Information processing apparatus, information processing method, and program

Abstract

To provide a technique which can improve quality of a finally generated response in retrieval-augmented generation.SOLUTION: An information processing apparatus comprises: first acquisition means for acquiring a query; first retrieval means for retrieving an initial passage related to the query from a passage set including a plurality of passages; second retrieval means for retrieving an additional passage from the passage set with reference to association information including strengths of association between passages included in the passage set and the initial passage; and third retrieval means for executing retrieval processing using the initial passage and the additional passage.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] In recent years, machine-learned language models have been utilized in retrieval processing, generation processing, etc. As an example, Non-Patent Document 1 discloses a technology related to Retrieval-Augmented Generation (RAG). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] "Reliable, Adaptable, and Attributable Language Models with Retrieval", A. Asai et al, 2024 / 3, https: / / arxiv.org / abs / 2403.03187 Summary of the Invention [Problem to be solved by the invention]

[0004] In search expansion generation (RAG), a language model is typically used to search for a text group for an input question, and the searched text group is also referenced for generation processing. In such search expansion generation, conventional techniques often have the problem that the searched text group is not self-contained, resulting in poor quality of the final generated response.

[0005] The present disclosure has been made in view of the above-mentioned problems, and an exemplary purpose thereof is to provide a technique that can improve the quality of responses that are ultimately generated in search expansion generation. [Means for solving the problem]

[0006] An information processing device according to one exemplary aspect of the present disclosure includes a first acquisition means for acquiring a query, a first search means for searching for an initial passage related to the query from a set of passages including a plurality of passages, a second search means for searching for an additional passage from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage, and a third search means for performing a search process using the initial passage and the additional passage.

[0007] An information processing device according to an exemplary aspect of the present disclosure includes an acquisition means for acquiring input data including a group of sentences, a generation means for generating a set of passages including a plurality of passages included in the group of sentences, a calculation means for calculating association information using a language model, the association information including the strength of association between a plurality of passages included in the set of passages, the association information being referenced in a search process, and a storage means for storing the association information in association with the plurality of passages.

[0008] An information processing method according to an exemplary aspect of the present disclosure includes obtaining a query, searching for an initial passage related to the query from a collection of passages including a plurality of passages, searching for an additional passage from the collection of passages by referring to the initial passage and association information including the strength of association between passages included in the collection of passages, and performing a search process using the initial passage and the additional passage.

[0009] An information processing method according to an exemplary aspect of the present disclosure includes: acquiring input data including a group of sentences; generating a set of passages including a plurality of passages included in the group of sentences; calculating association information using a language model, the association information including the strength of association between the plurality of passages included in the set of passages, the association information being referenced in a search process; and storing the association information in association with the plurality of passages.

[0010] A program according to an exemplary aspect of the present disclosure is a program that causes a computer to function as an information processing device, and the program causes the computer to execute a first acquisition process that acquires a query, a first search process that searches for an initial passage related to the query from a set of passages that includes a plurality of passages, a second search process that searches for an additional passage from the set of passages by referring to the initial passage and association information that includes the strength of association between passages included in the set of passages, and a search process using the initial passage and the additional passage.

[0011] A program according to an exemplary aspect of the present disclosure is a program that causes a computer to function as an information processing device, and the program causes the computer to execute an acquisition process that acquires input data including a group of sentences, a generation process that generates a set of passages that includes a plurality of passages included in the group of sentences, a calculation process that calculates association information using a language model, the association information including the strength of association between a plurality of passages included in the set of passages, and which is referenced in a search process, and a storage process that stores the association information in association with the plurality of passages. [Effects of the Invention]

[0012] According to an exemplary aspect of the present disclosure, an exemplary effect is achieved in that a technique can be provided that can improve the quality of the response that is ultimately generated in search expansion generation. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 2] FIG. 1 is a flow diagram showing the flow of an information processing method according to the present disclosure. [Figure 3] 1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 4] FIG. 1 is a flow diagram showing the flow of an information processing method according to the present disclosure. [Figure 5]1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 6] FIG. 1 is a flow diagram showing the flow of an information processing method according to the present disclosure. [Figure 7] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 8] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 9] FIG. 1 is a flow diagram showing the flow of an information processing method according to the present disclosure. [Figure 10] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 11] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 12] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 13] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 14] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 15] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 16] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 17] FIG. 1 is a block diagram illustrating a configuration of a computer that functions as an information processing device according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0014] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technologies (part or all of the products or methods) employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, embodiments obtained by appropriately omitting some of the technologies employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, the effects mentioned in the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, embodiments that do not exhibit the effects mentioned in the exemplary embodiments shown below may also be included in the scope of the present invention.

[0015] [First embodiment] A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is the basic form of each exemplary embodiment described later. The scope of application of each technology employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise. Furthermore, each technology shown in the drawings referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise.

[0016] (Configuration of information processing device 1) The configuration of an information processing device 1 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the information processing device 1. As shown in Fig. 1, the information processing device 1 includes an acquisition unit 11, a generation unit 12, a calculation unit 13, and a storage unit 14. As an example, the information processing device 1 is configured to calculate and store in advance, using a language model, association information to be referenced in a search process executed in an information processing device 2, which will be described later.

[0017] (Acquisition part 11) The acquisition unit 11 acquires input data including a group of sentences. Here, the group of sentences includes, as an example, one or more documents including a plurality of sentences written in a natural language, but this does not limit the present exemplary embodiment. Furthermore, the language of the group of sentences is not particularly limited.

[0018] (Generation part 12) The generation unit 12 generates a collection of passages including a plurality of passages included in the group of sentences. As an example, the generation unit 12 extracts a plurality of passages from the group of sentences and includes the extracted plurality of passages in the collection of passages. Here, a passage may be a unit such as a paragraph, a sentence, a clause, a word, or a morpheme included in the group of sentences, or may be another unit. As an example, a passage may be a group of a predetermined number of characters extracted from the group of sentences.

[0019] (Calculation unit 13) The calculation unit 13 calculates association information, which includes the strength of association between a plurality of passages included in the passage collection and is referred to in the search process, using a language model. Here, a specific example of calculating the association information using a language model does not limit the present exemplary embodiment, but as an example, Using the embedding model, a passage p a For the passage p a Similar passages similar to p b Get Similar passages p b Given a passage p, the language model a The generation probability LM(p a |p b ) and calculate the generation probability for the passage p a and p b Score s indicates the strength of the association with ab Let's say You can also perform the following process: A passage p a and similar passages p bThe cosine similarity between the passage p a and p b Score s indicates the strength of the association with ab Let's say As an example, when the set of passages includes passages p1, p2, and p3, the calculation unit 13 may perform the following process: A score s indicating the strength of the connection between passages p1 and p2 12 Score s indicates the strength of the connection between passages p2 and p3 23 Score s indicates the strength of the connection between passage p3 and p1 31 As described above, a score indicating the strength of the association defined in each passage pair may be calculated for a plurality of passage pairs, and the calculated scores may be included in the association information.

[0020] (Storage unit 14) The storage unit 14 stores the association information in association with the plurality of passages. As an example, the storage unit 14 stores the association information in a storage device (not shown). As an example, the stored association information is referenced in the above-mentioned search process. Note that the storage unit 14 may also be referred to as a storage control unit.

