Information processing device, information processing method, program, and recording medium

Abstract

To generate high-accuracy data while preventing a cost increase.SOLUTION: An information processing device (100) includes: a first acquisition unit (11) that acquires a query; a first generation unit (21) that generates input information including both one or a plurality of object chunks and an instruction sentence according to a first condition included in the query; and a second acquisition unit (12) that acquires output information that is output by a language model to which the input information is input.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a technique for generating information using a language model. [Background technology]

[0002] There is known a technology for providing information in response to a query received from a user. For example, Patent Literature 1 discloses a search device that receives a search query from a user and calculates the degree of compatibility between the search query and each of a plurality of search target data prepared in advance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2021 / 111769 Summary of the Invention [Problem to be solved by the invention]

[0004] Conventional techniques such as that disclosed in Patent Document 1 leave room for improvement in terms of cost and the accuracy of the data provided.

[0005] One aspect of the present invention is to provide a technique that can generate highly accurate data while suppressing increases in costs. [Means for solving the problem]

[0006] In order to solve the above problem, an information processing device according to one embodiment of the present invention includes a first acquisition unit that acquires a query, a first generation unit that generates input information including one or more target chunks and an instruction sentence according to a first condition included in the query, and a second acquisition unit that acquires output information output by a language model to which the input information is input.

[0007] In order to solve the above problems, an information processing device according to one embodiment of the present invention includes a condition generation unit that generates conditions used to extract one or more target chunks from a plurality of chunks, and a chunk acquisition unit that acquires one or more of the target chunks using the conditions.

[0008] In order to solve the above problems, an information processing method according to one embodiment of the present invention is an information processing method executed by an information processing device, and includes a first acquisition step of acquiring a query, a first generation step of generating input information including one or more target chunks and an instruction sentence according to a first condition included in the query, and a second acquisition step of acquiring output information output by a language model to which the input information is input.

[0009] In order to solve the above problems, an information processing method according to one embodiment of the present invention is an information processing method executed by an information processing device, and includes a condition generation step of generating conditions used to extract one or more target chunks from a plurality of chunks, and a chunk acquisition step of acquiring one or more of the target chunks using the conditions.

[0010] The information processing device according to each aspect of the present invention may be realized by a computer. In this case, the present invention also includes a program for realizing the information processing device on a computer by causing the computer to operate as each unit (software element) of the information processing device, and a computer-readable recording medium on which the program is recorded. Effect of the Invention

[0011] According to one aspect of the present invention, highly accurate data can be generated while suppressing increases in costs. [Brief description of the drawings]

[0012] [Figure 1] 1 is a block diagram showing a configuration of an information processing system according to a first embodiment of the present invention. [Diagram 2]1 is a flowchart showing a process flow of an information processing device according to a first embodiment of the present invention. [Diagram 3] FIG. 2 is a diagram for explaining processing by the information processing device according to the first embodiment of the present invention. [Figure 4] FIG. 11 is a block diagram showing a configuration of an information processing system according to a second embodiment of the present invention. [Diagram 5] FIG. 11 is a flowchart showing a process flow by an information processing device according to a second embodiment of the present invention. [Figure 6] FIG. 11 is a diagram showing an example of display information displayed by an information processing device according to a second embodiment of the present invention. [Figure 7] FIG. 11 is a diagram showing an example of display information displayed by an information processing device according to a second embodiment of the present invention. [Figure 8] FIG. 11 is a diagram for explaining a process performed by an information processing device according to a second embodiment of the present invention. [Figure 9] FIG. 11 is a diagram showing an example of input information generated by an information processing device according to a second embodiment of the present invention. [Figure 10] FIG. 11 is a diagram showing an example of answer information generated by the information processing device according to the second embodiment of the present invention. [Figure 11] FIG. 11 is a diagram showing an example of answer information generated by the information processing device according to the second embodiment of the present invention. [Figure 12] FIG. 11 is a flowchart showing a process flow by an information processing device according to a second embodiment of the present invention. [Figure 13] FIG. 11 is a diagram for explaining a process performed by an information processing device according to a second embodiment of the present invention. [Figure 14] FIG. 11 is a flowchart showing a process flow by an information processing device according to a second embodiment of the present invention. [Figure 15] FIG. 11 is a block diagram showing a configuration of an information processing system according to a third embodiment of the present invention. [Figure 16] FIG. 11 is a block diagram showing a configuration of an information processing system according to a fourth embodiment of the present invention. [Figure 17] FIG. 1 is a diagram illustrating a hardware configuration of an information processing system according to each embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] [Embodiment 1] <Outline of Information Processing System 1> Although details will be described later, an information processing system 1 according to one embodiment of the present invention includes an information processing device 100 and a server 200. Here, the server 200 includes a language model LM that generates output data (output information OUT) from a prompt PR that is input data (input information) provided from the information processing device 100.

[0014] On the other hand, the information processing device 100 is roughly Get the query generating input information including one or more target chunks and a directive corresponding to a first condition included in the query; Obtaining output information output by a language model to which the input information is input The configuration is as follows.

[0015] As described above, the information processing device 100 acquires a query, generates input information including one or more target chunks and a directive sentence according to a first condition included in the query, and acquires output information output by a language model to which the input information is input. Therefore, the information processing device 100 can generate highly accurate data while suppressing an increase in cost by appropriately using a language model.

[0016] <Configuration of Information Processing System 1> An information processing system 1 according to an embodiment of the present invention will be described in detail below with reference to the drawings. Fig. 1 is a block diagram showing the configuration of the information processing system 1. As shown in Fig. 1, the information processing system 1 includes an information processing device 100 and a server 200.

[0017] (Configuration of Server 200) First, a description will be given of the configuration of the server 200 included in the information processing system 1. The server 200 includes a control unit 240, a storage unit 220, and a communication unit 230, as shown in FIG.

[0018] (Communication unit 230) The communication unit 230 communicates with a device external to the server 200. As an example, the communication unit 230 communicates with the information processing device 100 included in the information processing system 1. The communication unit 230 transmits data supplied from the control unit 240 to the information processing device 100, and supplies data received from the information processing device 100 to the control unit 240. Note that the data received by the communication unit 230 from the information processing device 100 includes a prompt PR that is input information generated by the information processing device 100. Furthermore, the data provided by the communication unit 230 to the information processing device 100 may include output information OUT generated by the language model LM with reference to the prompt PR.

[0019] (Storage unit 220) The storage unit 220 stores a language model LM. More specifically, the storage unit 220 stores a plurality of parameters that define the language model LM. These parameters are, for example, parameters that have been learned (updated) in advance by machine learning, but this does not limit the present embodiment. In addition, the language model LM is a model that has been machine-learned using documents in various fields as training data, but the details of the language model LM do not limit the present embodiment.

[0020] (Control unit 240) The control unit 240 executes the language model LM to obtain an output result from the language model LM. As an example, the control unit 240 inputs a prompt PR received from the information processing device 100 to the language model LM and obtains output information OUT generated by the language model LM. The control unit 240 also provides the output information OUT to the information processing device 100 via the communication unit 230.

[0021] In the present embodiment, each component of the server 200 is illustrated as a component separate from the information processing device 100, but this does not limit the present embodiment. For example, the control unit 240 of the server 200 or the function of the language model execution unit in the control unit 240 may be configured to be included in the control unit of the information processing device 100. Similarly, the language model LM stored in the storage unit 220 of the server 200 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.

[0022] (Configuration of information processing device 100) Next, a description will be given of the configuration of the information processing device 100 included in the information processing system 1. As shown in FIG.

[0023] In the description here, the information processing device 100 is configured to include the above-mentioned units, but this does not limit the present embodiment. As an example, the above-mentioned units may be distributed among a plurality of devices connected to each other so as to be able to communicate with each other.

[0024] (Communication unit 130) The communication unit 130 communicates with devices external to the information processing device 100. As an example, the communication unit 130 communicates with the server 200. The communication unit 130 transmits data supplied from the control unit 110 to the server 200, and supplies data received from the server 200 to the control unit 110. Note that the data transmitted by the communication unit 130 to the server 200 may include a prompt PR generated by the control unit 110. Furthermore, the data received by the communication unit 130 from the server 200 may include output information OUT generated by the language model LM of the server 200 by referring to the prompt PR.

