 10 and the recommendation device 20 may be a single recommendation device 20 incorporated into the same housing. For example, the components of the integration device 10 may be incorporated inside the recommendation device 20.

[0018] Terminal 30[k] is, for example, a device used by user U[k] to purchase goods or services. An electronic payment application is installed on terminal 30[k], for example.

[0019] Terminal 30[k] is preferably a portable device. Terminal 30[k] may be, for example, a smartphone or a tablet. Alternatively, terminal 30[k] may be a PC (Personal Computer).

[0020] 1-2: Diagram 2 of the integrated device configuration is a block diagram showing an example configuration of the integrated device 10. As shown in Figure 2, the integrated device 10 comprises a processing unit 11, a storage device 12, an input device 14, and a communication device 15. Each element of the integrated device 10 is interconnected by one or more buses for communicating information.

[0021] The processing unit 11 is a processor that controls the entire integrated device 10. The processing unit 11 is configured using, for example, one or more chips. The processing unit 11 is also configured using a central processing unit (CPU) that includes, for example, interfaces with peripheral devices, arithmetic units, and registers. Some or all of the functions of the processing unit 11 may be implemented by hardware such as a DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), PLD (Programmable Logic Device), and FPGA (Field Programmable Gate Array). The processing unit 11 executes various processes in parallel or sequentially.

[0022] The storage device 12 is a recording medium that can be read from and written to by the processing device 11. The storage device 12 includes, for example, non-volatile memory and volatile memory. Non-volatile memory includes, for example, ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory), and EEPROM (Electrically Erasable Programmable Read Only Memory). Volatile memory includes, for example, RAM (Random Access Memory). The storage device 12 stores the control program PR1 executed by the processing device 11. The storage device 12 functions as a work area for the processing device 11. Furthermore, the storage device 12 stores the advertising database MDB. The advertising database MDB is a database of advertisements for goods or services that the recommendation device 20 provides to the terminal 30[k] when recommending such goods or services to the terminal 30[k].

[0023] Figure 3 shows an example of the structure of the advertising database MDB. The advertising database MDB stores advertising data MD. Each advertising data MD has a "Product / Service Name" field and a "Summary" field. The "Product / Service Name" field stores the name of the product or service. The "Summary" field stores a summary related to the advertisement for the product or service indicated by the "Product / Service Name" field.

[0024] As an example, the advertising database MDB shown in Figure 3 stores advertising data MD where the "product / service name" is "Lemino (registered trademark)" and the "summary" is the text "Lemino is a flat-rate video streaming service (internet television) provided by NTT Docomo." The advertising database MDB also stores advertising data MD where, as another example, the "product / service name" is "d Anime Store (registered trademark)" and the "summary" is the text "d Anime Store is a flat-rate video-on-demand service specializing in anime, operated by Docomo and Docomo Anime Store."

[0025] In Figure 2, the input device 14 is a device that receives operations from the administrator of the integrated device 10. For example, the input device 14 is configured to include a keyboard, touchpad, touch panel, or pointing device such as a mouse. Here, if the input device 14 is configured to include a touch panel, it may also function as a display device.

[0026] The communication device 15 is hardware acting as a transmitting and receiving device for communicating with other devices. The communication device 15 is also called, for example, a network device, a network controller, a network card, or a communication module. The communication device 15 may be equipped with a connector for wired connection and an interface circuit corresponding to the above-mentioned connector. The communication device 15 may also be equipped with a wireless communication interface. Examples of connectors and interface circuits for wired connection include products compliant with wired LAN, IEEE 1394, and USB. Examples of wireless communication interfaces include products compliant with wireless LAN and Bluetooth®.

[0027] The processing unit 11 functions as a communication control unit 111, a graphing unit 112, a first acquisition unit 113, a determination unit 114, a second acquisition unit 115, a decision unit 116, and an integration unit 117, for example, by reading and executing the control program PR1 from the storage device 12.

[0028] The communication control unit 111 causes the communication device 15 to send and receive various information, data, and signals with other devices.

[0029] The graphing unit 112 uses a large-scale language model to create a knowledge graph of the summary sentences contained in the advertising data MD stored in the advertising database MDB. For example, the large-scale language model is Llama. The knowledge graph IG2 generated by the graphing unit 112, which creates a knowledge graph of the summary sentences contained in the advertising data MD, is an example of a "second knowledge graph".

[0030] Figure 4 shows a first example of a knowledge graph IG2. The knowledge graph IG2 has multiple nodes N2. Each of the multiple nodes N2 has a central node CN2 and n individual nodes IN2. That is, the knowledge graph IG2 has multiple nodes N2, each containing a central node CN2 and n individual nodes IN2. Here, n is an integer greater than or equal to 2. In the example shown in Figure 4, n = 5. The knowledge graph IG2 also has n edges ED2. Each of the n edges ED2 connects the central node CN2 to each of the n individual nodes IN2. Here, an individual node IN2 is an example of a "second node". Here, an individual node IN2 is an example of a "second individual node". The central node CN2 is an example of a "third node". Here, an edge ED2 is an example of a "second edge".

[0031] The central node CN2 corresponds to the "product / service name" included in the advertising data MD exemplified in Figure 3. The individual node IN2 corresponds to the keywords included in the text indicated by the "summary text" within the advertising data MD.

[0032] In the knowledge graph IG2 illustrated in Figure 4, the central node CN2 corresponds to the fact that the "product / service name" is "Lemino". The first individual node IN2[1] corresponds to the fact that "Lemino" is a "subscription-based video streaming service". The second individual node IN2[2] corresponds to the fact that the former name of "Lemino" was "dTV (registered trademark)". The third individual node IN2[3] corresponds to the fact that "Lemino" was launched as a "new release rental service". The fourth individual node IN2[4] corresponds to the fact that one of the genres handled by "Lemino" is "Japanese films". The fifth individual node IN2[5] corresponds to the fact that one of the genres handled by "Lemino" is "Western films".

[0033] Figure 5 shows a second example of the knowledge graph IG2. In the example shown in Figure 5, n = 4.

[0034] In the knowledge graph IG2 illustrated in Figure 5, the central node CN2 corresponds to the fact that the "product / service name" is "d Anime Store". The first individual node IN2[1] corresponds to the fact that "d Anime Store" is a "video-on-demand service". The second individual node IN2[2] corresponds to the fact that the monthly fee for "d Anime Store" is "550 yen". The third individual node IN2[3] corresponds to the fact that "d Anime Store" is "specialized in anime". The fourth individual node IN2[4] corresponds to the fact that "d Anime Store" supports "download playback".