[0021] (Effects of information processing device 1) As described above, in the information processing device 1, Obtain input data containing sentences, Generate a set of passages including a plurality of passages included in the set of sentences; calculating association information including the strength of association between a plurality of passages included in the collection of passages, the association information being referenced in the search process, using a language model; storing the association information in association with the plurality of passages; The above configuration is adopted. In this way, the information processing device 1 performs a process of calculating and storing association information to be referenced in the search process using a language model. Therefore, in the search process, it is not necessary to calculate the association information again using the language model. Therefore, with the above configuration, the processing speed of the search process is improved and user convenience is improved. Furthermore, with the above configuration, the association information can be used in the search process, thereby improving the self-containedness of the information and, as a result, improving the quality of the ultimately generated response.

[0022] (Flow of information processing method S1) Next, the flow of information processing method S1 according to this exemplary embodiment will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of information processing method S1. As shown in Fig. 2, information processing method S1 includes a step (process) S11 of acquiring input data, a step (process) S12 of generating a collection of passages, a step (process) S13 of calculating association information, and a step (process) S14 of storing the association information.

[0023] (Step S11) In step S11, the acquisition unit 11 acquires input data including a group of sentences. The specific processing by the acquisition unit 11 has been described above, and therefore will not be described here.

[0024] (Step S12) In step S12, the generation unit 12 generates a set of passages including a plurality of passages included in the set of sentences. The specific processing by the generation unit 12 has been described above, and therefore will not be described here.

[0025] (Step S13) In step S13, the calculation unit 13 calculates association information, which includes the strength of association between the plurality of passages included in the passage collection and is referenced in the search process, using a language model. The specific process performed by the calculation unit 13 has been described above, so a detailed description thereof will be omitted here.

[0026] (Step S14) In step S14, the storage unit 14 stores the association information in association with the plurality of passages. The specific processing by the storage unit 14 has been described above, and therefore will not be described here.

[0027] (Effect of information processing method S1) As described above, in the information processing method S1, Obtain input data containing sentences, Generate a set of passages including a plurality of passages included in the set of sentences; calculating association information including the strength of association between a plurality of passages included in the collection of passages, the association information being referenced in the search process, using a language model; storing the association information in association with the plurality of passages; According to the above configuration, the same effects as those of the information processing device 1 are achieved.

[0028] (Configuration of information processing device 2) The configuration of an information processing device 2 according to this exemplary embodiment will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the information processing device 2. As shown in Fig. 3, the information processing device 2 includes a first acquisition unit 21, a first search unit 22, a second search unit 23, and a third search unit 24. As an example, the information processing device 2 is configured to execute a search process by referring to the association information calculated by the information processing device 1 described above.

[0029] (First acquisition unit 21) The first acquisition unit 21 acquires a query. Here, the query is written in a natural language as an example, but this does not limit the present exemplary embodiment. Furthermore, the language of the query is not particularly limited.

[0030] (First search unit 22) The first search unit 22 searches for an initial passage related to the query from a set of passages including a plurality of passages. Here, the set of passages may be configured to use a set of passages generated by the generation unit 12 included in the information processing device 1, as an example, but this does not limit the present exemplary embodiment. The term "initial passage" is used merely for convenience in explaining the process, and does not limit the present exemplary embodiment. The "initial passage" may also be expressed as a "first passage," a "first type of passage," or the like.

[0031] (Second search unit 23) The second search unit 23 searches for an additional passage from the set of passages by referring to the initial passage and association information including the strength of association between passages included in the set of passages. Here, the association information may be calculated in advance using a language model, for example. More specifically, the association information may be calculated by the calculation unit 13 included in the information processing device 1. The term "additional passage" is used merely for the sake of convenience in explaining the process, and does not limit the present exemplary embodiment. The "additional passage" may also be expressed as a "second passage," a "second type of passage," or the like.

[0032] (Third search unit 24) The third search unit 24 executes a search process using the initial passage and the additional passage. As an example, the third search unit 24 may execute the search process by generating a prompt including the initial passage and the additional passage and inputting the prompt into a language model or a generative model. The search results by the third search unit 24 are presented to the user via, for example, a presentation (not shown). The search process may also be referred to as a generative process.

[0033] (Effects of information processing device 2) As described above, in the information processing device 2, Get the query, Searching for an initial passage related to the query from a collection of passages (first search process); Searching for additional passages from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage (second search process); Execute a search process (third search process) using the initial passage and the additional passage. The information processing device 2 employs the following configuration. In this way, the information processing device 2 searches for additional passages from the collection of passages by referring to the initial passage and association information including the strength of association between passages included in the collection of passages, and then executes a search process using the initial passage and the additional passage. This improves the self-containedness of the information used in the third search process, thereby improving the quality of the response (the result of the third search process) that is finally generated. Furthermore, in the above configuration, the association information referenced in the search process may be calculated and stored in advance using a language model. Therefore, there is no need to recalculate the association information using a language model in the search process. Therefore, the above configuration improves the processing speed of the search process and improves user convenience.

[0034] (Flow of information processing method S2) Next, the flow of information processing method S2 according to this exemplary embodiment will be described with reference to Fig. 4. Fig. 4 is a flow diagram showing the flow of information processing method S2. As shown in Fig. 4, information processing method S2 includes a step (process) S21 of acquiring a query, a step (process) S22 of searching for a processing passage, a step (process) S23 of searching for an additional passage, and a step (process) S24 of executing a search process using the initial passage and the additional passage.

[0035] (Step S21) In step S21, the first acquisition unit 21 acquires a query. The specific processing by the first acquisition unit 21 has been described above, and therefore will not be described here.

[0036] (Step S22) In step S22, the first search unit 22 searches a collection of passages containing a plurality of passages for an initial passage related to the query. The specific processing by the first search unit 22 has been described above, and therefore will not be described here.

[0037] (Step S23) In step S23, the second search unit 23 searches for additional passages from the set of passages by referring to the initial passage and association information including the strength of association between the passages included in the set of passages. The specific processing by the second search unit 23 has been described above, so a detailed description thereof will be omitted here.

[0038] (Step S24) In step S24, the third search unit 24 executes a search process using the initial passage and the additional passage. The specific process performed by the third search unit 24 has been described above, and therefore will not be described here.

[0039] (Effect of information processing method S2) As described above, in the information processing method S2, Get the query, Searching for an initial passage related to the query from a collection of passages (first search process); Searching for additional passages from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage (second search process); Execute a search process (third search process) using the initial passage and the additional passage. The above configuration provides the same effects as the information processing device 2.

[0040] Second Embodiment A second exemplary embodiment, which is one example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be assigned the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technology employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs. Furthermore, each technology shown in each drawing referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs.

[0041] (Configuration of information processing system 1A) The configuration of an information processing system 1A according to this exemplary embodiment will be described with reference to Fig. 5. Fig. 5 is a block diagram showing the configuration of the information processing system 1A. As shown in Fig. 5, the information processing system 1A includes an information processing device 100, and a first server device 50 and a second server device 60 connected to the information processing device 100 via a network N. Here, the specific configuration of the network N does not limit this exemplary embodiment, but as an example, a wireless LAN (Local Area Network), a wired LAN, a WAN (Wide Area Network), a public line network, a mobile data communication network, or a combination of these networks can be used.