[0025] (I / O section 140) The input / output unit 140 is configured to include at least one of an input / output device such as a keyboard, a mouse, a display panel, a touch pad, a touch panel, etc. Alternatively, the input / output unit 140 may be configured as an input / output interface to which an input / output device such as a keyboard, a mouse, a display panel, a touch pad, a touch panel, etc. is connected. In the case of this configuration, the input / output unit 140 accepts input of various information to the information processing device 100 from the connected input device. In addition, the input / output unit 140 outputs various information to the connected output device.

[0026] Note that the input / output unit 140 is not an essential component of the information processing device 100. As an example, the information processing device 100 may not include the input / output unit 140, and another device configured to be able to communicate with the information processing device 100 may include the input / output unit 140.

[0027] (Storage unit 120) The storage unit 120 stores various data referenced by the control unit 110 and various data generated by the control unit 110. As an example, the storage unit 120 stores: Document Group DG Chunk group CG Vector group VG · Query QR ·Instruction IN Target chunk TC Output information OUT ·Answer information RI are stored. Here, the document group DG is composed of a plurality of documents acquired in advance. Here, the type, language, acquisition source, size, etc. of the document do not limit this embodiment, and documents having any type, language, acquisition source, size, etc. can be used. Note that, examples of the document group DG include published free papers, paid papers, patent specifications, technical reports, in-house documents, etc. Note that the concepts of "synonyms" and "similar words" described later in this embodiment also include "synonyms" and "similar words" belonging to different languages.

[0028] The chunk group CG is composed of one or more chunks extracted from one or more documents included in the document group DG. Here, a chunk refers to the document itself, or one or more components included in the document. Examples of chunks include the document itself, sentences, paragraphs, phrases, words, characters, numbers, symbols, etc. included in the document, but these examples do not limit the present embodiment.

[0029] The vector group VG is composed of vectors (also called feature vectors in the feature space) for one or more chunks included in the chunk group CG. Here, the conversion from the chunks to the feature vectors (embedding of the chunks in the feature space) can be performed in advance using a predetermined algorithm.

[0030] In this embodiment, it is not essential that the information processing device 100 holds the vector group VG, and that it performs various processes using the vectors included in the vector group VG.

[0031] In the explanation of each embodiment, an example is given in which each unit of control unit 110 mainly performs processing with reference to a "chunk," but this example does not limit each embodiment, and it is not essential to the processing of each embodiment that a document is divided into one or more documents in advance. For example, in each processing described in each embodiment, the processing will still be valid even if a "chunk" and a "target chunk" are replaced with a "document" and a "target document," respectively, and such an example is also included in each embodiment.

[0032] The query QR is data indicating an inquiry, request, demand, etc. from a user, and is acquired, for example, via the input / output unit 140 of the information processing device 100. A specific example of the data included in the query QR does not limit the present embodiment, but an example of the query QR is data indicating a condition (also called a first condition) COND1 and data indicating one or more items (also written as ITEM1, etc.) input in association with the condition. The data included in the query QR may be expressed in the form of text data, for example, but this does not limit the present embodiment.

[0033] Furthermore, when the query QR is expressed in the form of text data, the language of the text does not limit this embodiment. Furthermore, the language of the query QR may be a language different from that of the documents included in the document group DG described above. Furthermore, the language of the query QR may be a language different from that of the chunks included in the chunk group CG described above.

[0034] The prompt PR is information generated by a first generating unit 21, which will be described later. For example, At least one of the chunks included in the chunk group CG is included as the target chunk TC, and It further comprises an instruction IN obtained from said query QR. The prompt PR is an example of input information that is input to the language model LM.

[0035] The target chunk TC and the instruction sentence IN are, for example, text data, and the prompt PR is also configured to include these text data, but this does not limit the present embodiment. The prompt PR may be configured to include data of any format, such as image data, graph data, and table data, in addition to text data.

[0036] The output information OUT is information output by the language model LM to which the prompt PR is input, and is acquired by a second acquisition unit 12 (described later) via the communication unit 130, for example. A specific example of the output information OUT will be described later.

[0037] The answer information RI is information generated by the answer information generating unit 23 described later with reference to the output information OUT. The answer information RI is, for example, information including an answer to the above-mentioned query QR. A specific example of the answer information RI will be described later.

[0038] (Control unit 110) As shown in FIG. 1, the control unit 110 includes a first acquisition unit 11, a first generation unit 21, a second acquisition unit 12, and a response information generation unit 23.

[0039] (First Acquisition Unit 11) The first acquisition unit 11 acquires a query QR. As an example, the first acquisition unit 11 acquires a query QR according to an input from a user via the input / output unit 140. The acquired query QR is, as an example, stored in the storage unit 120 and is referred to by the first generation unit 21. As an example, the query QR may include the first condition COND1 as described above.

[0040] (First generation unit 21) The first generation unit 21 generates a prompt PR (input information) including one or more target chunks TC and an instruction sentence IN corresponding to a first condition COND1 included in the query QR. The generated prompt PR is provided to the language model LM of the server 200 via the communication unit 130, for example.

[0041] 1, the first generation unit 21 includes, for example, a command statement generation unit 211 and a first display data generation unit 212. However, the structure of the first generation unit 21 is merely an example, and the first generation unit 21 may have any configuration as long as it is configured to be able to execute the processes by these components.

[0042] The instruction statement generation unit 211 included in the first generation unit 21 generates an instruction statement IN from the query QR. As an example, the instruction statement generation unit 211 generates an instruction statement IN according to a first condition COND1 and one or more items (ITEM1, etc.) included in the query QR. Specific examples of instruction statements generated by the instruction statement generation unit 211 will be described later.

[0043] On the other hand, the first display data generating unit 212 included in the first generating unit 21 generates display data including at least a part of the prompt PR (input information). The display data generated by the first display data generating unit 212 is visually presented to the user via a display panel or a touch panel included in the input / output unit 140, for example.

[0044] The first generating unit 21 may be configured to obtain a correction instruction from a user who has viewed the display data generated by the first display data generating unit 212, and to correct the prompt PR in accordance with the obtained correction instruction. In this case, the first generating unit 21 may be configured to provide the corrected prompt PR to the language model LM of the server 200.

[0045] (Second Acquisition Unit 12) The second acquisition unit 12 acquires output information OUT output by the language model LM to which the prompt PR or the corrected prompt PR is input. The acquired output information OUT is stored in the storage unit 120, for example, and is referred to by the answer information generation unit 23, which will be described later. Specific examples of the output information OUT will be described later.

[0046] (Answer information generation unit 23) The answer information generating unit 23 generates answer information RI by referring to the output information OUT output by the language model LM. A specific example of the answer information RI will be described later. The answer information generating unit 23 includes an answer information display data generating unit 231 as shown in FIG. 1 as an example. However, the structure of the answer information generating unit 23 is merely an example, and the answer information generating unit 23 may have any configuration as long as it is configured to be able to execute processing by the components. The answer information display data generating unit 231 included in the answer information generating unit 23 generates display data including the answer information RI. A display example of the answer information RI will be described later.

[0047] (Processing flow by information processing device 100) Next, the flow of processing by the information processing device 100 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of an information processing method S100 by the information processing device 100.

[0048] (Step S101) In step S101, the first acquisition unit 11 acquires a query QR. The upper part of Fig. 3 is a diagram showing an example of the query QR acquired by the control unit 110 via the input / output unit 140 in this step. As shown in Fig. 3, the query QR in this example includes a conditional statement (first condition) COND1 and an answer item ITEM1. As an example, the first acquisition unit 11 acquires the query QR via a UI (User Interface) screen configured to be able to accept the conditional statement COND1 and the answer item ITEM1.

[0049] Here, the first condition COND1 is a condition for specifying what kind of information the user wants to obtain using the information processing device 100. The example shown in FIG. 3 shows a case where the information processing device 100 is applied to the field of materials informatics, and the user specifies, as the first condition COND1, "1.60×10 -6 cm 23 shows an example where the user has input "a composite membrane having a methanol permeability of 0.1 μm to 100 μm / s." Furthermore, answer item ITEM1 indicates an item that the user desires as an answer by the information processing device 100. In the example shown in FIG. 3, an example where the user has input "thin film composition" is shown as answer item ITEM1.

[0050] In this way, when the information processing device 100 is applied to the field of materials informatics, the query QR may include, for example, at least one of a property name, a substance name, a composition name, a property value, a unit, an experimental condition, and an experimental apparatus. However, these conditions and items are merely examples, and the first acquisition unit 11 may be configured to acquire other information as the query QR, and the information processing device 100 is applicable to any field.