[0035] In Figure 2, the first acquisition unit 113 acquires a general knowledge graph from an external device of the integration device 10. The general knowledge graph is a knowledge graph IG1 based on a summary text about general knowledge. Knowledge graph IG1 is an example of a "first knowledge graph".

[0036] Figure 6 shows a first example of knowledge graph IG1. Knowledge graph IG1 has multiple nodes N1. Each of the multiple nodes N1 has m individual nodes IN1. That is, knowledge graph IG1 has multiple nodes N1 containing m individual nodes IN1. Here, m is an integer greater than or equal to 2. In the example shown in Figure 6, m = 3. Knowledge graph IG2 also has one or more edges ED1. Each of the one or more edges ED1 connects two individual nodes IN1 from the m individual nodes IN1. Here, individual node IN1 is an example of a "first node". Here, individual node IN1 is an example of a "first individual node". Here, edge ED1 is an example of a "first edge".

[0037] In the knowledge graph IG1 illustrated in Figure 6, the first individual node IN1[1] corresponds to the fact that knowledge graph IG1 is knowledge graph IG about “Western films”. The second individual node IN1[2] corresponds to the fact that “Western films” are “produced in Europe or America”. The third individual node IN1[3] corresponds to the fact that the market size of “Western films” is “66.5 billion”.

[0038] Figure 7 shows a second example of knowledge graph IG1. In the example shown in Figure 7, m = 5.

[0039] In the knowledge graph IG1 illustrated in Figure 7, the first individual node IN1[1] corresponds to the fact that knowledge graph IG1 is knowledge graph IG about “anime”. The second individual node IN1[2] corresponds to the fact that “anime” is “composed using animation”. The third individual node IN1[3] corresponds to the fact that “anime” is an abbreviation for “animation”. The fourth individual node IN1[4] corresponds to the fact that “animation” is also called “movie”. The fifth individual node IN1[5] corresponds to the fact that the term “movie” has come into use “since the 2000s”.

[0040] Furthermore, in Figure 2, the first acquisition unit 113 acquires knowledge graph IG2 generated by the graphing unit 112 along with knowledge graph IG1.

[0041] The determination unit 114 determines whether or not identical nodes exist between multiple individual nodes IN1 included in the knowledge graph IG1 acquired by the first acquisition unit 113 and multiple individual nodes IN2 included in the knowledge graph IG2 acquired by the first acquisition unit 113.

[0042] For example, if the knowledge graph IG1 acquired by the first acquisition unit 113 is the knowledge graph IG1 exemplified in Figure 6, and the knowledge graph IG2 acquired by the first acquisition unit 113 is the knowledge graph IG2 exemplified in Figure 4, then the first individual node IN1[1] of knowledge graph IG1 and the fifth individual node IN2[5] of knowledge graph IG2 are the same node because they are both "Western films". In this case, the determination unit 114 determines that there are identical nodes between the multiple individual nodes IN1 included in the knowledge graph IG1 acquired by the first acquisition unit 113 and the multiple individual nodes IN2 included in the knowledge graph IG2 acquired by the first acquisition unit 113.

[0043] On the other hand, for example, when the knowledge graph IG1 acquired by the first acquisition unit 113 is the knowledge graph IG1 illustrated in FIG. 7, and the knowledge graph IG2 acquired by the first acquisition unit 113 is the knowledge graph IG2 illustrated in FIG. 5, there are no identical nodes between the plurality of individual nodes IN1 included in the knowledge graph IG1 and the plurality of individual nodes IN2 included in the knowledge graph IG2. In this case, the determination unit 114 determines that there are no identical nodes between the plurality of individual nodes IN1 included in the knowledge graph IG1 acquired by the first acquisition unit 113 and the plurality of individual nodes IN2 included in the knowledge graph IG2 acquired by the first acquisition unit 113.

[0044] In FIG. 2, when the determination unit 114 determines that there are no identical nodes between the plurality of individual nodes IN1 included in the knowledge graph IG1 acquired by the first acquisition unit 113 and the plurality of individual nodes IN2 included in the knowledge graph IG acquired by the first acquisition unit 113, the second acquisition unit 115 vectorizes all the nodes of the knowledge graph IG1 and the knowledge graph IG2 using a large language model. As an example, the large language model is BERT. The above-mentioned all nodes include a plurality of individual nodes IN1, a plurality of individual nodes IN2, and a central node CN2. That is, the second acquisition unit 115 vectorizes the plurality of individual nodes IN1, the plurality of individual nodes IN2, and the central node CN2.

[0045] By vectorizing the plurality of individual nodes IN1, the plurality of individual nodes IN2, and the central node CN2, the second acquisition unit 115 acquires vectors corresponding one-to-one to the plurality of individual nodes IN1, vectors corresponding one-to-one to the plurality of individual nodes IN2, and a vector corresponding to the central node CN2. Here, the vector corresponding one-to-one to the plurality of individual nodes IN1 is an example of the "first vector". The vector corresponding one-to-one to the plurality of individual nodes IN2 is an example of the "second vector". The vector corresponding to the central node CN2 is an example of the "third vector".

[0046] When there are m individual nodes IN1, the second acquisition unit 115 acquires m vectors corresponding one-to-one to the m individual nodes IN1. When there are n individual nodes IN2, the second acquisition unit 115 acquires n vectors corresponding one-to-one to the n individual nodes IN2.

[0047] The determination unit 116 determines, by exhaustive comparison, the similarity between each of the plurality of individual nodes IN1 and each of the plurality of individual nodes IN2 based on the plurality of vectors corresponding one-to-one to the plurality of individual nodes IN1 and the plurality of vectors corresponding one-to-one to the plurality of individual nodes IN2. As an example, when the number of individual nodes IN1 is m and the number of individual nodes IN2 is n, the determination unit 116 determines m×n similarities based on the m vectors corresponding one-to-one to the m individual nodes IN1 and the n vectors corresponding one-to-one to the n individual nodes IN2. The similarity is, for example, the cosine similarity between the vector corresponding to the individual node IN1 and the vector of the individual node IN2. Note that the similarity is an example of the "first similarity".

[0048] In addition, the determination unit 116 determines the similarity between each of at least two individual nodes IN2 among the plurality of individual nodes IN2 and the central node CN2. As an example, the determination unit 116 determines k similarities based on the k vectors corresponding one-to-one to the k individual nodes IN2 and the vector corresponding to the central node CN2. Here, k is an integer of 2 or more and n or less. The similarity is, for example, the cosine similarity between the vector of the individual node IN2 and the vector of the central node CN2. Note that the similarity is an example of the "second similarity".