[0042] (First server device 50) 5, the first server device 50 includes a control unit 51, a storage unit 52, and a communication unit 53. The communication unit 53 communicates with devices external to the first server device 50. As an example, the communication unit 53 communicates with an information processing device 100 included in the information processing system 1A. The communication unit 53 transmits data supplied from the control unit 51 to the information processing device 100, and supplies data received from the information processing device 100 to the control unit 51.

[0043] A language model LM is stored in the storage unit 52. As an example, a plurality of parameters that define the language model LM are stored in the storage unit 52. As an example, these parameters are parameters that have been learned in advance by machine learning (parameters that have undergone an update process by machine learning), but this does not limit the present exemplary embodiment.

[0044] The control unit 51 uses the language model LM to obtain an output result from the language model LM. As an example, the control unit 51 inputs data received from the information processing device 100 into the language model LM and obtains an output result from the language model LM. The control unit 51 also provides the output result to the information processing device 100 via the communication unit 53. Specific processing by the language model LM will be described later.

[0045] (Second server device 60) As shown in FIG. 5 , the second server device 60 includes a control unit 61, a storage unit 62, and a communication unit 63. The communication unit 63 communicates with devices external to the second server device 60. As an example, the communication unit 63 communicates with an information processing device 100 included in the information processing system 1A. The communication unit 63 transmits data supplied from the control unit 61 to the information processing device 100, and supplies data received from the information processing device 100 to the control unit 61. Note that the data received by the communication unit 63 from the information processing device 100 may include a prompt generated by the information processing device 100. Furthermore, the data provided by the communication unit 63 to the information processing device 100 may include a generation result generated by a generative model GM (described later) based on the prompt.

[0046] The storage unit 62 stores a generative model GM. As an example, the storage unit 62 stores a plurality of parameters that define the generative model GM. These parameters are, as an example, parameters that have been learned in advance by machine learning (parameters that have undergone an update process by machine learning), but this does not limit the present exemplary embodiment. A large-scale language model learned by machine learning can also be used as the generative model GM, but this does not limit the present exemplary embodiment.

[0047] The control unit 61 acquires information generated by the generative model GM by using the generative model GM. As an example, the control unit 61 acquires a generation result generated by the generative model GM based on a prompt received from the information processing device 100. The control unit 61 also provides the generation result to the information processing device 100 via the communication unit 63. Specific processing by the generative model GM will be described later.

[0048] In the present exemplary embodiment, the first server device 50 and the second server device 60 are illustrated as devices separate from the information processing device 100, but this does not limit the present exemplary embodiment. The control unit 51 of the first server device 50 or the function of the language model execution unit in the control unit 51 may be configured to be provided in the control unit of the information processing device 100. Furthermore, the control unit 61 of the second server device 60 or the function of the generative model execution unit in the control unit 61 may be configured to be provided in the control unit of the information processing device 100. Similarly, the language model LM stored in the storage unit 52 of the first server device 50 may be stored in the storage unit of the information processing device 100, and the information processing device 100 itself may be configured to be able to execute the language model LM. Furthermore, the generative model GM stored in the storage unit 62 of the second server device 60 may be stored in the storage unit of the information processing device 100, and the information processing device 100 itself may be configured to be able to execute the generative model GM.

[0049] In the above example, the language model LM and the generative model GM are described as separate models, but this does not limit the present exemplary embodiment. The language model LM and the generative model GM may be configured to be realized by a single machine-learned model.

[0050] (Configuration of information processing device 100) Next, the configuration of the information processing device 100 according to this exemplary embodiment will be described with reference to Fig. 5. As shown in Fig. 5, the information processing device 100 includes a control unit 10, a storage unit 20, a communication unit 30, and an input / output unit 40.

[0051] (Communication unit 30) The communication unit 30 communicates with devices external to the information processing device 100. As an example, the communication unit 30 communicates with a first server device 50 and a second server device 60. The communication unit 30 transmits data supplied from the control unit 10 to the first server device 50 and the second server device 60, and supplies data received from the first server device 50 and the second server device 60 to the control unit 10.

[0052] (Input / output section 40) The input / output unit 40 is configured to include at least one of input / output devices such as a keyboard, a mouse, a display, a printer, and a touch panel. Alternatively, the input / output unit 40 may be configured to have input / output devices such as a keyboard, a mouse, a display, a printer, and a touch panel connected to it. In this configuration, the input / output unit 40 accepts various types of information input to the information processing device 100 from the connected input devices. Furthermore, the input / output unit 40 outputs various types of information to connected output devices under the control of the control unit 10. An example of the input / output unit 40 is an interface such as a USB (Universal Serial Bus).

[0053] (Storage unit 20) The storage unit 20 stores various data referenced by the control unit 10 and various data generated by the control unit 10. As an example, the storage unit 20 stores: ·Text group IND Passage Group PG ·Association Information AI Query QR Directed graph OG Initial Passage IP Additional Passage AP Prompt PR ·Generation result GR Output data OUT ·Model group MG Specific examples of each data item will be described later, so a detailed description will be omitted here.

[0054] (Control unit 10) 5, the control unit 10 includes an acquisition unit 11, a generation unit 12, a calculation unit 13, a first search unit 22, a second search unit 23, a third search unit 24, and an output data generation unit 25. Here, the acquisition unit 11 has the functions of the acquisition unit 11 and the first acquisition unit 21 described in exemplary embodiment 1, and therefore the acquisition unit 11 will also be referred to as the acquisition unit 11(21). The acquisition unit 11 may also be referred to as the second acquisition unit 11. The calculation unit 13 has the functions of the calculation unit 13 and the storage unit 14 described in exemplary embodiment 1, and therefore the calculation unit 13 will also be referred to as the calculation unit 13(14).

[0055] In the control unit 10, roughly speaking, Obtain input data containing sentences, Generate a set of passages including a plurality of passages included in the set of sentences; calculating association information including the strength of association between a plurality of passages included in the collection of passages, the association information being referenced in the search process, using a language model; storing the association information in association with the plurality of passages; The process (first process group) Get the query, Searching for an initial passage related to the query from a collection of passages (first search process); Searching for additional passages from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage (second search process); Execute a search process (third search process) using the initial passage and the additional passage. In the following description, the first processing group is also referred to as processing in the indexing phase, and the second processing group is also referred to as processing in the search phase.

[0056] (Acquisition part 11(21)) In the indexing phase, the acquisition unit 11 (21) acquires input data including a set of sentences IND. The set of sentences IND acquired by the acquisition unit 11 (21) is stored in the storage unit 20, for example. Here, the set of sentences IND includes, for example, one or more documents including multiple sentences written in a natural language, but this does not limit the present exemplary embodiment. Furthermore, the language of the set of sentences IND is not particularly limited.

[0057] Furthermore, the acquisition unit 11 (21) acquires a query QR in the search phase. The query QR acquired by the acquisition unit 11 (21) is stored in the storage unit 20, for example. Here, the query QR is written in a natural language, for example, but this does not limit the present exemplary embodiment. Furthermore, the language of the query QR is not particularly limited.