[0051] In this embodiment, it is not essential that the query QR includes both the condition statement (first condition) COND1 and the answer item ITEM1 as shown in the upper part of Fig. 3. As an example, the query QR may include the condition statement (first condition) COND1 but not the answer item ITEM1, and the process in step S101 may be performed by sub-steps S1011 to S1013 as follows.

[0052] (Substep S1011) Query QR is "1.60 × 10 -6 cm 2 A query including the text "thin film composition of a composite membrane having a methanol permeability of 0.1 μm / s or less" is input, and the first acquisition unit 11 acquires the query QR.

[0053] (Substep S1012) The first acquisition unit 11 analyzes the input query QR and acquires the query QR in accordance with the first condition COND1 “1.60×10 -6 cm 2 / s” and answer item ITEM1 “Thin film composition.” This is based on the first condition COND1 “1.60×10 -6 cm2 It may also be expressed as extracting (generating) the composite membrane having a methanol permeability of 1.60×10⁻⁷ cm / s or less and the answer item ITEM1 "thin film composition". Further, the first acquisition unit 11 may perform the above processing using the language model provided in the information processing apparatus 100, or may perform the processing using the language model LM provided in the server 200. When the first acquisition unit 11 performs the processing of sub-steps S1011 and S1012, the expression "the first condition and the answer item included in the query QR" in the following description shall be read as "the first condition and the answer item extracted from the query QR".

[0054] (Step S102) In step S102, the first generation unit 21 selects a target chunk TC from a plurality of chunks included in the chunk group CG. Here, the method of selecting the target chunk TC is not limited to the present embodiment, but as an example, the target chunk TC may be selected in ascending or descending order of the chunk ID assigned to each of the plurality of chunks included in the chunk group CG.

[0055] (Step S103) In step S103, the first generation unit 21 generates an instruction text IN by referring to the query QR by the instruction text generation unit 211. Further, the first generation unit 21 generates a prompt PR including the generated instruction text IN and one or more target chunks TC selected in step S102.

[0056] In the lower part of FIG. 3, an example of the prompt PR generated by the first generation unit 21 in this step is shown. As shown in the lower part of FIG. 3, in this step, the instruction text generation unit 211 refers to · Conditional statement (first condition): "A composite membrane having a methanol permeability of 1.60×10⁻⁷ cm / s or less" -6 cm 2 / s or less" · Answer item: "Thin film composition" and, as an instruction text IN corresponding thereto, · "From the following text, '1.60×10⁻⁷ cm / s or less methanol permeability composite membrane' and 'thin film composition' are extracted." -6 cm2 Please extract the "thin film composition" of the "composite membrane with a methanol permeability of 100 / s or less" and answer in list form. is generated.

[0057] However, the example of the prompt shown in the lower part of Fig. 3 is merely an example and does not limit the present embodiment. The prompt PR generated by the first generation unit 21 may be in any format as long as it is a prompt that instructs the language model LM to output an answer item that satisfies a conditional statement (first condition).

[0058] Depending on the language model LM, it may be possible to set a role in the prompt PR. In such a case, the first generation unit 21 may be configured to set a role in the prompt PR according to the specifications of the language model LM, or to divide the prompt PR into a plurality of prompts and set a role in each prompt.

[0059] In addition, in this step, the first generation unit 21 Present the display data generated by the first display data generating unit 212 to the user; Acquire correction instructions from users who viewed the display data, - Modify the above prompt PR according to the obtained modification instructions The above process may be further carried out.

[0060] (Step S104) In step S104, the first generation unit 21 provides the prompt PR generated in step S103, or the prompt PR generated in step S103 and modified by a user instruction, to the server 200 via the communication unit 130, and inputs the prompt PR to the language model LM. The language model LM to which the prompt has been input generates output information OUT according to the prompt PR.

[0061] (Step S105) In step S105, the second acquisition unit 12 acquires the output information OUT output by the language model LM. The acquired output information OUT is stored in the storage unit 120, for example, and is referred to by the answer information generation unit .

[0062] (Step S106) In step S106, the answer information generating unit 23 adds the output information OUT acquired in step S105 to the result list.

[0063] (Step S107) In step S107, the answer information generation unit 23 determines whether there are any unprocessed chunks among the multiple chunks included in the chunk group, or among the multiple chunks extracted from the multiple chunks and held as candidates for the target chunk TC.

[0064] If there is an unprocessed chunk (YES in step S107), the process returns to step S102, a target chunk TC is selected from the unprocessed chunks, and the process from step S103 onward is performed. On the other hand, if there is no unprocessed chunk (NO in step S107), the process proceeds to step S108.

[0065] (Step S108) In step S108, the answer information generating unit 23 generates answer information RI with reference to the output information OUT. Generate answer information RI by referring to at least a part of the output information OUT stored in the result list; · The output information OUT accumulated in the result list is integrated, and answer information RI is generated by referring to the integrated answer information.

[0066] In this way, when the information processing device 100 is applied to the field of materials informatics, the response information RI may include literature information and at least any of physical property names, substance names, composition names, physical property values, units, experimental conditions, and experimental equipment.

[0067] As described above, the information processing device 100 acquires a query, generates input information including one or more target chunks and a directive sentence according to a first condition included in the query, and acquires output information output by a language model to which the input information is input. Therefore, the information processing device 100 can generate highly accurate data while suppressing an increase in cost by appropriately using a language model.

[0068] (Additional Notes for the First Embodiment) The information processing device 100 according to the first embodiment can be applied to data generation in any field. In the above description, the information processing device 100 is mainly applied to data generation in the field of materials informatics, but the present invention is not limited to this.

[0069] As an example, the first acquisition unit 11 receives the following as the query QR: Conditional sentence COND1: "Fire in a house inhabited only by people aged 70 or older" - Answer item 1 "Date of occurrence" The first generating unit 21 may generate a command sentence IN by referring to the query QR, and generate a prompt PR including the generated command sentence IN and one or more target chunks TC. Here, as one example, the target chunk TC may be a chunk selected from a chunk group including news. The second acquiring unit 12 may acquire output information OUT output by the language model LM to which the prompt PR is input, and the answer information generating unit 23 may generate answer information RI by referring to the output information OUT.

[0070] On the other hand, in materials informatics and related fields, there are many papers and technical reports that can be referred to when generating data to be provided to users. For this reason, it is not easy to ensure the accuracy of data while suppressing costs. According to the above-mentioned information processing device 100, a query is acquired, input information including one or more target chunks and a directive statement corresponding to a first condition included in the query is generated, and output information output by a language model to which the input information is input is acquired. Therefore, according to the information processing device 100, by suitably using a language model, it is possible to generate highly accurate data while suppressing an increase in costs in materials informatics and related fields.

[0071] [Embodiment 2] Other embodiments of the present invention will be described below. For the sake of convenience, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will be omitted.

[0072] <Overview of Information Processing System 1A> Although details will be described later, an information processing system 1A according to one embodiment of the present invention includes an information processing device 100A and a server 200. Here, similar to the first embodiment, the server 200 includes a language model LM that generates output data (output information OUT) from a prompt PR that is input data (input information) provided from the information processing device 100A.

[0073] On the other hand, the information processing device 100A is, roughly speaking, Get the query generating criteria used to extract one or more target chunks from the plurality of chunks; Using said condition, obtain one or more target chunks; generating input information including at least a portion of the one or more target chunks and a directive obtained from the query; Obtaining output information output by a language model to which the input information is input The configuration is as follows.

[0074] As described above, the information processing device 100A generates extraction conditions, extracts a target chunk from a plurality of chunks using the extraction conditions, and inputs input information including at least a part of the target chunk into a language model. Therefore, the information processing device 100A can generate highly accurate data while suppressing an increase in cost.

[0075] <Configuration of Information Processing System 1A> Hereinafter, the information processing system 1A according to this embodiment will be described in detail with reference to the drawings. Fig. 4 is a block diagram showing the configuration of the information processing system 1A. As shown in Fig. 4, the information processing system 1A includes an information processing device 100A and a server 200. Here, the configuration of the server 200 is the same as that described in the first embodiment, and therefore the description thereof will be omitted.

[0076] (Configuration of information processing device 100A) 4, the information processing device 100A includes a control unit 110A, a storage unit 120A, a communication unit 130, and an input / output unit 140. Here, the communication unit 130 and the input / output unit 140 have the same configuration as those described in the first embodiment, and therefore the description thereof will be omitted.