[0049] The integration unit 117 integrates the knowledge graph IG1 and the knowledge graph IG2 acquired by the first acquisition unit 113.

[0050] As described above, if the determination unit 114 determines that there are identical nodes between multiple individual nodes IN1 included in knowledge graph IG1 and multiple individual nodes IN2 included in knowledge graph IG2, the integration unit 117 integrates knowledge graph IG1 and knowledge graph IG2 by integrating the identical nodes included in knowledge graph IG1 and the identical nodes included in knowledge graph IG2.

[0051] If the number of individual nodes IN1 in knowledge graph IG1 is m, and the number of individual nodes IN2 in knowledge graph IG2 is n, and one individual node IN1 in the m individual nodes IN1 and one individual node IN2 in the n individual nodes IN2 are the same node, the integration unit 117 integrates knowledge graph IG1 and knowledge graph IG2 by integrating the one individual node IN1 and the one individual node IN2. Here, one individual node IN1 in the m individual nodes IN1 is an example of a "fourth node". One individual node IN2 in the n individual nodes IN2 is an example of a "fifth node". Note that the number of "fourth nodes" in the m individual nodes IN1 and the number of "fifth nodes" in the n individual nodes IN2 are not limited to one. Furthermore, it is preferable that the number of "fourth nodes" and the number of "fifth nodes" are the same. However, the number of "fourth nodes" and the number of "fifth nodes" do not have to be the same.

[0052] Figure 8 shows an example of knowledge graph IG1 and knowledge graph IG2 integrated by the integration unit 117. As described above, when knowledge graph IG1 is the knowledge graph IG1 exemplified in Figure 6 and knowledge graph IG2 is the knowledge graph IG2 exemplified in Figure 4, the first individual node IN1[1] included in knowledge graph IG1 and the fifth individual node IN2[5] included in knowledge graph IG2 are the same node. Therefore, the integration unit 117 integrates knowledge graph IG1 and knowledge graph IG2 by integrating the first individual node IN1[1] and the fifth individual node IN2[5]. Here, the first individual node IN1[1] is an example of a "fourth node". The fifth individual node IN2[5] is an example of a "fifth node".

[0053] On the other hand, if the determination unit 114 determines that there are no identical nodes between the multiple individual nodes IN1 included in the knowledge graph IG1 and the multiple individual nodes IN2 included in the knowledge graph IG2, the integration unit 117 integrates the knowledge graph IG1 and the knowledge graph IG2 by connecting the individual nodes IN1 and IN2 with a connecting edge CE.

[0054] In this case, the integration unit 117 determines individual nodes IN1 and IN2 to be joined at the joining edge CE based on the multiple similarities between each of the multiple nodes N1 of the knowledge graph IG1 and each of the multiple nodes N2 of the knowledge graph IG2, and the multiple similarities between the multiple nodes N2 of the knowledge graph IG2. Here, the multiple similarities between each of the multiple nodes N1 of the knowledge graph IG1 and each of the multiple nodes N2 of the knowledge graph IG2 are examples of the "first similarity" described above. The multiple similarities between the multiple nodes N2 of the knowledge graph IG2 are examples of the "second similarity" described above.

[0055] Specifically, if knowledge graph IG1 contains m individual nodes IN1 and knowledge graph IG2 contains n individual nodes IN2, the integration unit 117 determines which individual nodes IN1 and IN2 to be joined by a joining edge CE, based on m × n first similarities between each of the m individual nodes IN1 and each of the n individual nodes IN2, and k second similarities between each of the k individual nodes IN2 and the central node CN2.

[0056] As an example, the integration unit 117 extracts j first similarity values ​​from the above m × n first similarity values ​​in descending order, along with the combinations of individual nodes IN1 and individual nodes IN2 corresponding to each first similarity value. Here, j is an integer between 2 and m × n, inclusive. The integration unit 117 also determines multiple individual nodes IN2 from the n individual nodes IN2 that correspond to the j first similarity values, as the above k individual nodes IN2. In this case, k = j.

[0057] Subsequently, the integration unit 117 determines the j products by calculating the product of one first similarity out of the j first similarities and one second similarity out of k=j second similarities for each of the j first similarities. These j products represent the "finally calculated similarity". Based on the j products, the integration unit 117 determines the individual nodes IN1 and IN2 that will be joined at the joining edge CE. As an example, the integration unit 117 determines the individual nodes IN1 and IN2 that correspond to the product with the highest value among the j products as the individual nodes IN1 and IN2 that will be joined at the joining edge CE.

[0058] Here, the "single first similarity" mentioned above is the similarity between one individual node IN1 included in m individual nodes IN1 and one individual node IN2 included in k individual nodes IN2. The "single second similarity" mentioned above is the similarity corresponding to the said "single individual node IN2".

[0059] Figures 9 to 12 show an example of how the integration unit 117 integrates knowledge graph IG1 and knowledge graph IG2.

[0060] Figure 9 shows an example of knowledge graph IG1 and knowledge graph IG2 before integration by the integration unit 117. In Figure 9, knowledge graph IG1 is the knowledge graph IG1 exemplified in Figure 7. Knowledge graph IG2 is the knowledge graph IG2 exemplified in Figure 5. As described above, there are no identical nodes between knowledge graph IG1 exemplified in Figure 7 and knowledge graph IG2 exemplified in Figure 5. As a result, as shown in Figure 9, no integration of identical nodes occurs between knowledge graph IG1 and knowledge graph IG2.

[0061] Figure 10 shows an example of j first similarity values ​​extracted by the integration unit 117. As described above, the integration unit 117 extracts j first similarity values ​​in descending order of value from the m × n first similarity values ​​between each of the m individual nodes IN1 and each of the n individual nodes IN2, along with the combination of individual nodes IN1 and individual nodes IN2 corresponding to the first similarity value. In the example shown in Figure 10, the first similarity value with the largest value among the j first similarity values ​​is the similarity between the 5th individual node IN1[5] and the 2nd individual node IN2[2]. The second largest first similarity value is the similarity between the 1st individual node IN1[1] and the 3rd individual node IN2[3]. The jth largest first similarity value is the similarity between the 4th individual node IN1[4] and the 1st individual node IN2[1].

[0062] Figure 11 shows an example of j products determined by the integration unit 117. The integration unit 117 multiplies the similarity between the fifth individual node IN1[5] and the second individual node IN2[2], which is the first similarity with the largest value, by the second similarity between the second individual node IN2[2] and the central node CN2. The integration unit 117 also multiplies the similarity between the first individual node IN1[1] and the third individual node IN2[3], which is the second largest first similarity, by the second similarity between the third individual node IN2[3] and the central node CN2. The integration unit 117 also multiplies the similarity between the fourth individual node IN1[4] and the first individual node IN2[1], which is the jth largest first similarity, by the second similarity between the first individual node IN2[1] and the central node CN2.