[0058] (Generation part 12) In the indexing phase, the generation unit 12 generates a passage collection PG including a plurality of passages PS included in the sentence group IND. As an example, the generation unit 12 performs a process of extracting a plurality of passages PS from the sentence group IND and including the extracted plurality of passages PS in the passage collection PG. Here, a passage PS may be a unit such as a paragraph, sentence, clause, word, or morpheme included in the sentence group IND, or may be another unit. As an example, a passage may be a group of a predetermined number of characters extracted from the sentence group. (Calculation unit 13) In the indexing phase, the calculation unit 13 calculates association information AI, which includes the strength of association between the plurality of passages PS included in the passage set PG, and is referred to in the search process, using the language model LM. Here, a specific example of calculation of the association information using the language model LM does not limit this exemplary embodiment, but as an example, Using the embedding model EM, a passage p a For the passage p a Similar passages similar to p b Get Similar passages p b Given a passage p, the language model LM a The generation probability LM(p a |p b ) and calculate the generation probability for the passage p a and p b Score s indicates the strength of the association with ab Let's say You can also perform the following process: A passage p a and similar passages p b The cosine similarity between the passage p a and p b Score s indicates the strength of the association with ab Let's say Here, the embedding model EM is an example of a model included in the model group MG stored in the storage unit 20. In addition, the generation probability LM(p a |p b ), the calculation unit 13 calculates ·The above passage p a and the similar passage p b The generation probability LM(p a |p b ) to the first server device 50 via the communication unit 30; A response to the instruction by the language model LM with a generation probability LM(p a |pb ) via the communication unit 30. The following configuration may be adopted.

[0059] Furthermore, as a more specific example, when the set of passages includes passages p1, p2, and p3, the calculation unit 13 calculates A score s indicating the strength of the connection between passages p1 and p2 12 Score s indicates the strength of the connection between passages p2 and p3 23 Score s indicates the strength of the connection between passage p3 and p1 31 As described above, a score indicating the strength of the association defined in each passage pair may be calculated for a plurality of passage pairs, and the calculated scores may be included in the association information.

[0060] Furthermore, the calculation unit 13 stores the association information AI in association with the plurality of passages PS. As an example, the calculation unit 13 stores the association information AI in the storage unit 20. As an example, the stored association information AI is referenced in the above-mentioned search phase. A more specific example of processing by the calculation unit 13 will be described later.

[0061] (First search unit 22) In the search phase, the first search unit 22 searches a passage set PG including a plurality of passages PS for an initial passage IP related to the query QR. Here, as an example, the passage set PG may be the passage set generated by the generator 12 in the indexing phase. Note that in this exemplary embodiment, the term "initial passage" is used merely for the sake of convenience in explaining the process, and is not intended to limit the present exemplary embodiment. The "initial passage" may also be expressed as a "first passage" or a "first type of passage." Specific examples of the process performed by the first search unit 22 will be described later.

[0062] (Second search unit 23) In the search phase, the second search unit 23 searches for an additional passage AP from the passage set PG by referring to the initial passage IP and association information AI, which includes the strength of association between passages PS included in the passage set PG. Here, the association information AI may be, for example, information calculated in advance by the calculation unit 13 (14). Note that, in this exemplary embodiment, the term "additional passage" is used merely for the sake of convenience in explaining the process, and is not intended to limit the present exemplary embodiment. The "additional passage" may also be expressed as a "second passage," a "second type of passage," or the like. Specific examples of the process performed by the second search unit 23 will be described later.

[0063] (Third search unit 24) The third search unit 24 executes a search process using the initial passage and the additional passage in the search phase. As shown in Fig. 5, the third search unit 24 includes a prompt generation unit 241 and a generation result acquisition unit 242. As an example, the third search unit 24 A prompt generation unit 241 generates a prompt PR including the initial passage IP and the additional passage AP; Sending the prompt PR to the second server device 60 via the communication unit 30, The prompt PR is input to the generative model GM to execute the search process (generative process), The search results (generation results) based on the generation model GM are acquired by the generation result acquisition unit 242. A specific example of the process performed by the third search unit 24 will be described later.

[0064] (Output data generation unit 25) The output data generation unit 25 generates output data OUT by referring to the search results (generation results) using the generative model GM. The generated output data OUT is, for example, visually presented to the user via the input / output unit 40.

[0065] (Process flow during the indexing phase) Next, a process flow in the indexing phase executed by the information processing device 100 will be described with reference to Fig. 6 to Fig. 8. Fig. 6 is a flow diagram showing an example of the process flow in the indexing phase.

[0066] (Step S11) In step S11, the acquisition unit 11 acquires input data including a sentence group IND.

[0067] (Steps S121 and S122) In step S121, the generation unit 12 generates a collection of passages PG including a plurality of passages PS included in the group of sentences IND. Then, in step S122, the generation unit 12 stores the generated collection of passages PG in the storage unit 20.

[0068] FIG. 7 is a diagram for explaining a specific example of a passage set generation process by the generation unit 12. As shown in FIG. 7, a sentence group IND (also referred to as a document set in FIG. 7) is input to the generation unit 12 (also referred to as a passage conversion unit in FIG. 7), and the generation unit 12 divides the document set into a plurality of passages PS, and stores the passages PS obtained by the division as elements of a passage set PG in the storage unit 20 (also referred to as a database in FIG. 7). The passage set generation process may also be referred to as a passage conversion process, a passage extraction process, etc. The document set IND is also referred to as a collection of documents d i As a set of

number

number

[0069] (Steps S13 and S14) Next, in step S13, the calculation unit 13 calculates association information AI, which includes the strength of association between the plurality of passages PS included in the passage set PG and is referenced in the search process, using the language model LM. Then, in step S14, the calculation unit 13 stores the association information AI in the storage unit 20 in association with the plurality of passages PS.

[0070] 8 is a diagram for explaining a specific example of the association information calculation process by the calculation unit 13. As shown in FIG. 8, the calculation unit 13 (also referred to as the passage association calculation unit in FIG. 8) Read out a plurality of passages PS from the storage unit 20; Using the passage embedding model EM and the language model LM, calculate association information AI including the strength of association between the plurality of passages PS; The calculated association information AI is stored in the storage unit 20. The following process is performed.

[0071] Here, the passage set PG (plurality of passages PS) input to the calculation unit 13 is

number

number

[0072] More specifically, the association information calculation process by the calculation unit 13 may be configured to include the following processes.

[0073] (Step S131: Process of identifying sets of related passages) The calculation unit 13 calculates an embedded model M sim Embedding of all passages using

number

[0074] Then, the calculation unit 13 calculates all the passage pairs.

number

[0075] Then, the calculation unit 13 calculates, for example, each passage p a Alternatively, the calculation unit 13 may also select a set of passages that are extracted from the same document, thereby obtaining the k most similar passages for each passage p a The k passages with the highest similarity may be acquired for the given sentence.