[0077] (Storage unit 120A) The storage unit 120 stores various data referenced by the control unit 110A and various data generated by the control unit 110A. Document Group DG Chunk group CG Vector group VG · Query QR ·Condition Information CI ·Instruction IN Target chunk TC Output information OUT ·Answer information RI Here, the document group DG, chunk group CG, and vector group VG have been explained in the first embodiment, so explanations thereof will be omitted here.

[0078] As in the first embodiment, the query QR is data indicating an inquiry, request, demand, etc. from a user, and is acquired via the input / output unit 140 of the information processing device 100A, as an example. A specific example of the data included in the query QR does not limit this embodiment, but an example of the query QR is data indicating a condition (also called a first condition) COND1 input via a display screen DIS0 described later, and data indicating one or more items (ITEM1, ITEM2, etc.) input in association with the condition. As an example, the data included in the query QR may be expressed in the form of text data, but this does not limit this embodiment.

[0079] The condition information CI is information including a condition (also called a second condition) COND2 used to extract one or more target chunks TC from a plurality of chunks. The second condition COND2 is generated by a second generating unit 22 (described later) and is referred to by a third acquiring unit 13 (described later), for example.

[0080] Although the specific configuration of the second condition COND2 does not limit the present embodiment, as an example, the second condition COND2 is configured by one or more keywords. A more specific example of the second condition COND2 will be described later.

[0081] The prompt PR is information generated by the first generator 21, as in the first embodiment, and includes, for example, at least a part of the one or more target chunks TC and an instruction sentence IN obtained from the query QR. The prompt PR is an example of input information input to the language model LM.

[0082] As in the first embodiment, the target chunk TC and the instruction sentence IN are, for example, text data, and the prompt PR is also configured to include these text data, but this does not limit the present embodiment. The prompt PR may be configured to include data of any format, such as image data, graph data, and table data, in addition to text data.

[0083] The output information OUT is information output by the language model LM to which the prompt PR is input, and, as an example, is acquired by the second acquisition unit 12 via the communication unit 130, similarly to the first embodiment. A specific example of the output information OUT will be described later.

[0084] The answer information RI is information generated by the answer information generating unit 23 with reference to the output information OUT, as in the first embodiment. The answer information RI is, for example, information including an answer to the query QR, as in the first embodiment. A specific example of the answer information RI will be described later.

[0085] (Control unit 110A) 4, the control unit 110A includes a first acquisition unit 11, a first generation unit 21, a second acquisition unit 12, a second generation unit 22, a third acquisition unit 13, and a response information generation unit 23. Here, the first acquisition unit 11, the first generation unit 21, and the second acquisition unit 12 have the same configurations as those described in the first embodiment, and therefore the description thereof will be omitted here.

[0086] (Second generation unit 22) 4, the second generating unit 22 includes, as an example, a condition generating unit 221, a condition modifying unit 222, and a second display data generating unit 223. However, the structure of the second generating unit 22 is merely an example, and the second generating unit 22 may have any configuration as long as it is configured to be able to execute the processes by these components.

[0087] The condition generating unit 221 included in the second generating unit 22 generates a condition (second condition COND2) used to extract one or more target chunks TC from a plurality of chunks included in the chunk group CG.

[0088] As an example, the condition generating unit 221 generates the second condition COND2 by referring to the first condition COND1 included in the query QR. As a more specific example, the condition generating unit 221 generates the second condition COND2 so as to correspond to the first condition COND1 included in the query QR.

[0089] The second display data generating unit 223 included in the second generating unit 22 generates display data including at least a part of the second condition COND2. The display data generated by the second display data generating unit 223 is visually presented to the user via a display panel or a touch panel included in the input / output unit 140, for example.

[0090] The condition modification unit 222 included in the second generation unit 22 modifies the second condition COND2 in response to an instruction from a user. As an example, the condition modification unit 222 acquires a modification instruction from a user who has viewed the display data generated by the second display data generation unit 223, and modifies the second condition COND2 in response to the acquired modification instruction. The condition modification unit 222 modifies the second condition COND2 in response to the modification instruction from the user, thereby making the second condition COND2 more preferable to the user.

[0091] The second generating unit 22 may be configured not to include the condition modifying unit 222 and the second display data generating unit 223, and such a configuration is also included in the present embodiment.

[0092] (Third Acquisition Section 13) The third acquisition unit 13 acquires one or more target chunks TC from multiple chunks included in the chunk group CG using the second condition COND2 generated by the condition generation unit 221, or the second condition COND2 generated by the condition generation unit 221 and modified by the condition modification unit 222.

[0093] As an example, the third acquisition unit 13 acquires one or more target chunks TC by applying an extraction process using the second condition COND2 to a plurality of chunks included in the chunk group CG. Here, the extraction process is ·Word matching using keywords included in the second condition COND2 · Neighborhood vector search by referring to the feature vector corresponding to the second condition COND2 and the feature vector of each chunk The above processing may be included.

[0094] Furthermore, the third acquisition unit 13 may be configured to acquire, as the extraction process, one or more chunks that match the second condition COND2 from among the chunks included in the chunk group CG as one or more target chunks TC. Here, the "chunk that matches the second condition COND2" refers, for example, to a chunk that includes one or more keywords included in the second condition COND2, or a synonym or similar word of the one or more keywords (the same applies below). However, this example does not limit the present embodiment.

[0095] According to the above configuration, since the chunks that match the second condition COND2 are acquired as the target chunks, the third acquisition unit 13 can preferably narrow down the chunks. Therefore, the cost of the process that refers to the chunks can be preferably reduced.

[0096] In addition, as another example, the third acquisition unit 13 Among the chunks included in the chunk group CG, one or more chunks that match the second condition COND2 are obtained, Obtain one or more target chunks TC by applying a dilation process to one or more chunks that meet the second condition COND2. The above configuration may also be used.

[0097] Here, the "enlargement process" refers to, for example, the following process for the target chunk TC: One or more chunks CK that match the second condition COND2 In addition, - A chunk that is contextually located either before or after the chunk CK This refers to the process of including.

[0098] As a more specific example, "enlargement processing" includes the following: Calculating a similarity between one or more chunks CK1 that satisfy the second condition COND2 and a chunk CK2 that is contextually located before or after the chunks CK1, If the calculated similarity is equal to or greater than a predetermined threshold, in addition to the chunk CK1, the chunk CK2 is also included in the target chunk TC. In addition, as an "enlargement process", these processes may be repeated for multiple chunks CK1. For example, as an "enlargement process", Add the chunk CK2 to the target chunk TC, and calculate the similarity between the chunk CK3 that is contextually located before or after the chunk CK2 and the chunk CK1 or chunk CK2, If the calculated similarity is equal to or greater than a certain threshold, the chunk CK3 is included in the target chunk TC. Furthermore, depending on the similarity between the chunk CK3 and the chunks before and after it, the chunks before and after it are added to the target chunk. Here, in the process of calculating the similarity between chunks, the similarity between feature vectors of each chunk may be used.

[0099] As another example, "enlargement processing" includes determining whether one or more chunks CK that match the second condition COND2 contain an anaphor; If the word contains an anaphoric word, a process is performed to determine what the anaphoric word refers to; If the target can be determined, the chunk containing the target is included in the target chunk TC. If the target cannot be determined, add the chunk contextually preceding the chunk CK to the target chunk TC. Here, the above-mentioned "processing of determining what the anaphor points to" may or may not be a process using the language model LM.

[0100] As another example, "enlargement processing" includes The target chunk TC may be configured to include up to N chunks that are contextually before one or more chunks CK that match the second condition COND2, and up to N chunks that are contextually after the one or more chunks CK.

[0101] As described above, by applying the enlargement process, the third acquisition unit 13 can prepare an appropriate amount of target chunk TC, thereby improving the accuracy of the data generated by referencing the chunk.

[0102] (Processing flow by information processing device 100A) Next, the flow of processing by the information processing device 100A according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a flow diagram showing the flow of an information processing method S200 by the information processing device 100A.

[0103] (Step S101) In step S101, the first acquisition unit 11 acquires a query QR. Fig. 6 is a diagram showing a display screen DIS0 which is an example of a display screen displayed by the control unit 110 on the input / output unit 140 in this step. As shown in Fig. 6, the display screen DIS0 is a UI (User Interface) screen configured to be able to accept a first condition COND1, an answer item ITEM1, and an associated item ITEM2. The first acquisition unit 11 acquires the first condition COND1, the answer item ITEM1, and the associated item ITEM2 accepted via the display screen DIS0 as a query QR.