[0063] Furthermore, if the first similarity is cosine similarity C1 and the second similarity is cosine similarity C2, the integration unit 117 preferably converts the cosine similarity C2 to a similarity SC2 scaled to a value between 0 and 1 using the following formula (1), and then calculates the product of cosine similarity C1 and similarity SC2. SC2 = (C2 + 1) / 2 (1)

[0064] The integration unit 117 increases the similarity between each of the multiple individual node IN1s and each of the multiple individual node IN2s, specifically between the individual node IN1s and the individual node IN2s whose names and meanings are similar to those of the goods or services, by multiplying the first similarity by the second similarity. In other words, it shortens the distance between the individual node IN1s and individual node IN2s corresponding to that similarity.

[0065] In the following discussion, we will assume that the second product, i.e., the case where the value obtained by multiplying the similarity between the first individual node IN1[1] and the third individual node IN2[3] by the similarity between the third individual node IN2[3] and the central node CN2 is the largest among the j products exemplified in Figure 11, is the second product.

[0066] Figure 12 shows an example of knowledge graphs IG1 and IG2 after integration by the integration unit 117. As described above, as an example, the value obtained by multiplying the similarity between the first individual node IN1[1] and the third individual node IN2[3] by the similarity between the third individual node IN2[3] and the central node CN2 was the largest. Therefore, the integration unit 117 integrates knowledge graphs IG1 and IG2 by joining individual nodes IN1[1] and IN2[3] with a joining edge CE.

[0067] 1-3: Recommendation Device Configuration Diagram 13 is a block diagram showing an example configuration of the recommendation device 20. As shown in Figure 13, the recommendation device 20 comprises a processing device 21, a storage device 22, an input device 24, and a communication device 25. Each element of the recommendation device 20 is interconnected by one or more buses for communicating information.

[0068] The processing unit 21 is a processor that controls the entire recommendation device 20. The processing unit 21 is configured using, for example, one or more chips. The processing unit 21 is also configured using, for example, a central processing unit (CPU) that includes interfaces with peripheral devices, an arithmetic unit, and registers. Some or all of the functions of the processing unit 21 may be implemented by hardware such as a DSP, ASIC, PLD, and FPGA. The processing unit 21 executes various processes in parallel or sequentially.

[0069] The storage device 22 is a recording medium that can be read from and written to by the processing device 21. The storage device 22 includes, for example, non-volatile memory and volatile memory. Non-volatile memory is, for example, ROM, EPROM, and EEPROM. Volatile memory is, for example, RAM. The storage device 22 stores the control program PR2 executed by the processing device 21. The storage device 22 functions as a work area for the processing device 21. Furthermore, the storage device 22 stores the user database UDB. The user database UDB stores information indicating the episodic memory of each user U[1] to user U[n]. Here, "episodic memory" means memories of events experienced by users U[1] to user U[n] as individuals. The information indicating episodic memories stored in the user database (UDB) preferably includes not only information describing the content of an event, but also accompanying information such as the time and space in which the event occurred, the physical state of users U[1] to U[n], and the psychological state of users U[1] to U[n].

[0070] The input device 24 is a device that receives operations from the administrator of the recommendation device 20. For example, the input device 24 is configured to include a keyboard, touchpad, touch panel, or pointing device such as a mouse. If the input device 24 is configured to include a touch panel, it may also function as a display device.

[0071] The communication device 25 is hardware acting as a transmitting and receiving device for communicating with other devices. The communication device 25 is also called, for example, a network device, network controller, network card, or communication module. The communication device 25 may be equipped with a connector for wired connection and an interface circuit corresponding to the above-mentioned connector. The communication device 25 may also be equipped with a wireless communication interface. Examples of connectors and interface circuits for wired connection include products compliant with wired LAN, IEEE 1394, and USB. Examples of wireless communication interfaces include products compliant with wireless LAN and Bluetooth®.

[0072] The processing unit 21 functions as a communication control unit 211, a first acquisition unit 212, a second acquisition unit 213, a determination unit 214, a search unit 215, and a recommendation unit 216, for example, by reading and executing the control program PR2 from the storage device 22.

[0073] The communication control unit 211 causes the communication device 25 to send and receive various information, data, and signals with other devices.

[0074] The first acquisition unit 212 acquires information indicating the utterances of user U[k] from the terminal 30[k] used by user U[k]. For example, user U[k] uses terminal 30[k] to send and receive emails and chat with other users U. Another example is that user U[k] uses terminal 30[k] to post text on websites on the internet, including social networking services (SNS). The first acquisition unit 212 acquires these emails, chats, and posts on websites by user U[k] from terminal 30[k] as information indicating the utterances of user U[k].

[0075] Furthermore, the first acquisition unit 212 uses information indicating the utterance of user U[k] acquired from terminal 30[k] to refer to information indicating the episodic memory of user U[k] stored in the user database UDB, thereby acquiring information indicating past utterances related to user U[k]'s utterance.

[0076] The second acquisition unit 213 acquires knowledge graph IG1 and knowledge graph IG2 from the integration device 10 in an integrated state.

[0077] The decision unit 214 determines, based on the user U[k]'s actions, an individual node IN1 from among a plurality of individual nodes IN1 included in the knowledge graph IG1 that is presumed to be of interest to user U[k]. For example, the decision unit 214 determines an individual node IN1 that is presumed to be of interest to user U[k] based on the utterance of user U[k] acquired by the first acquisition unit 212 and past utterances of user U[k] related to the said utterance.

[0078] The search unit 215 searches the knowledge graph IG2 starting from the individual node IN1 determined by the decision unit 214. For example, if there are multiple knowledge graphs IG2 acquired by the second acquisition unit 213, the search unit 215 searches for a knowledge graph IG2 that includes the same node as the individual node IN1 determined by the decision unit 214, or a knowledge graph IG2 that includes an individual node IN2 connected to the individual node IN1 by a connecting edge CE. In this case, as an example, the search unit 215 searches for the individual node IN2 that is most similar to the utterance of user U[k] from among the multiple individual nodes IN2 included in the knowledge graph IG2 being searched.