[0076] (Step S132: Processing for calculating association information for a set of passages) Then, the calculation unit 13 Passage p a For the passage p a Similar passages similar to p b Given p, the language model LM a The generation probability LM(p a |p b ) is calculated, ·Calculated generation probability LM(p a |p b ) in passage p a and passage p b Score s that indicates the strength of the association with ab The score is as follows: Passage p b appears in a sentence, followed by p a The naturalness of the appearance It can also be seen as a score, Passage p b Passage p a How good a context for It can also be seen as a score. a and passage pb Score s that indicates the strength of the association with ab As the embedding e a and embedding e b Alternatively, the cosine similarity between the two may be used.

[0077] (Processing flow in the search phase) Next, a process flow in the search phase executed by information processing device 100 will be described with reference to Fig. 9 to Fig. 13. Fig. 9 is a flow diagram showing an example of the process flow in the search phase.

[0078] (Steps S21 and S22) In step S21, the first acquisition unit 21 acquires a query QR. Then, in step S22, the first search unit 22 searches a passage set PG including a plurality of passages PS for an initial passage IP related to the query QR. In the following description, the query QR will also be referred to as a query q.

[0079] 10 is a diagram showing an example of an initial passage search process by the first search unit 22. As shown in FIG. 10, the first search unit 22 (also referred to as an initial passage search unit in FIG. 10) searches for an initial passage IP by referring to a query QR(q) and a plurality of passages PS included in a passage set PS(P) stored in the storage unit 20. As an example, the first search unit 22 Find the k most relevant passages to query q from a set of passages P, Output the acquired k passages as the initial passage IP. Here, the initial passage IP is called P init

number

[0080] More specifically, the first search unit 22 · Embedding model for search M idxTransform the query q into a search vector using Using the transformed search vector, we compute the embedding vector E for each passage calculated in the indexing phase. idx Calculate the similarity with A predetermined number (e.g., 5) of passages similar to the query q are extracted as an initial passage P init Output as The embedded model M idx is an example of a model included in the model group MG stored in the storage unit 20.

[0081] (Step S23) Next, in step S23, the second search unit 23 searches for an additional passage AP from the passage set PG by referring to the initial passage IP and association information AI including the strength of association between passages included in the passage set PG. Fig. 11 is a diagram showing an example of an additional passage search process by the second search unit 23. As shown in Fig. 11, the second search unit 23 (also referred to as an additional passage search unit in Fig. 11) Query QR(q) ·Passage embedded model EM(M sim ) - Multiple passages PS(P) included in the passage set PG ·Inter-passage correlation information AI(C) Initial Passage IP(P init ) Search for additional passages AP by referring to the additional passages AP. second It is sometimes written as:

[0082] The process by the second search unit 23 is to search for a query q, an initial passage P init ,Using the correlation information C between passages, the initial passage P related to the query q is init It can also be expressed as a process of acquiring additional passages on which P strongly depends. Here, the number of additional passages acquired (additional passages P second The number of ( ) can be determined arbitrarily.

[0083] More specifically, the additional passage search process by the second search unit 23 may be configured to include the following processes.

[0084] (Step S231: Processing to calculate information (directed graph) used to search for additional passages) The second search unit 23 searches for the correlation information C between the passages.

number

number

number

[0085] (Step S232: Processing for searching for additional passages using the above information (directed graph)) Then, the second search unit 23 executes a search algorithm including the following processes. Early Passage P init For each node, set the "transition probability (weight) when randomly transitioning nodes" uniformly. And, for the edges coming out of each node, set the edge weight uniformly. Then, identify the m passages other than the initial passage that have the highest score PR(p) according to the algorithm. Then, the identified m passages are divided into additional passages P second Output as

[0086] Here, the method of calculating the score PR(p) does not limit the present exemplary embodiment, but as an example, the score PR(p) for a certain passage p may be calculated as follows: -Weight assigned to the edge ending at the passage - Weight assigned to the passage at the start point of the edge that ends at the passage The score PR(p) may be determined by referring to at least one of the following: Furthermore, the PageRank algorithm may be used as a method for calculating the score PR(p).

[0087] FIG. 12 shows an example of the directed graph G(V,E) referred to by the second search unit 23. Early Passage P init (Indicated as IP in Figure 12) Additional passages P found by the above algorithm second (denoted as AP in Figure 12) shows a directed graph G(V,E) containing

[0088] In this way, the second search unit 23 searches for one or more passage pairs (p a , p b ) a , p b ) and select an additional passage AP(P second ) may be searched for. In this step, the second search unit 23 searches each passage p a The edge (p a ,p b) for which the association score s ab The above algorithm may be executed using the value of as the weight of the edge. For example, any one of the following processing examples or a combination thereof may be executed.

[0089] (Processing example 1) The second search unit 23 searches for s ab The value of

number

[0090] (Processing example 2) Alternatively, the second search unit 23 may search for s ab Using the value of , passage p b Rank r b For example, the second search unit 23 may be configured to calculate the position of the searched item (counting from the top). Rank R b The reciprocal of 1 / r b Calculate Normalize the reciprocal, The normalized inverse is expressed as the edge (p a ,p b ) weight may be set to

[0091] Moreover, the above-mentioned processing examples 1 and 2 may be used in combination with the following processing example 3.

[0092] (Processing example 3) The second search unit 23 searches the initial passage P init The "transition probability when transitioning nodes randomly" may be set using the scores or orders assigned to the nodes belonging to . For example, the initial passage P initIf you select as P init Cosine similarity can be used as the score given to nodes belonging to

[0093] In each of the above processes, P init Nodes other than those belonging to may be assigned a transition probability of 0.

[0094] Moreover, instead of or together with the above-described processing examples, the following processing example 4 may be executed.

[0095] (Processing example 4) In this processing example, the second search unit 23 searches the initial passage IP(P init ) and the association information AI(C), and an additional passage AP(P second ) is searched for. This processing example 4 includes the following steps S41 and S42.

[0096] (Step S41) First, the second search unit 23 searches for the correlation information C between the passages.

number

number

number

[0097] (Step S41) Next, the second search unit 23 searches for the initial passage P initFrom the above, a set of passages that can be reached in n moves is identified. Here, the set of passages that can be reached in n moves is also denoted as passage set A.

[0098] Then, the second search unit 23 searches for the node V'

number

number

[0099] 13 shows an example of a partial directed graph G'(V', E') referred to by the second search unit 23 that executes this processing example. In FIG. 13, the partial directed graph is Early Passage P init (Indicated as IP in Figure 12) Additional passages P found by the above algorithm second (denoted as AP in FIG. 12) and the initial passage P init From, additional passage P that can be reached in one move second and The graph shows a directed partial graph G'(V',E') that includes

[0100] At least a part of the digraph G(V,E) shown in FIG. 12 or at least a part of the partial digraph G'(V',E') shown in FIG. 13 may be visually presented to the user via the input / output unit 40. As an example, the second search unit 23 Present the directed graph G(V,E) to the user, Accepting a user instruction regarding the directed graph G(V,E) (for example, inputting a specific value for n), Construct a partial directed graph G'(V',E') that reflects the user's instructions and present it to the user. The following processing may be performed.

[0101] Moreover, instead of or together with the above-described processing examples, the following processing example 5 may be executed.