[0104] Here, the first condition COND1 is a condition for specifying what information the user wants to obtain using the information processing device 100A. The example shown in FIG. 6 shows a case where the information processing device 100A is applied to the field of material informatics, and an example is shown in which the user inputs "increase the hardness of polylactic acid" as the first condition COND1. Furthermore, the answer item ITEM1 indicates an item desired by the user as an answer by the information processing device 100A. In the example shown in FIG. 6, an example is shown in which the user inputs "filler" as the answer item ITEM1. Furthermore, the auxiliary item ITEM2 indicates an item that the user wants to additionally obtain as an answer by the information processing device 100A. In the example shown in FIG. 6, an example is shown in which the user inputs "altitude, weight ratio" as the auxiliary item ITEM2.

[0105] In this way, when the information processing device 100A is applied to the field of materials informatics, the query QR may include, for example, at least one of a property name, a substance name, a composition name, a property value, a unit, an experimental condition, and an experimental apparatus. However, these conditions and items are merely examples, and the first acquisition unit 11 may be configured to acquire other information as the query QR, and the information processing device 100A is applicable to any field.

[0106] Also, as shown in Fig. 6, the display screen DIS0 may include an area (COST1 in Fig. 6) for displaying an estimated cost for processing using the query QR. Also, as shown in Fig. 6, the display screen DIS0 may include an area (COST1 in Fig. 6) for displaying an estimated cost for processing using the query QR. "Target Designation" button B1 for designating targets "Start Extraction" button B2 to start extraction "Download" button B3 for downloading extracted documents, etc. "Scoring" button B4 for scoring extracted documents, etc. The configuration may include at least a part of the above.

[0107] (Steps S202 to S203) Next, in step S202, the second generation unit 22 generates a second condition COND2 by referring to the first condition COND1 included in the query QR. Then, in step S203, the second generation unit 22 presents the second condition COND2 to the user via the input / output unit 140.

[0108] 7 is a diagram showing a display screen DIS1 including the second condition COND2 generated in step S202 and presented to the user in step S203. The display screen DIS1 is an example of display data generated by the second display data generating unit 223. By presenting the display data, the user can easily grasp the content of the second condition COND2.

[0109] As shown in Fig. 7, the second condition COND2 includes one or more keywords for narrowing down the chunks included in the target chunk group CG. For example, in the example shown in Fig. 7, the second condition COND2 includes "polylactic acid" and "hardness" as the keywords. Here, these keywords are generated by the second generation unit 22 with reference to the first condition and one or more items included in the query QR.

[0110] For example, keyword 1 "polylactic acid" is a keyword generated (obtained) by the second generation unit 22 as a synonym or similar word of "resin" included in the query QR. Also, keyword 2 "hardness" is a keyword generated (obtained) by the second generation unit 22 as a synonym or similar word of "hardness" included in the query QR.

[0111] Also, as shown in FIG. 7, the second condition COND2 may be configured to include information indicating whether to perform an "AND" or "OR" search for each keyword (corresponding to "AND" and "OR" on the display screen DIS1 in FIG. 7).

[0112] The above-mentioned keyword generation (acquisition) process may include a process using the language model LM. For example, the second generation unit 22 may inquire the language model LM about one or more keywords for thoroughly searching each piece of information included in the query QR, and generate the second condition COND2 by referring to a response from the language model LM. As a more specific example, the second generation unit 22 may generate the following as an inquiry prompt using an arbitrary natural number N (e.g., N=3) that can be set by the user: "Please select N keywords that are appropriate for searching information on "increasing the hardness of polylactic acid." may be input to the language model LM, and the language model LM may output a second condition COND2 including the keywords "polylactic acid" and "hardness."

[0113] Alternatively, the second generation unit 22 · Manage multiple entity types in advance, Determining whether or not an entity corresponding to at least any of the plurality of entity types is present in a first condition COND1 included in the query QR; If it is determined that the entity exists, the entity is included as a keyword in the second condition COND2. Here, the entity type may be, for example, a substance name, a physical property name, a unit, etc., but this is not a limitation of the present embodiment. At least a part of the above process may be performed using a language model LM.

[0114] The second generating unit 22 may be configured to further add a synonym of the generated (acquired) keyword as a keyword to the second condition COND2. In this case, the added keyword may be included in the second condition COND2 as a search keyword based on the "OR" condition.

[0115] (Step S204) In step S204, the second generation unit 22 accepts an input from the user and modifies the second condition COND2 in response to the input. In the example shown in Fig. 7, the second generation unit 22 accepts a keyword 4 from the user and modifies the second condition COND2 to include the accepted keyword.

[0116] In addition, if the second generation unit 22 is configured not to include the condition modification unit 222 and the second display data generation unit 223, the process of presenting the second condition COND2 to the user and the process of modifying the second condition COND2 in accordance with input from the user in the above-mentioned steps S203 to S204 are omitted.

[0117] (Step S205) In step S205, the third acquisition unit 13 acquires one or more target chunks TC from the multiple chunks included in the chunk group CG, using the second condition COND2 generated by the second generation unit 22 in step S202, or the second condition COND2 generated by the second generation unit 22 in step S202 and modified by the second generation unit 22 in step S204. As an example, the third acquisition unit 13 acquires, from the multiple chunks included in the chunk group CG, one or more chunks that match the second condition COND2 as the one or more target chunks TC.

[0118] Fig. 8 is a diagram showing a schematic diagram of the process by the third obtaining unit 13 in this step. In the example shown in Fig. 8, the chunk group CG includes components included in a plurality of papers as chunks. More specifically, the chunk group CG shown in Fig. 8 includes the description of "experimental results explanation" in each of papers 1 to 3 as chunks C1 to C3.

[0119] In this step, the third acquisition unit 13 narrows down the search using the keywords included in the second condition COND2 described above, thereby excluding chunk C1 and extracting chunks C2 and C3. In this example, chunk C1 does not include a wording that corresponds to a keyword included in the second condition COND2, and is therefore excluded by the third acquisition unit 13. On the other hand, chunks C2 and C3 include a wording that corresponds to a keyword included in the second condition COND2, and are therefore extracted by the third acquisition unit 13. The chunks C2 and C3 extracted in this manner constitute the target chunk TC or the target chunk group TCG1 as the target chunk TC2 and the target chunk TC3.

[0120] (Step S206) In step S206, the third acquisition unit 13 expands the target chunk TC acquired in step S205. As an example, the third acquisition unit 13 One or more chunks CK (chunks C2 and C3 in FIG. 8) that are acquired in step S205 and match the second condition COND2 In addition, A chunk that is contextually located either before or after the chunk CK (chunks C2 and C3 in FIG. 8). is included in the target chunk TC.

[0121] In the example shown in FIG. 8, the third acquisition unit 13 includes in the target chunk TC, for Paper 2, the description of "experimental apparatus description" that is contextually located before chunk C2, and the description of "conclusion" that is contextually located after chunk C2. Also, in the example shown in FIG. 8, the third acquisition unit 13 further includes in the target chunk TC, for Paper 3, the description of "experimental apparatus description" that is contextually located before chunk C3, and the description of "conclusion" that is contextually located after chunk C3. In FIG. 8, the expanded target chunk obtained by expanding chunk C2 is represented as EC2 or TC2', and the expanded target chunk obtained by expanding chunk C3 is represented as EC3 or TC3'. However, these notations do not limit the present embodiment.

[0122] The third acquisition unit 13 may also perform chunk expansion processing by referring to anaphoric words. For example, assume that the chunk C2 obtained as the target chunk group TCG1 has a description "The strength of the aforementioned sample is...". Then, assume that the chunk C2a located before the chunk C2 has a description "In the sample, XX is added at a weight ratio of △%". In this case, the third acquisition unit 13 may identify that "the aforementioned sample" and "in the sample" are in an anaphoric relationship, and expand the target chunk C2 to include the chunk C2a. In other words, the third acquisition unit may expand the target chunk C2 to include another chunk including a word that is in an anaphoric relationship with the word included in the target chunk.

[0123] In addition, the specific content of the "enlargement process" executed by the third acquisition unit 13 in this step will not be described here because a detailed example has already been described above. In addition, this step does not limit the present embodiment, and a process flow that does not include this step is also included in the present embodiment.