[0079] The recommendation unit 216 recommends a product or service to user U[k] based on the search results of the search unit 215. For example, the recommendation unit 216 recommends a product or service to user U[k] that corresponds to the central node CN2 included in the knowledge graph IG2 that is being searched. When the recommendation unit 216 recommends such a product or service, for example, it recommends the product or service to user U[k] along with an advertisement summary text that corresponds to the individual node IN2 which was the search result of the search unit 215. Specifically, the recommendation unit 216 causes the communication device 25 to transmit information indicating the product or service and information indicating the advertisement summary text to terminal 30[k].

[0080] 2: Operation of the Embodiment 2-1: Operation of the Integrated Device Figures 14 and 15 are flowcharts showing examples of operation of the integrated device 10.

[0081] In step S1 of Figure 14, the processing unit 11 in the integration device 10 functions as a graphing unit 112. The processing unit 11 uses a large-scale language model to create a knowledge graph of the summary sentences contained in the advertising data MD stored in the advertising database MDB. As a result, the processing unit 11 generates a knowledge graph IG2.

[0082] In step S2, the processing unit 11 functions as a first acquisition unit 113. The processing unit 11 acquires the knowledge graph IG1, which is a general knowledge graph, from an external device of the integration device 10. The processing unit 11 also acquires the knowledge graph IG2 generated in step S1.

[0083] In step S3, the processing unit 11 functions as a determination unit 114. The processing unit 11 determines whether there are identical nodes between a plurality of individual nodes IN1 included in the knowledge graph IG1 and a plurality of individual nodes IN2 included in the knowledge graph IG2. If identical nodes exist (YES in step S3), the processing unit 11 executes the process in step S4. If identical nodes do not exist (NO in step S3), the processing unit 11 executes the process in step S5.

[0084] In step S4, the processing unit 11 functions as an integration unit 117. The processing unit 11 integrates the knowledge graph IG1 and the knowledge graph IG2 by integrating the same nodes included in the knowledge graph IG1 and the same nodes included in the knowledge graph IG2.

[0085] In step S5, the processing unit 11 functions as a second acquisition unit 115. The processing unit 11 vectorizes all nodes. These nodes include a plurality of individual nodes IN1, a plurality of individual nodes IN2, and a central node CN2. By vectorizing the plurality of individual nodes IN1, a plurality of individual nodes IN2, and a central node CN2, the processing unit 11 obtains vectors that correspond one-to-one with the plurality of individual nodes IN1, vectors that correspond one-to-one with the plurality of individual nodes IN2, and vectors that correspond to the central node CN2.

[0086] In step S6, the processing unit 11 functions as an integration unit 117. The processing unit 11 integrates knowledge graph IG1 and knowledge graph IG2 by joining together nodes with high similarity scores among combinations of one individual node IN1 from a plurality of individual nodes IN1 and one individual node IN2 from a plurality of individual nodes IN2 using a joining edge CE.

[0087] Figure 15 is a flowchart illustrating an example of a substep that constitutes step S6.

[0088] In substep S6(1), the processing unit 11 functions as a determination unit 116. Based on m vectors that correspond one-to-one to m individual nodes IN1 and n vectors that correspond one-to-one to n individual nodes IN2, the processing unit 11 determines m × n first similarity values ​​between each of the multiple individual nodes IN1 and each of the multiple individual nodes IN2 in a brute-force manner.

[0089] In substep S6(2), the processing unit 11 functions as an integration unit 117. The processing unit 11 extracts j first similarity values ​​from the m × n first similarity values ​​in descending order of value, along with the combination of individual node IN1 and individual node IN2 corresponding to the first similarity value.

[0090] In substep S6(3), the processing unit 11 functions as a determination unit 116. The processing unit 11 determines a second similarity between each of the j individual nodes IN2 and the central node CN2.

[0091] In substep S6(4), the processing unit 11 functions as an integration unit 117. The processing unit 11 determines the j products by calculating the product of one first similarity out of the j first similarities and one second similarity out of the j second similarities for each of the j first similarities. Here, the "one first similarity" is the similarity between one individual node IN1 included in m individual nodes IN1 and one individual node IN2 included in k individual nodes IN2. The "one second similarity" is the similarity corresponding to the "one individual node IN2".

[0092] In substep S6(5), the processing unit 11 functions as an integration unit 117. The processing unit 11 determines that the individual nodes IN1 and IN2 corresponding to the product with the highest value among the j products determined in substep S6(4) are to be joined by the joining edge CE. That is, the processing unit 11 determines that the individual nodes IN1 and IN2 with the highest calculated similarity are to be joined by the joining edge CE. Furthermore, the processing unit 11 integrates the knowledge graph IG1 and the knowledge graph IG2 by joining the determined individual nodes IN1 and IN2 with the joining edge CE.

[0093] 2-2: Recommendation device operation diagram 16 is a flowchart illustrating an example of the operation of the recommendation device 20.

[0094] In step S11, the processing unit 21 of the recommendation device 20 functions as the first acquisition unit 212. The processing unit 21 acquires information indicating the utterance of user U[k] from the terminal 30[k] used by user U[k].

[0095] In step S12, the processing unit 21 functions as a first acquisition unit 212. The processing unit 21 uses information indicating the utterance of user U[k] acquired from terminal 30[k] to refer to information indicating the episodic memory of user U[k] stored in the user database UDB, thereby acquiring information indicating past utterances related to user U[k]'s utterance.

[0096] In step S13, the processing unit 21 functions as a second acquisition unit 213. The processing unit 21 acquires the knowledge graph IG1 and the knowledge graph IG2 from the integration device 10 in an integrated state.

[0097] In step S14, the processing unit 21 functions as a determination unit 214. Based on the utterance of user U[k] acquired in step S11 and the past utterances of user U[k] acquired in step S12, the processing unit 21 determines an individual node IN1 that is presumed to be of interest to user U[k].

[0098] In step S15, the processing unit 21 functions as a search unit 215. The processing unit 21 searches the knowledge graph IG2 starting from the individual node IN1 determined in step S13.

[0099] In step S16, the processing unit 11 functions as a recommendation unit 216. Based on the search results in step S15, the processing unit 11 recommends a product or service to user U[k].

[0100] 3: Effects of the Embodiment The knowledge graph IG integration device 10 according to this embodiment comprises a first acquisition unit 113 and an integration unit 117. The first acquisition unit 113 acquires a knowledge graph IG1 as a first knowledge graph having a plurality of nodes N1 and a knowledge graph IG2 as a second knowledge graph having a plurality of nodes N2. The integration unit 117 integrates the knowledge graph IG1 and the knowledge graph IG2 by joining an individual node IN1 as a first individual node included in the plurality of nodes N1 of the knowledge graph IG1 and an individual node IN2 as a second individual node included in the plurality of nodes N2 of the knowledge graph IG2 with a joining edge CE. The integration unit 117 determines the individual node IN1 and the individual node IN2 based on a plurality of first similarities between each of the plurality of nodes N1 of the knowledge graph IG1 and each of the plurality of nodes N2 of the knowledge graph IG2, and a plurality of second similarities between the plurality of nodes N2 of the knowledge graph IG2.