[0102] (Processing example 5) In this processing example 5, the second search unit 23 searches the one or more edges (p a , p b ) for each of The edge (p a , p b a first score indicating the strength of association between the pair of passages defining the query QR(q), the first score being calculated in advance without reference to the query QR(q); The edge (p a , p b a second score indicating the strength of association between the pair of passages defining the query QR(q), the second score being calculated with reference to the query QR(q); , and select an additional passage AP(P second ) The following processing is performed in this processing example 5. The following steps S51 and S52 are included.

[0103] (Step S51: Processing for calculating associations between passages using a query) The second search unit 23 retrieves the second score f(q,p a , p b )of,

number

[0104] The second search unit 23 searches for the correlation information C between the passages.

number

[0105] Note that the second score f(q,p a , p b ) calculation process, ·e a : Passage p a Embedding ·e a+b : Passage p a and passage p b Embedding a string of The calculation process of can be performed in advance in the indexing phase. In such a configuration, in the search phase, ·e q : Embedding of query q sim: Cosine similarity between embeddings Since this calculation process can be performed without using the language model LM, even in the configuration that utilizes the second score as described above, it is possible to suppress an increase in the amount of processing in the search phase, and user convenience is not impaired.

[0106] (Step S52: Processing for aggregating the association information between passages calculated using the query and the association information between passages calculated in advance)

[0107] In this step, each node (passage) p of the directed graph G(V,E) a For the node p a The set of nodes N(p a ) is considered. a ) is the node p a Here, the edge (p a , p b )

number

[0108] The second search unit 23, for example, Using each score (the first score above, the second score above), set N(p a ) passage p b Rank r b Calculate each (which is the largest from the top) - Compile two rankings and create a single ranking. The second search unit 23 performs the process of: b The score of |N(p a )|-r b +1 Here, |N(pa )| is the set N(p a ) and the second search unit 23 performs the following search, for example: Add up the Borda scores in the two rankings, ·Total score: p b Sort By doing so, it is possible to derive a ranking that aggregates the first score and the second score. Then, the second search unit 23 recalculates the score (aggregated score) for each edge based on the aggregated ranking. For example, the second search unit 23 calculates the aggregated score by using the inverse of the aggregated ranking.

[0109] The second search unit 23 performs the above process on all passages p a In other words, by executing the above process, the second search unit 23 can construct a new graph G that takes the query q into consideration when evaluating the association between passages, and can use the new graph G to search for the additional passage AP(P second ) can be searched for.

[0110] In this way, in this processing example, the second search unit 23 searches for the one or more edges (p a , p b ) for each of the first scores s ab and the second score f(q,p a , p b ) and the score obtained by aggregating them, an additional passage AP(P second ) is searched for.

[0111] As described above, in this processing example, ·e a : Passage p a Embedding ·e a+b : Passage p a and passage pb Embedding a string of The calculation process of can be performed in advance in the indexing phase. For example, the calculation process can be performed using a language model LM, taking advantage of the abundant information of the language model LM. ·e q : Embedding of query q sim: Cosine similarity between embeddings The calculation process of and the aggregation process of the first score and the second score can be performed using a lightweight embedding model without using a language model.

[0112] Therefore, according to the above configuration, the evaluation of the passage taking into account the query q and the additional passage AP(P second ) can be searched appropriately without reducing user convenience.

[0113] Returning to FIG. 9, the explanation will be continued.

[0114] (Steps S241, S242, S243) In step S241, the prompt generator 241 generates a prompt PR using the initial passage IP retrieved in step S22 and the additional passage AP retrieved in step S23. Here, the prompt includes the initial passage IP and the additional passage AP.

[0115] Then, in step S242, the prompt generation unit 241 inputs the generated prompt PR to the generative model GM via the communication unit 30. Then, in step S243, the generation result acquisition unit 242 acquires the generation result GR generated by the generative model GM based on the prompt PR.

[0116] <Application example> An application example of the information processing system 1A according to this exemplary embodiment will be described below with reference to Fig. 14 to Fig. 16. Fig. 14 mainly shows a processing example in the indexing phase, and Fig. 15 and Fig. 16 mainly show a processing example in the search phase.

[0117] First, as shown in Fig. 14, a document set IND is input and acquired by the acquisition unit 11 (21). Then, the document set IND is converted into a plurality of passages by the generation unit 12 (corresponding to step S121 described above). Here, the conversion process is carried out as follows, as shown in Fig. 14: A process of dividing the document set IND into multiple passages (PS1, PS2, PS3, ...) (S1211 in FIG. 14), and Converting each passage into an embedding vector (S1212 in Figure 14) Here, the process of converting into an embedding vector can be performed by the embedding model EM described above, for example.

[0118] Then, either or both of each divided passage and the embedding vector corresponding to each passage are stored in the storage unit 20 (database in FIG. 14) (S122 in FIG. 14).

[0119] Also, as shown in FIG. 14, the correlation information AI(C) between each passage (PS1, PS2, PS3, ...) and the passages connected to the passage via edges is calculated by the calculation unit 13 (14) (S13 in FIG. 14).

[0120] In the example shown in Figure 14, ·Association information (association score) between passage PS1 and passage PS4 is s 14 =0.8 is calculated, · As the association information (association score) between passage PS1 and passage PS5, s 15 =0.01 is calculated.

[0121] This association information AI(C) is stored in the storage unit 20 (database in FIG. 14) together with the directed graph OG made up of a plurality of passages (PS1, PS2, PS3, . . . ).

[0122] On the other hand, as shown in FIG. 15, when a query QR(q) is acquired, the first search unit 22 searches the plurality of passages (PS1, PS2, PS3, . . . ) for an initial passage IP(P) related to the query QR(q). init In the example shown in FIG. 15, the first search unit 22 searches for · Embedding vector of query QR(q) Search for vectors similar to the vector in the above passages (PS1, PS2, PS3, ...). By doing so, the initial passage IP(P init ) is identified (S22 in FIG. 15).

[0123] Then, the second search unit 23 searches the initial passage IP(P init ) and the association information AI(C) associated with the directed graph OG, and second )) is searched for (S23 in FIG. 15). In the example shown in FIG. 15, the following is searched for as an additional passage to the initial passage IP (PS1) "bbb came to aaa's place..." "The twin sisters eee and fff ggg..." (AP1 (PS4)) "I've been hearing whistles at night lately from KKK" (AP2 (PS6)) is being searched for.

[0124] Then, the third search unit 24 performs the following search as shown in FIG. Query QR(q) Initial Passage IP Additional Passage AP The third search unit 24 then inputs the generated prompt PR to the generative model GM or the language model LM to obtain text as the generation result (search result) (S242, S243 in FIG. 16). The output data generation unit 25 then generates output data OUT including the obtained generation result (search result), and presents it to the user via the input / output unit 40.