[0124] (Step S102) Next, in step S102, the first generating unit 21 selects a target chunk TC to be included in the prompt PR in step S103, which will be described later, from at least one of the one or more target chunks TC acquired in step S205 and one or more target chunks TC expanded in step S206 (these target chunks may also be called target chunk candidates). Note that in this step, all of the one or more target chunk candidates may be selected as the target chunk TC.

[0125] (Step S103) In step S103, the first generating unit 21 generates an instruction sentence IN by referring to the query QR, and The instruction IN, Among the target chunk TC acquired in step S205 or the expanded chunk EC expanded in step S206, the target chunk TC selected in step S102 and Generate a prompt PR containing:

[0126] 9 is a diagram showing an example of a prompt PR, PR1, generated in this step by the first generation unit 21. The prompt PR1 shown in FIG. 9 is also an example of display data (display screen DIS2) generated by the first display data generation unit 212.

[0127] As shown in FIG. 9, the prompt PR1 is an example of an instruction sentence IN, which is as follows: From the search results below, A "filler" that has the property of "increasing the hardness of polylactic acid" - The "hardness" and "weight ratio" of the "filler" Please extract it. Also, output the chunk ID to be searched for as the source of extraction. Please answer the extracted results in a format using "|" as a delimiter. The directive IN1 includes the following directive: - First condition: "Increase the hardness of polylactic acid" Answer: "Filler" -Additional items: "Hardness, weight ratio" This is an instruction sentence generated by the first generation unit 21 with reference to the query QR including the above.

[0128] More specifically, the first generation unit 21 refers to the first condition “increase the hardness of polylactic acid” and the answer item “filler” included in the query QR, and generates the instruction information in the instruction sentence IN1 as Instructions to extract a "filler" that has the property of "increasing the hardness of polylactic acid" In addition, the first generating unit 21 refers to the answer item "filler" and the additional item "hardness, weight ratio" included in the query QR to generate instruction information in the instruction sentence IN1. Instructions to extract the "hardness" and "weight ratio" of the "filler" Generate.

[0129] In addition, in the instruction sentence IN1 generated by the first generation unit 21, "Please output the chunk ID to be extracted." This instruction information is included in the instruction statement IN1 so that the control unit 110A can specify which chunk "filler", "hardness", and "weight ratio" were extracted from. For example, the control unit 110A can specify which document "filler", "hardness", and "weight ratio" were extracted from by managing the chunk ID in association with the document ID from which the target chunk was extracted.

[0130] In addition, in the instruction sentence IN1 generated by the first generation unit 21, - "Please reply with the extracted results in a format using "|" as a separator." The instruction information is included in the instruction sentence IN1 by the first generating unit 21 so that the output information by the language model LM conforms to the answer format specified by the user. The instruction information can be determined in advance by the first generating unit 21 so that the answer format is as desired by the user.

[0131] 9, the prompt PR1 includes the target chunk EC2 (TC2') and the target chunk EC3 (TC3') expanded in step S206 as the "search target" mentioned in the instruction IN1. Since the prompt PR1 includes these target chunks as search targets, the language model LM to which the prompt PR1 is input searches these target chunks according to the instruction IN1, A "filler" that has the property of "increasing the hardness of polylactic acid" - "Hardness" and "weight ratio" of the "filler" The following will be extracted.

[0132] In this step, the first display data generation unit 212 included in the first generation unit 21 may present the prompt PR1 to the user as a display screen DIS2 via a display panel or a touch panel included in the input / output unit 140. According to this configuration, the user can easily check the generated prompt PR before inputting it into the language model LM.

[0133] In addition, in this step, the first generation unit 21 may receive a correction instruction from a user who views the display screen DIS2, and may correct the prompt PR1 based on the correction instruction. With this configuration, the prompt PR can be made more preferable to the user.

[0134] 9 is merely an example and does not limit the present embodiment. The prompt PR generated by the first generation unit 21 may be in any form as long as it is a prompt that instructs the language model LM to output a response to an instruction indicated by the instruction information generated by referring to the first condition and answer item included in the query QR.

[0135] Depending on the language model LM, it may be possible to set a role in the prompt PR. In such a case, the first generation unit 21 may be configured to set a role in the prompt PR according to the specifications of the language model LM, or to divide the prompt PR into a plurality of prompts and set a role in each prompt.

[0136] (Step S104) In step S104, the first generation unit 21 provides the prompt PR generated in step S103, or the prompt PR generated in step S103 and modified by a user instruction, to the server 200 via the communication unit 130, and inputs the prompt PR to the language model LM. The language model LM to which the prompt has been input generates output information OUT according to the prompt PR.

[0137] (Step S105) In step S105, the second acquiring unit 12 acquires the output information OUT output by the language model LM. The acquired output information OUT is stored in the storage unit 120A, for example, and is referred to by the answer information generating unit .

[0138] (Step S106) In step S106, the answer information generating unit 23 adds the output information OUT acquired in step S105 to the result list.

[0139] (Step S107) In step S107, the answer information generation unit 23 determines whether there are any unprocessed chunks among the multiple chunks included in the chunk group, or among the multiple chunks extracted from the multiple chunks and held as candidates for the target chunk TC.

[0140] If there is an unprocessed chunk (YES in step S107), the process returns to step S102, a target chunk TC is selected from the unprocessed chunks, and the process from step S103 onward is performed. On the other hand, if there is no unprocessed chunk (NO in step S107), the process proceeds to step S108.

[0141] (Step S108) In step S108, the answer information generating unit 23 generates answer information RI with reference to the output information OUT. Generate answer information RI by referring to at least a part of the output information OUT stored in the result list; · The output information OUT accumulated in the result list is integrated, and answer information RI is generated by referring to the integrated answer information.

[0142] Fig. 10 is a diagram showing display data (display screen DIS3) including the answer information RI generated by the answer information generating unit 23 in this step. The display data is generated, for example, by the answer information display data generating unit 231. As shown in Fig. 10, the answer information RI includes "filler", "hardness", "weight ratio", and "literature information". Here, "filler", "hardness", and "weight ratio" included in the answer information RI are extracted by the answer information generating unit 23 from the output information OUT by the language model LM.

[0143] Moreover, the "literature information" in the answer information RI indicates information on the literature from which "filler", "hardness", and "weight ratio" are extracted. The answer information generating unit 23 can specify the "literature information" by specifying the literature ID associated with the chunk ID included in the output information OUT.

[0144] In this way, when the information processing device 100A is applied to the field of materials informatics, the response information RI may include literature information and at least any of physical property names, substance names, composition names, physical property values, units, experimental conditions, and experimental equipment.

[0145] Moreover, the answer information RI generated by the answer information generation unit 23 may include an area (COST2 in FIG. 10) that displays the actual cost used. As shown in FIG. 10, the display screen DIS3 includes a "download" button B3, and when the user presses this button, the answer information generation unit 23 downloads the document specified in the answer information RI.

[0146] The answer information generating unit 23 may be configured to score each piece of information included in the answer information RI in response to pressing of a "scoring" button shown in Fig. 10. As an example, the answer information generating unit 23 may perform such scoring processing as follows: Regarding at least a part of one or more items (in this example, answer items and additional items) included in the query QR, Scoring is performed on at least a part of one or more elements included in the output information OUT acquired by the second acquisition unit 12 or the answer information RI generated by the answer information generation unit 23. Here, the above "one or more elements" refers to, for example, an element corresponding to the above "one or more items" or literature information included in the answer information.

[0147] Fig. 11 shows a display screen DIS4 after scoring, which is generated by the answer information generating unit 23 in response to pressing of the "scoring" button shown in Fig. 10. The answer information generating unit 23 may be configured to execute the scoring process without pressing the "scoring" button, and generate the display screen DIS4 shown in Fig. 11. Also, each element shown on the display screen DIS4 may be displayed in place of the output information OUT shown in Fig. 10.

[0148] In the example shown in Figure 11, the "fillers" are scored using the number of times they appear in the literature, and the "fillers" are displayed in descending order of score. By performing such scoring, the user can easily identify information that is important or reliable.

[0149] As described above, the information processing device 100A according to the present embodiment generates extraction conditions, extracts a target chunk from a plurality of chunks using the extraction conditions, and inputs input information including at least a part of the target chunk into a language model. Therefore, the information processing device 100A can generate highly accurate data while suppressing an increase in cost.