[0101] The knowledge graph IG integration device 10, with the above configuration, can more effectively recommend products or services, for example, when using a graph that integrates multiple knowledge graphs IG to recommend products or services.

[0102] More specifically, the knowledge graph IG integration device 10 uses both the first and second similarity measures to increase the similarity between each of the multiple individual node IN1 and each of the multiple individual node IN2, specifically between each of the multiple individual node IN1 and the individual node IN2 whose name and meaning are similar to those of the product or service. As a result, the knowledge graph IG integration device 10 becomes able to recommend products or services more effectively.

[0103] Furthermore, in the knowledge graph IG integration device 10 according to this embodiment, the plurality of nodes N1 of the knowledge graph IG1 as the first knowledge graph include m (where m is an integer of 2 or more) individual nodes IN1 as first nodes. The knowledge graph IG1 includes m individual nodes IN1 and one or more edges ED1 as first edges. The plurality of nodes N2 of the knowledge graph IG2 as the second knowledge graph include n (where n is an integer of 2 or more) individual nodes IN2 as second nodes. The knowledge graph IG2 includes n individual nodes IN2, a central node CN2 as a third node, and n edges ED2 as second edges that connect the central node CN2 and each of the n individual nodes IN2. The integration unit 117 determines the individual nodes IN1 and IN2 based on m × n first similarity scores between each of the m individual nodes IN1 and each of the n individual nodes IN2, and k second similarity scores between each of the k (where k is an integer between 2 and n) individual nodes IN2 out of the n individual nodes IN2 and the central node CN2.

[0104] The knowledge graph IG integration device 10, with the above configuration, makes it possible to use a knowledge graph IG2 in which a central node CN2 corresponding to the name of a product or service is at the center, and multiple individual nodes IN2 corresponding to the summary text of an advertisement are arranged radially.

[0105] Furthermore, in the knowledge graph IG integration device 10 according to this embodiment, the integration unit 117 extracts j (where j is an integer of 2 or more) first similarity values ​​from the m × n first similarity values ​​in descending order of value. The integration unit 117 also determines a plurality of individual nodes IN2 from the n individual nodes IN2 as second nodes, which correspond to the j first similarity values, as the k individual nodes IN2 mentioned above.

[0106] The knowledge graph IG integration device 10, with the above configuration, can reduce the load on the processing device 11 by extracting j first similarity values ​​from m × n first similarity values ​​in descending order of value.

[0107] Furthermore, in the knowledge graph IG integration device 10 according to this embodiment, the integration unit 117 determines the j products by calculating the product of one first similarity out of j first similarities and one second similarity out of k second similarities for each of the j first similarities. The integration unit 117 also determines the individual node IN1 and the individual node IN2 based on the j products. The "one first similarity" is the similarity between one individual node IN1 included in m individual nodes IN1 and one individual node IN2 included in k individual nodes IN2. The "one second similarity" is the similarity corresponding to the "one individual node IN2".

[0108] The knowledge graph IG integration device 10, with the above configuration, increases the similarity between each of the multiple individual nodes IN1 and each of the multiple individual nodes IN2, based on the product of the first similarity and the second similarity, among the combinations of each of the multiple individual nodes IN1 and each of the multiple individual nodes IN2, specifically between each of the multiple individual nodes IN1 and the individual node IN2 that has a similar name and meaning to the product or service. As a result, the knowledge graph IG integration device 10 becomes able to recommend products or services more effectively.

[0109] Furthermore, the knowledge graph IG integration device 10 according to this embodiment further comprises a second acquisition unit 115. The second acquisition unit 115 uses a large-scale language model to vectorize the m individual nodes IN1, the n individual nodes IN2, and the central node CN2, thereby acquiring m first vectors corresponding one-to-one to the m individual nodes IN1, n second vectors corresponding one-to-one to the n individual nodes IN2, and a third vector corresponding to the central node CN2. The integration unit 117 determines m × n first similarity scores based on the m first vectors and the n second vectors. The integration unit 117 also determines k second similarity scores based on each of the k second vectors that correspond one-to-one to the k individual nodes IN2 from among the n second vectors, and the third vector.

[0110] With the above configuration, the knowledge graph IG integration device 10 can determine the first and second similarity based on the cosine similarity between the vectors corresponding to each node.

[0111] Furthermore, in the knowledge graph IG integration device 10 according to this embodiment, if the fourth node included in m individual nodes IN1 and the fifth node included in n individual nodes IN2 are the same node, the integration unit 117 integrates the knowledge graph IG1 as the first knowledge graph and the knowledge graph IG2 as the second knowledge graph by integrating the fourth node and the fifth node. Also, if the m individual nodes IN1 and the n individual nodes IN2 do not include the same node, the integration unit 117 integrates the knowledge graph IG1 and the knowledge graph IG2 by connecting the individual nodes IN1 and IN2 with a coupling edge CE.

[0112] The knowledge graph IG integration device 10, having the above configuration, can integrate knowledge graph IG1 and knowledge graph IG2 by integrating the identical nodes present in both knowledge graph IG1 and knowledge graph IG2.

[0113] Furthermore, in the knowledge graph IG integration device 10 according to this embodiment, the knowledge graph IG1, as the first knowledge graph, is a graph based on a first summary sentence relating to general knowledge. The knowledge graph IG2, as the second knowledge graph, is a graph based on a second summary sentence relating to an advertisement for a product or service.

[0114] The knowledge graph IG integration device 10, with the above configuration, can recommend products or services more effectively by integrating the knowledge graph IG1 relating to general knowledge and the knowledge graph IG2 relating to products or advertisements.