[0125] (Effects of the information processing device 100) As described above, in the indexing phase, the information processing device 100 Obtain input data containing sentences IND, Generate a set of passages PG including a plurality of passages PS included in the set of sentences IND; Between the plurality of passages included in the passage set PG (p a ,p b ) including the strength of the connections between a , p b ,s ab ) and the association information C(p a , p b ,s ab ) is calculated using the language model LM, The plurality of passages (p a ,p b , , ) and the association information C(p a , p b ,s ab ) to remember In addition, in the search phase, Get the query QR, Searching for an initial passage IP related to the query GQ from a passage set PG including a plurality of passages (first search process); Association information C(p a , p b ,s ab) and the initial passage IP, searching for an additional passage AP from the passage set PG (second search process); Execute a search process (third search process) using the initial passage IP and the additional passage AP. In this way, the information processing device 100 calculates and stores association information referenced in the search process using a language model. Therefore, there is no need to recalculate association information using a language model in the search process. Therefore, the above configuration improves the processing speed of the search process and improves user convenience. Furthermore, the above configuration refers to association information including the strength of association between passages included in the passage set and the initial passage to search for an additional passage from the passage set, and executes the search process using the initial passage and the additional passage. Therefore, the self-containedness of the information used in the third search process is improved, and as a result, the quality of the finally generated response (the result of the third search process) is improved.

[0126] [Software implementation example] Some or all of the functions of the information processing devices 1, 2, and 100 (hereinafter also referred to as "each of the above devices") may be realized by hardware such as an integrated circuit (IC chip), or by software.

[0127] In the latter case, each of the above devices is realized by, for example, a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in Figure 17. Figure 17 is a block diagram showing the hardware configuration of computer C that functions as each of the above devices.

[0128] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for causing the computer C to operate as each of the above-mentioned devices. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing the functions of each of the above-mentioned devices.

[0129] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0130] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, mouse, display, and printer.

[0131] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can acquire the program P via such a recording medium M. The program P can also be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also acquire the program P via such a transmission medium.

[0132] Furthermore, the functions of each of the devices may be realized by a single processor provided in a single computer, by multiple processors provided in a single computer working in cooperation, or by multiple processors provided in each of multiple computers working in cooperation. Furthermore, the programs for causing each of the devices to realize the functions may be stored in a single memory provided in a single computer, or may be distributed and stored in multiple memories provided in a single computer, or may be distributed and stored in multiple memories provided in each of multiple computers. [Appendix A] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0133] (Appendix A1) a first obtaining means for obtaining a query; a first searching means for searching a collection of passages including a plurality of passages for an initial passage related to the query; a second search means for searching for an additional passage from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage; a third search means for executing a search process using the initial passage and the additional passage; An information processing device comprising:

[0134] (Appendix A2) The second search means A directed graph including one or more edges defined by one or more passage pairs included in the association information is referenced to search for an additional passage from the set of passages. 10. The information processing device according to claim 1,

[0135] (Appendix A3) The second search means, for each of the one or more edges, a first score indicating the strength of association between the pair of passages defining the edge, the first score being calculated in advance without reference to the query; and a second score indicating the strength of association between the pair of passages defining the edge, the second score being calculated with reference to the query; Search for additional passages from the set of passages by referring to 10. The information processing device according to claim 9, wherein the information processing device is a

[0136] (Appendix A4) The second search means and searching the collection of passages for an additional passage using a score obtained by aggregating the first score and the second score for each of the one or more edges. 10. The information processing device according to claim 9, wherein the information processing device is a

[0137] (Appendix A5) The second search means A partial directed graph is obtained by referring to the initial passage and the association information, and the partial directed graph constitutes a part of the directed graph, and an additional passage is searched for from the set of passages. 10. The information processing device according to claim 9, wherein the information processing device is a device for processing a plurality of data.

[0138] (Appendix A6) a second acquiring means for acquiring input data including a group of sentences; a generation means for generating a set of passages including a plurality of passages included in the set of sentences; a calculation means for calculating the association information including the strength of association between a plurality of passages included in the collection of passages; It also has An information processing device according to any one of appendices A1 to A5.

[0139] (Appendix A7) The calculation means calculates the association information using a language model. 10. The information processing device according to claim 9, wherein the information processing device is a

[0140] (Appendix A8) an acquisition means for acquiring input data including a group of sentences; a generation means for generating a set of passages including a plurality of passages included in the set of sentences; a calculation means for calculating association information including the strength of association between a plurality of passages included in the collection of passages, the association information being referenced in a search process, using a language model; a storage means for storing the association information in association with the plurality of passages; An information processing device comprising:

[0141] [Appendix B] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0142] (Appendix B1) a first acquisition process in which at least one processor acquires a query; a first search operation by the at least one processor to search a collection of passages including a plurality of passages for an initial passage related to the query; a second search process in which the at least one processor searches for an additional passage from the set of passages by referring to association information including strengths of association between passages included in the set of passages and the initial passage; a third search process in which the at least one processor executes a search process using the initial passage and the additional passage; An information processing method comprising:

[0143] (Appendix B2) The second search process includes: A directed graph including one or more edges defined by one or more passage pairs included in the association information is referenced to search for an additional passage from the set of passages. 1. The information processing method described in Appendix B1.

[0144] (Appendix B3) The second search process includes, for each of the one or more edges, a first score indicating the strength of association between the pair of passages defining the edge, the first score being calculated in advance without reference to the query; and a second score indicating the strength of association between the pair of passages defining the edge, the second score being calculated with reference to the query; Search for additional passages from the set of passages by referring to 1. The information processing method described in Appendix B2.

[0145] (Appendix B4) The second search process includes: and searching the collection of passages for an additional passage using a score obtained by aggregating the first score and the second score for each of the one or more edges. The information processing method described in Appendix B3.

[0146] (Appendix B5) The second search process includes: A partial directed graph is obtained by referring to the initial passage and the association information, and the partial directed graph constitutes a part of the directed graph, and an additional passage is searched for from the set of passages. 1. An information processing method according to any one of Appendices B2 to B4.

[0147] (Appendix B6) a second acquisition process in which the at least one processor acquires input data including a set of sentences; a generation process in which the at least one processor generates a set of passages including a plurality of passages included in the set of sentences; a calculation process in which the at least one processor calculates the association information including strengths of association between a plurality of passages included in the collection of passages; It further includes 1. An information processing method according to any one of Appendices B1 to B5.

[0148] (Appendix B7) In the calculation process, the at least one processor calculates the association information using a language model. An information processing method as described in Appendix B6.

[0149] (Appendix B8) an acquisition process in which the at least one processor acquires input data including a set of sentences; a generation process in which the at least one processor generates a set of passages including a plurality of passages included in the set of sentences; a calculation process in which the at least one processor calculates association information including strengths of association between a plurality of passages included in the collection of passages, the association information being referenced in a search process, using a language model; a storage process in which the at least one processor stores the association information in association with the plurality of passages; An information processing method comprising:

[0150] [Appendix C] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0151] (Appendix C1) A program that causes a computer to function as an information processing device, The computer a first obtaining means for obtaining a query; a first searching means for searching a collection of passages including a plurality of passages for an initial passage related to the query; a second search means for searching for an additional passage from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage; a third search means for executing a search process using the initial passage and the additional passage; An information processing program that functions as a

[0152] (Appendix C2) The second search means A directed graph including one or more edges defined by one or more passage pairs included in the association information is referenced to search for an additional passage from the set of passages. An information processing program as described in Appendix C1.

[0153] (Appendix C3) The second search means, for each of the one or more edges, a first score indicating the strength of association between the pair of passages defining the edge, the first score being calculated in advance without reference to the query; and a second score indicating the strength of association between the pair of passages defining the edge, the second score being calculated with reference to the query; Search for additional passages from the set of passages by referring to An information processing program as described in Appendix C2.