[0150] (Additional Note 1 of the Second Embodiment) As described above, when there are multiple target chunks TC, the information processing device 100A may perform the above-mentioned steps S102 to S106 multiple times while changing the target chunks to be included in the prompt PR. Here, as an example, the information processing device 100A may perform the above-mentioned steps S102 to S106 multiple times while adjusting the length of the target chunks to be included in the prompt PR so as to satisfy the specifications of the language model LM. As an example, when there are eight target chunks TC01 to TC08 as the target chunks TC, the control unit 110A of the information processing device 100A may perform the above-mentioned steps S102 to S106 multiple times. A prompt PR01 is generated that includes the four target chunks TC01 to TC04 and the instruction sentence IN described above, and output information OUT01 output by the language model based on the prompt PR is obtained. A prompt PR02 is generated that includes the four target chunks TC05 to TC08 and the instruction sentence IN described above, and output information OUT02 output by the language model based on the prompt PR02 is obtained. The above configuration may also be used.

[0151] In the case of the above configuration, in step S108, the answer information generating unit 23 may first integrate the output information OUT01 to OUT02, and generate the answer information RI by referring to the integrated output information. At least a part of the integration process may be performed using the language model LM.

[0152] In this way, the first generation unit 21 a first input (prompt 01) including a first part of the one or more target chunks TC and an instruction sentence IN obtained from the query QR; a second input (prompt 02) including a second part of the one or more target chunks TC and an instruction IN obtained from the query QR; Generate The second acquisition unit 12 a first output information (OUT01) outputted from the language model LM to which the first input information (prompt 01) is input; and and acquiring second output information (OUT02) outputted from the language model LM to which the second input information (prompt 02) is input; The answer information generating unit 23 Generate the answer information RI by integrating at least a part of the first output information (OUT01) and at least a part of the second output information (OUT02). The above configuration may also be used.

[0153] According to the above configuration, even when there are many target chunks, it is possible to obtain the output of the language model LM related to the entirety of the many target chunks by repeatedly inputting a prompt including only some of the target chunks. Therefore, it is possible to preferably input the prompt PR and obtain the output information OUT without placing an excessive load on the language model LM or in a manner that conforms to the specifications of the language model LM.

[0154] In the second embodiment, an example in which one prompt PR contains multiple chunks has been given, but this does not limit the second embodiment and the related description. A configuration in which one prompt PR contains only one chunk is also included in the contents described in this specification. More specifically, The first generation unit 21 is configured so that the number of target chunks included in one prompt PR is only one, The prompt PR does not include the chunk ID described above, while the control unit 110A manages the target chunk in association with the document ID from which the target chunk was extracted, The answer information generating unit 23 includes the document information identified by the document ID in the answer information RI. The above configuration may also be used.

[0155] (Additional Note 2 to Embodiment 2) The process of generating the second condition COND2 by the second generating unit 22 is not limited to the above example. In the following, a process using the language model LM will be described as a specific example of the process of generating the second condition COND2 by the second generating unit 22.

[0156] FIG. 12 is a flow diagram showing an example of the flow of a process of generating the second condition COND2 by the second generating unit 22.

[0157] (Step S301) In step S301, the second generation unit 22 determines whether or not there is an unprocessed word class. Here, the word class refers to a category to be extracted or narrowed down, for example. If there is an unprocessed word class (YES in step S301), the process proceeds to step S302, and if not (NO in step S301), the process ends.

[0158] (Step S302) In step S302, the second generation unit 22 selects one of the unprocessed word classes and sets it as a word class to be processed. The upper part of FIG. 13 is a diagram showing a word class WC including "physical property name" and "unit" as an example of the above-mentioned word class, together with a query QR including a conditional statement and an answer item. The example shown in the upper part of FIG. 13 can be presented to the user via the input / output unit 140 as an example, but this does not limit the present embodiment. The second generation unit 22 may be configured to accumulate the frequency of selection of each word class by the user, and may be configured to exclude a word class with a relatively low selection frequency from the word classes to be presented to the user.

[0159] In this step, the second generating unit 22 sets an unprocessed word class out of the word classes "physical property name" and "unit" as a word class to be processed.

[0160] (Step S303) In step S303, the second generation unit 22 generates a prompt for extracting words belonging to the word class to be processed from the conditional statements and answer items included in the query QR. The lower left side of Fig. 13 shows prompt 1 (PR01), which is a prompt generated by the second generation unit 22 when "physical property name" is selected as the word class to be processed. The lower right side of Fig. 13 shows prompt 2 (PR02), which is a prompt generated by the second generation unit 22 when "unit" is selected as the word class to be processed.

[0161] 13 are merely examples and do not limit the present embodiment. The prompt generated by the second generation unit 22 may be in any format as long as it is a prompt for extracting words belonging to a word class to be processed from the conditional statements and answer items included in the query QR.

[0162] (Step S304) In step S304, the second generation unit 22 inputs the prompt generated in step S303 to the language model LM, and acquires output information output by the language model LM. The lower left side of Fig. 13 shows answer 1 (R01), which is output information output by the language model LM to which prompt 1 (PR01) is input. More specifically, answer 1 shows "methanol permeability", which is a phrase belonging to the phrase class "physical property name" and extracted from the conditional statement and answer items included in the query QR.

[0163] On the other hand, the lower right side of Fig. 13 shows answer 2 (R02), which is output information output by the language model LM to which prompt 2 (PR02) is input. More specifically, answer 2 includes "cm", which is a phrase belonging to the phrase class "unit" and extracted from the conditional sentence and answer item included in the query QR. 2 / s" is shown.

[0164] (Step S305) In step S305, the second generation unit 22 adds the words (answer) included in the output information of the language model LM to the narrowing down condition (second condition) by "AND". The bottom part of FIG. 13 shows the narrowing down condition (second condition) COND2 generated in this step. As shown in the bottom part of FIG. 13, the narrowing down condition (second condition) COND2 is formed by adding answer 1 "methanol permeability" from the language model LM to prompt 1 and answer 2 "cm 2 / s" is included using "AND".

[0165] In this step, the second generating unit 22 may be configured to further include synonyms of each word included in the conditional statement included in the query QR in the narrowing condition (second condition) COND2 with "OR".

[0166] The second generating unit 22 can suitably generate the second condition COND2 by performing the above-mentioned process. Note that a specific example of the process of generating the second condition COND2 by the second generating unit 22 is not limited to the above example, and for example, the process shown in FIG. 14 may be performed.

[0167] FIG. 14 is a flow diagram showing another example of the flow of the process of generating the second condition COND2 by the second generating unit 22.

[0168] (Step S401) Next, in step S401, the second generation unit 22 generates a prompt for generating a refinement condition (second condition) COND2 from the conditional statement and answer item included in the query QR. In other words, the second generation unit 22 An instruction to the language model LM to generate a refinement condition (second condition) COND2 from the conditional statements and answer items included in the query QR. Generates a prompt containing

[0169] (Step S304) Step S304 is the same as the process described with reference to FIG. 12, so the description will be omitted.

[0170] (Step S402) In step S402, the second generation unit 22 refers to the output information output by the language model LM to which the prompt generated in step S401 is input, and sets the output information as a refinement condition (second condition) COND2.

[0171] The second generating unit 22 can also suitably generate the second condition COND2 by performing the above-mentioned processing.

[0172] (Additional Note 3 of Embodiment 2) The information processing device 100A according to embodiment 1A can be applied to data generation in any field. In the above description, the information processing device 100A is mainly applied to data generation in the field of materials informatics, but the present invention is not limited to this.

[0173] As an example, the first acquisition unit 11 receives the following as the query QR: Condition COND1: Fire in a residence inhabited only by people aged 70 or older -Answer item ITEM1 "Date of occurrence" -Additional item ITEM2 "City of occurrence" The first generating unit 21 may generate a command sentence IN by referring to the query QR, and generate a prompt PR including the generated command sentence IN and one or more target chunks TC. Here, as one example, the target chunk TC may be a chunk selected from a chunk group including news. The second acquiring unit 12 may acquire output information OUT output by the language model LM to which the prompt PR is input, and the answer information generating unit 23 may generate answer information RI by referring to the output information OUT.