[0115] Furthermore, the recommendation device 20 according to this embodiment includes a first acquisition unit 113, an integration unit 117, a decision unit 214, a search unit 215, and a recommendation unit 216. The first acquisition unit 113 acquires a knowledge graph IG1 as a first knowledge graph having a plurality of nodes N1, and a knowledge graph IG2 as a second knowledge graph having a plurality of nodes N2. The integration unit 117 integrates the knowledge graph IG1 and the knowledge graph IG2 by connecting an individual node IN1 as a first individual node included in the plurality of nodes N1 of the knowledge graph IG1 and an individual node IN2 as a second individual node included in the plurality of nodes N2 of the knowledge graph IG2 with a connecting edge CE. The integration unit 117 determines the individual nodes IN1 and IN2 based on a plurality of first similarities between each of the plurality of nodes N1 in knowledge graph IG1 and each of the plurality of nodes N2 in knowledge graph IG2, and a plurality of second similarities between the plurality of nodes N2 in knowledge graph IG2. Knowledge graph IG1, as the first knowledge graph, is a graph based on a first summary sentence relating to general knowledge. Knowledge graph IG2, as the second knowledge graph, is a graph based on a second summary sentence relating to an advertisement for a product or service. The decision unit 214 determines an individual node IN1 from among the plurality of individual nodes IN1 in knowledge graph IG1 that is presumed to be of interest to user U[k], based on information regarding user U[k]'s behavior. The search unit 215 uses knowledge graph IG1 and knowledge graph IG2 integrated by the integration unit 117 to search knowledge graph IG2, starting from the individual node IN1 determined by the decision unit 214. The recommendation unit 216 recommends products or services to user U[k] based on the search results of the search unit 215.

[0116] By having the above configuration, the recommendation device 20 can more effectively recommend products or services, for example, when using a graph in which multiple knowledge graphs IG are integrated to recommend products or services.

[0117] More specifically, the recommendation device 20 uses both the first and second similarity measures to increase the similarity between each of the multiple individual node IN1 and each of the multiple individual node IN2, specifically between each of the multiple individual node IN1 and the individual node IN2 whose name and meaning are similar to those of the product or service. As a result, the recommendation device 20 becomes able to recommend products or services more effectively.

[0118] 3. Modifications The present disclosure is not limited to the embodiments illustrated above. Specific examples of modifications are given below.

[0119] 3-1: Modified Example 1 In the above embodiment, the integration unit 117 extracted j first similarities from the m × n first similarities in descending order of value, along with the combination of individual node IN1 and individual node IN2 corresponding to each first similarity. Subsequently, the integration unit 117 determined j products by calculating the product of one first similarity from the j first similarities and one second similarity from the j second similarities for each of the j first similarities. Furthermore, the integration unit 117 determined the individual node IN1 and individual node IN2 corresponding to the product with the highest value among the j products to be joined by the joining edge CE.

[0120] However, the integration unit 117 does not have to extract some of the first similarities from the m × n first similarities along with the combination of individual nodes IN1 and individual nodes IN2 corresponding to those first similarities. Specifically, the integration unit 117 may determine individual nodes IN1 and individual nodes IN2 that maximize the product of the m × n first similarities and the n second similarities, and these will be joined by a joining edge CE.

[0121] 3-2: Modification 2 In the above embodiment, as illustrated in Figure 1, the integration device 10 and the recommendation device 20 are connected to each other so as to be able to communicate with each other via a communication network NET. However, the integration device 10 and the recommendation device 20 may be connected directly without going through a communication network NET.

[0122] 4. Other (1) In the embodiments described above, ROM and RAM were given as examples for the storage device 12 and storage device 22, but other suitable storage media include flexible disks, magneto-optical disks (e.g., compact disks, digital multipurpose disks, Blu-ray® disks), smart cards, flash memory devices (e.g., cards, sticks, key drives), CD-ROMs (Compact Disc-ROMs), registers, removable disks, hard disks, floppy® disks, magnetic strips, databases, servers, and other appropriate storage media.

[0123] (2) In the embodiments described above, the information, signals, etc. may be represented using any of the various different techniques. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0124] (3) In the embodiments described above, the input and output information may be stored in a specific location (e.g., memory) or managed using a management table. The input and output information may be overwritten, updated, or appended to. The output information may be deleted. The input information may be transmitted to other devices.

[0125] (4) In the embodiments described above, the determination may be made by a value represented using one bit (0 or 1), by a boolean value (true or false), or by a numerical comparison (for example, a comparison with a predetermined value).

[0126] (5) The processing procedures, sequences, flowcharts, etc., exemplified in the embodiments described above may be rearranged in order, as long as there is no contradiction. For example, the methods described in this disclosure present various step elements using an exemplary order and are not limited to the specific order presented.

[0127] (6) Each function illustrated in Figures 1 to 16 is realized by any combination of at least one of hardware and software. Furthermore, the method of realizing each function block is not particularly limited. That is, each function block may be realized using one device that is physically or logically coupled, or it may be realized using two or more physically or logically separated devices that are directly or indirectly connected (for example, using wired or wireless connections). A function block may also be realized by combining the above one device or the above multiple devices with software.

[0128] (7) The programs illustrated in the embodiments described above should be broadly interpreted to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, functions, etc., whether they are called software, firmware, middleware, microcode, hardware description languages ​​or by other names.

[0129] Furthermore, software, instructions, information, etc., may be transmitted and received via a transmission medium. For example, if software is transmitted from a website, server, or other remote source using at least one of wired technology (such as coaxial cable, fiber optic cable, twisted pair, or digital subscriber line (DSL)) and wireless technology (such as infrared or microwave), then at least one of these wired and wireless technologies is included in the definition of a transmission medium.

[0130] (8) In each of the above-mentioned forms, the terms “system” and “network” shall be used interchangeably.

[0131] (9) The information, parameters, etc. described in this disclosure may be expressed using absolute values, relative values ​​from a given value, or other corresponding information.

[0132] (10) In the embodiments described above, the terminal 30 may be a mobile station (MS). A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or several other appropriate terms. In this disclosure, terms such as “mobile station,” “user terminal,” “user equipment (UE),” and “terminal” may be used interchangeably.

[0133] (11) In the embodiments described above, the terms “connected,” “coupled,” or any variation thereof, mean any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are “connected” or “coupled” with each other. The coupling or connection between elements may be a physical coupling or connection, a logical coupling or connection, or a combination thereof. For example, “connection” may be reinterpreted as “access.” As used in this disclosure, two elements may be considered to be “connected” or “coupled” with each other using at least one of one or more wires, cables and printed electrical connections, and, in some non-limiting and non-exclusive examples, electromagnetic energy having wavelengths in the radio frequency domain, microwave domain and optical (both visible and invisible) domain.

[0134] (12) In the embodiments described above, the phrase "based on" does not mean "based solely on" unless otherwise specified. In other words, the phrase "based on" means both "based solely on" and "based at least on".

[0135] (13) The terms “determining” and “determining” as used in this disclosure may encompass a wide variety of actions. “Determining” may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, or inquiring (e.g., searching in a table, database, or other data structure), or ascertaining. “Determining” may also include receiving (e.g., receiving information), transmitting (e.g., sending information), inputting, outputting, or accessing (e.g., accessing data in memory). Furthermore, "judgment" and "decision" can include considering something as having been "judged" or "decided" after resolving, selecting, choosing, establishing, comparing, etc. In other words, "judgment" and "decision" can include considering something as having been "judged" or "decided" after some action. Also, "judgment (decision)" can be reinterpreted as "assuming," "expecting," or "considering."