[0154] (Appendix C4) The second search means and searching the collection of passages for an additional passage using a score obtained by aggregating the first score and the second score for each of the one or more edges. An information processing program as described in Appendix C3.

[0155] (Appendix C5) The second search means A partial directed graph is obtained by referring to the initial passage and the association information, and the partial directed graph constitutes a part of the directed graph, and an additional passage is searched for from the set of passages. An information processing program according to any one of appendices C2 to C4.

[0156] (Appendix C6) a second acquiring means for acquiring input data including a group of sentences; a generation means for generating a set of passages including a plurality of passages included in the set of sentences; The computer a calculation means for calculating the association information including the strength of association between a plurality of passages included in the collection of passages; Further function as An information processing program according to any one of appendices C1 to C5.

[0157] (Appendix C7) The calculation means calculates the association information using a language model. An information processing program as described in Appendix C6.

[0158] (Appendix C8) The computer an acquisition means for acquiring input data including a group of sentences; a generation means for generating a set of passages including a plurality of passages included in the set of sentences; a calculation means for calculating association information including the strength of association between a plurality of passages included in the collection of passages, the association information being referenced in a search process, using a language model; a storage process for storing the association information in association with the plurality of passages; An information processing program that functions as a

[0159] [Appendix D] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0160] (Appendix D1) at least one processor, a first acquisition process for acquiring a query; a first search process for searching a collection of passages, the collection including a plurality of passages, for an initial passage related to the query; a second search process for searching for an additional passage from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage; a third search process that executes a search process using the initial passage and the additional passage; An information processing device that executes the above.

[0161] The information processing device may further include a memory, and the memory may store a program for causing the at least one processor to execute each of the processes.

[0162] (Appendix D2) The second search process includes: A directed graph including one or more edges defined by one or more passage pairs included in the association information is referenced to search for an additional passage from the set of passages. 10. The information processing device according to claim 9, wherein the information processing device is an information processing device according to claim 1.

[0163] (Appendix D3) The second search process includes, for each of the one or more edges, a first score indicating the strength of association between the pair of passages defining the edge, the first score being calculated in advance without reference to the query; and a second score indicating the strength of association between the pair of passages defining the edge, the second score being calculated with reference to the query; Search for additional passages from the set of passages by referring to 10. The information processing device according to claim 9, wherein the information processing device is a

[0164] (Appendix D4) The second search process includes: and searching the collection of passages for an additional passage using a score obtained by aggregating the first score and the second score for each of the one or more edges. 10. The information processing device according to claim 9, wherein the information processing device is an information processing device according to claim 1, wherein the information processing device is an information processing device.

[0165] (Appendix D5) The second search process includes: A partial directed graph is obtained by referring to the initial passage and the association information, and the partial directed graph constitutes a part of the directed graph, and an additional passage is searched for from the set of passages. An information processing device according to any one of appendices D2 to D4.

[0166] (Appendix D6) a second acquisition process for acquiring input data including a group of sentences; a generation process for generating a set of passages including a plurality of passages included in the set of sentences; the at least one processor: a calculation process for calculating the association information including the strength of association between a plurality of passages included in the collection of passages; Run the following again: An information processing device according to any one of appendices D1 to D5.

[0167] (Appendix D7) In the calculation process, the at least one processor calculates the association information using a language model. 10. The information processing device according to claim 9, wherein the information processing device is an information processing device according to claim 8.

[0168] (Appendix D8) The at least one processor: an acquisition process for acquiring input data including a group of sentences; a generation process for generating a set of passages including a plurality of passages included in the set of sentences; a calculation process of calculating, using a language model, association information including strengths of association between a plurality of passages included in the collection of passages, the association information being referenced in the search process; a storage process for storing the association information in association with the plurality of passages; An information processing device that executes the above.

[0169] [Appendix E] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0170] (Appendix E1) A program that causes a computer to function as an information processing device, The computer, a first acquisition process for acquiring a query; a first search process for searching a collection of passages, the collection including a plurality of passages, for an initial passage related to the query; a second search process for searching for an additional passage from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage; a third search process that executes a search process using the initial passage and the additional passage; A non-transitory recording medium on which an information processing program for executing the above is recorded. [Explanation of symbols]

[0171] 1, 100 Information processing equipment 1A Information Processing System 10, 51, 61 Control unit 11(21) Acquisition unit (first acquisition means, second acquisition means) 12 Generation unit (generation means) 13(14) Calculation unit (calculation means, storage means) 22 First search unit (first search means) 23 Second search unit (second search means) 24 Third search unit (third search means) 25 Output data generation unit C1 processor C2 Memory

Claims

1. a first obtaining means for obtaining a query; a first searching means for searching a collection of passages including a plurality of passages for an initial passage related to the query; a second search means for searching for an additional passage from the set of passages by referring to association information including the strength of association between passages included in the set of passages and the initial passage; a third search means for executing a search process using the initial passage and the additional passage; An information processing device comprising:

2. The second search means A directed graph including one or more edges defined by one or more passage pairs included in the association information is referenced to search for additional passages from the set of passages. The information processing device according to claim 1 .

3. The second search means, with respect to each of the one or more edges, a first score indicating the strength of association between the pair of passages defining the edge, the first score being calculated in advance without reference to the query; and a second score indicating the strength of association between the pair of passages defining the edge, the second score being calculated with reference to the query; and Search for additional passages from the set of passages by referring to The information processing device according to claim 2 .

4. The second search means and searching the collection of passages for an additional passage using a score obtained by aggregating the first score and the second score for each of the one or more edges. The information processing device according to claim 3 .

5. The second search means A partial directed graph is obtained by referring to the initial passage and the association information, and the partial directed graph constitutes a part of the directed graph, and an additional passage is searched for from the set of passages. The information processing device according to claim 2 .

6. a second acquiring means for acquiring input data including a group of sentences; a generation means for generating a set of passages including a plurality of passages included in the set of sentences; a calculation means for calculating the association information including the strength of association between a plurality of passages included in the collection of passages; It also has The information processing device according to claim 1 .

7. The calculation means calculates the association information using a language model. The information processing device according to claim 6 .

8. an acquisition means for acquiring input data including a group of sentences; a generation means for generating a set of passages including a plurality of passages included in the set of sentences; a calculation means for calculating association information including the strength of association between a plurality of passages included in the collection of passages, the association information being referenced in a search process, using a language model; a storage means for storing the association information in association with the plurality of passages; An information processing device comprising:

9. Obtaining a query; retrieving an initial passage related to the query from a collection of passages including a plurality of passages; retrieving additional passages from the set of passages by referring to association information including strengths of associations between passages included in the set of passages and the initial passage; performing a search process using the initial passage and the additional passage; An information processing method comprising:

10. A program that causes a computer to function as an information processing device, The program causes the computer to: a first acquisition process for acquiring a query; a first search process for searching a collection of passages, the collection including a plurality of passages, for an initial passage related to the query; a second search process for searching for an additional passage from the set of passages by referring to association information including strengths of associations between passages included in the set of passages and the initial passage; a search process using the initial passage and the additional passage; A program that executes the following.

Images