[0136] (14) In the embodiments described above, where “include,” “including,” and variations thereof are used, these terms are intended to be inclusive, as is the term “comprising.” Furthermore, the term “or” as used in this disclosure is not intended to be exclusive OR.

[0137] (15) In the present disclosure, if articles are added by translation, such as a, an, and the in English, the present disclosure may include the fact that the noun following these articles is plural.

[0138] (16) In this disclosure, the term “A and B are different” may mean “A and B are different from each other.” The term may also mean “A and B are each different from C.” Terms such as “separate” and “combine” may be interpreted in the same way as “different.”

[0139] (17) Each aspect / embodiment described herein may be used individually, in combination, or switched between as needed during implementation. Furthermore, notification of the specified information (e.g., notification that "it is X") is not limited to explicit notification, but may also be implicit (e.g., by not providing such notification).

[0140] Although the present disclosure has been described in detail above, it will be clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the intent and scope of the present disclosure as defined by the claims. Accordingly, the descriptions in the present disclosure are illustrative and not restrictive in any way.

[0141] 1...Recommendation system, 10...Integration device, 11...Processing device, 12...Storage device, 14...Input device, 15...Communication device, 20...Recommendation device, 21...Processing device, 22...Storage device, 24...Input device, 25...Communication device, 30...Terminal, 111...Communication control unit, 112...Graphing unit, 113...First acquisition unit, 11...Determination unit, 115...Second acquisition unit, 116...Decision unit, 117...Integration unit, 211...Communication control unit, 212...First acquisition unit, 213...Second acquisition unit, 214...Decision unit, 215...Search unit, 2 16...Recommendation section, C1...Cosine similarity, C2...Cosine similarity, CE...Joint edge, CN2...Center node, ED1...Edge, ED2...Edge, IG...Knowledge graph, IG1...Knowledge graph, IG2...Knowledge graph, IN1...Individual node, IN2...Individual node, MD...Advertising data, MDB...Advertising database, N1...Node, N2...Node, NET...Communication network, PR1...Control program, PR2...Control program, SC2...Similarity, U...User, UDB...User database

Claims

1. A knowledge graph integrator comprising: a first acquisition unit that acquires a first knowledge graph having a plurality of nodes and a second knowledge graph having a plurality of nodes; and an integration unit that integrates the first knowledge graph and the second knowledge graph by joining a first individual node included in the plurality of nodes of the first knowledge graph and a second individual node included in the plurality of nodes of the second knowledge graph with a joining edge, wherein the integration unit determines the first individual node and the second individual node based on a plurality of first similarities between each of the plurality of nodes of the first knowledge graph and each of the plurality of nodes of the second knowledge graph and a plurality of second similarities between the plurality of nodes of the second knowledge graph.

2. The knowledge graph integrator according to claim 1, wherein the plurality of nodes of the first knowledge graph include m (where m is an integer of 2 or more) first nodes, the first knowledge graph includes the m first nodes and one or more first edges, the plurality of nodes of the second knowledge graph include n (where n is an integer of 2 or more) second nodes, the second knowledge graph includes the n second nodes, a third node, and n second edges connecting each of the third node and the n second nodes, and the integrator determines the first individual node and the second individual node based on m × n first similarities between each of the m first nodes and each of the n second nodes, and k second similarities between each of the k (where k is an integer of 2 or more and n or less) second nodes out of the n second nodes and the third node.

3. The knowledge graph integration device according to claim 2, wherein the integration unit extracts j (where j is an integer of 2 or more) first similarities from the m × n first similarities in descending order of their values, and determines a plurality of second nodes from the n second nodes that correspond to the j first similarities as the k second nodes.

4. The integrating unit determines the j product by calculating the product of one first similarity among the j first similarities and one second similarity among the k second similarities for each of the j first similarities; determines the first individual node and the second individual node based on the j product; the one first similarity is the similarity between one first node included in the m first nodes and one second node included in the k second nodes; and the one second similarity is the similarity corresponding to the one second node, as described in claim 3.

5. The knowledge graph integrator according to claim 2, further comprising a second acquisition unit that uses a large-scale language model to vectorize the m first nodes, the n second nodes, and the third node to obtain m first vectors corresponding one-to-one to the m first nodes, n second vectors corresponding one-to-one to the n second nodes, and a third vector corresponding to the third node, wherein the integration unit determines the m × n first similarity scores based on the m first vectors and the n second vectors, and determines the k second similarity scores based on each of the k second vectors from the n second vectors that correspond one-to-one to the k second nodes and the third vector.

6. The knowledge graph integrating device according to claim 2, wherein the integrating unit integrates the first knowledge graph and the second knowledge graph by integrating the fourth node included in the m first nodes and the fifth node included in the n second nodes when the fourth node is the same node, and integrates the first knowledge graph and the second knowledge graph by joining the first individual node and the second individual node with the joining edge when the m first nodes and the n second nodes do not include the same node.

7. The knowledge graph integration device according to claim 1, wherein the first knowledge graph is a graph based on a first summary sentence relating to general knowledge, and the second knowledge graph is a graph based on a second summary sentence relating to an advertisement for a product or service.

8. The system comprises: a first acquisition unit that acquires a first knowledge graph having a plurality of nodes and a second knowledge graph having a plurality of nodes; an integration unit that integrates the first knowledge graph and the second knowledge graph by joining a first individual node included in the plurality of nodes of the first knowledge graph and a second individual node included in the plurality of nodes of the second knowledge graph with a joining edge; the integration unit determines the first individual node and the second individual node based on a plurality of first similarities between each of the plurality of nodes of the first knowledge graph and each of the plurality of nodes of the second knowledge graph and a plurality of second similarities between the plurality of nodes of the second knowledge graph; the first knowledge graph is a graph based on a first summary sentence relating to general knowledge; the second knowledge graph is a graph based on a second summary sentence relating to second knowledge relating to advertising for goods or services; and a determination unit that determines, based on information relating to user behavior, a node among the plurality of nodes of the first knowledge graph that is presumed to be of interest to the user. A recommendation device further comprising: a search unit that searches the second knowledge graph starting from a node determined by the decision unit, using the first knowledge graph and the second knowledge graph integrated by the integration unit; and a recommendation unit that recommends products or services to the user based on the search results of the search unit.

Images