[0174] On the other hand, in the field of material informatics and related fields, there are many papers and technical reports that can be referred to when generating data to be provided to users. Therefore, the amount of data of chunks that can be candidates for input to the language model LM also tends to increase. According to the above-mentioned information processing device 100A, a condition (second condition COND2) used to extract one or more target chunks from a plurality of chunks is generated, one or more target chunks TC are acquired using the condition (second condition COND2), and input information (prompt PR) including at least a part of the one or more target chunks TC and an instruction sentence IN obtained from the query QR is generated. Therefore, while suitably suppressing the amount of data of the chunks input to the language model LM, it is possible to use target chunks that are highly useful in data generation by the language model LM (generation of output information OUT). Therefore, according to the information processing device 100A, in the field of material informatics and related fields, it is possible to generate highly accurate data while suppressing an increase in cost.

[0175] (Additional Notes for Embodiments 1 and 2) In each of the above embodiments, the length of the target chunk TC to be included in the prompt PR is not particularly limited. On the other hand, the accuracy of the output information OUT output by the language model LM and the cost for acquiring the output information OUT may vary depending on the length, number, type, etc. of the target chunk TC to be included in the prompt PR. For this reason, the information processing device 100, 100A may be configured to allow the length, number, type, etc. of the target chunk TC to be set in advance.

[0176] As an example, the information processing device 100, 100A may be configured to be able to receive, via the input / output unit 140, information specifying at least one of the length, number, type, etc. of the target chunk TC from the user.

[0177] For example, the information processing device 100, 100A may perform the following via the input / output unit 140: - The minimum and maximum number of characters for the target chunk TC to be included in the prompt PR - The minimum and maximum number of target chunk TCs to be included in the prompt PR -The type of chunk TC to be included in the prompt PR (e.g., phrase, sentence, paragraph, chapter, etc.) and the first generating unit 21 may generate a prompt PR so as to satisfy the received information.

[0178] Alternatively, the information processing device 100 Each time a prompt PR is generated, the length, number, type, etc. of the target chunk TC included in the prompt PR are stored, - Manage this information in relation to the accuracy and cost of the output information OUT based on the prompt PR, Determine (update) the length, number, and type of the target chunk TC that is preferable for the accuracy and cost to approach the values ​​desired by the user, and reflect them in the generation process of the next prompt PR. The above configuration may also be used.

[0179] [Embodiment 3] Other embodiments of the present invention will be described below. For convenience of explanation, the same reference numerals are used for components having the same functions as those described in the above embodiment, and the explanations will not be repeated. In the following explanation, as an example, the registration unit 35 is described as an additional configuration to the configuration described in the second embodiment, but this does not limit the present embodiment, and the configuration in which the registration unit 35 is added to the information processing device 100 described in the first embodiment is also included in the present embodiment.

[0180] (Information Processing System 1B) Fig. 15 is a diagram showing the configuration of an information processing system 1B according to this embodiment. As shown in Fig. 15, the information processing system 1B according to this embodiment includes a plurality of document servers (a first document server 300, a second document server 400, ... in Fig. 15) in addition to the components of the information processing system 1A according to the second embodiment.

[0181] Here, the first document server 300 is, for example, a paper distribution server that distributes free or paid papers. The second document server is, for example, a server managed by a target company, and is a server that manages technical reports and internal documents of the target company. However, these examples do not limit the present embodiment.

[0182] (Information processing device 100B) 15, the information processing device 100B included in the information processing system 1B according to this embodiment has the same configuration as the information processing device 100A according to the second embodiment, except for the following points. That is, the information processing device 100B according to this embodiment has a registration unit 35 in addition to each configuration included in the information processing device 100A according to the second embodiment.

[0183] (Registration Section 35) The registration unit 35 acquires a target document from the first document server 300 and the second document server 400, acquires a plurality of chunks by breaking down the acquired document into chunks, and stores the acquired plurality of chunks in the storage unit 120. These plurality of chunks constitute a chunk group CG in FIG.

[0184] Furthermore, the registration unit 35 generates a vector (feature vector in feature space) for each of one or more chunks included in the chunk group CG, and stores the generated vector in the storage unit 120. These vectors constitute the vector group VG in FIG. 15. The registration unit 35 can convert the chunk into a feature vector (embedding the chunk into feature space) by applying a predetermined algorithm to each of one or more chunks. Furthermore, at least a part of the above-mentioned processing by the registration unit 35 may be performed by a device separate from the information processing device 100B, such as the server 200, for example.

[0185] According to the information processing device 100B configured as above, it is possible to preferably prepare a plurality of chunks that are the source of the target chunk TC referenced in data generation (generation of output information OUT) by the language model LM. Moreover, according to the above configuration, documents are acquired from a plurality of servers and chunks are extracted, so that data generation (generation of output information OUT) can be performed using papers in various fields, documents from various companies, and the like. Moreover, according to the information processing device 100B, the same effects as those of the information processing devices 100 and 100A according to the first and second embodiments can be achieved.

[0186] [Embodiment 4] Other embodiments of the present invention will be described below. For ease of explanation, the same reference numerals are given to components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0187] (Information processing device 100C) Fig. 16 is a block diagram showing the configuration of an information processing device 100C according to this embodiment. As shown in Fig. 16, the information processing device 100C includes a control unit 110C and a storage unit 120C. The query QR, condition information CI, and target chunk TC stored in the storage unit 120C are the same as those described in the first to third embodiments, and therefore will not be described repeatedly here.

[0188] (Control unit 110C) The control unit 110C includes a query acquisition unit 41, a condition generation unit 42, and a chunk acquisition unit 43. Here, the query acquisition unit 41 is configured to perform the same process as the first acquisition unit 11 in the first to third embodiments, and is configured to acquire a query.

[0189] Moreover, the condition generating unit 42 is configured to perform the same processing as the second generating unit 22 in the second to third embodiments, and generates a condition (the second condition COND2 in the second to third embodiments) used to extract one or more target chunks from a plurality of chunks. As an example, the condition generating unit 42 generates the condition by referring to the query. Moreover, the chunk acquiring unit 43 is configured similarly to the third acquiring unit 13 in the second to third embodiments, and acquires one or more target chunks using the condition.

[0190] According to the information processing device 100C configured as above, conditions used for extracting one or more target chunks from a plurality of chunks are generated, and the one or more target chunks are acquired using the conditions, so that the target chunks to be searched or processed can be appropriately narrowed down. Therefore, by using the target chunks TC narrowed down by the information processing device 100C, it is possible to perform processing for generating highly accurate data while suppressing an increase in costs.

[0191] [Software implementation example] Each functional block of the information processing device 100, 100A, 100B, 100C (particularly, the first acquisition unit 11, the second acquisition unit 12, the third acquisition unit 13, the first generation unit 21, the second generation unit 22, the answer information generation unit 23, the registration unit 35, the query acquisition unit 41, the condition generation unit 42, and the chunk acquisition unit 43) and each functional block of the server 200 (particularly, the control unit 240) may be realized by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or may be realized by software. In the latter case, each of the information processing device 100, 100A, 100B, 100C and the server 200 can be configured, for example, using a computer (electronic calculator). FIG. 17 is a block diagram illustrating a physical configuration of a computer used as the information processing device 100 (or the information processing device 100A, 100B, 100C) and the server 200.

[0192] (Physical configuration of information processing device 100) 17, the information processing device 100 (or the information processing devices 100A, 100B, 100C) can be configured by a computer including a bus 110, a processor 101, a main memory 102, an auxiliary memory 103, a communication interface 104, and an input / output interface 105. The processor 101, the main memory 102, the auxiliary memory 103, the communication interface 104, and the input / output interface 105 are connected to each other via the bus 110. The input device 40 and the output device 50 are connected to the input / output interface 105.

[0193] For example, a microprocessor, a digital signal processor, a microcontroller, or a combination of these, etc. is used as the processor 101. For example, a semiconductor RAM (random access memory) etc. is used as the main memory 102.

[0194] The auxiliary memory 103 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these. The auxiliary memory 103 stores a program for causing the processor 101 to execute the operation of the information processing device 100 (or the information processing devices 100A, 100B, 100C) described above. The processor 101 loads the program stored in the auxiliary memory 103 onto the main memory 102, and executes each instruction included in the loaded program. The auxiliary memory 103 also stores various data referenced by the processor 101 to operate the computer as the information processing device 100 (or the information processing devices 100A, 100B, 100C).

[0195] The communication interface 104 is an interface that connects to the network 91. As the input / output interface 105, for example, a Universal Serial Bus (USB) interface, a short-range communication interface such as infrared or Bluetooth (registered trademark), or a combination of these is used.

[0196] For example, a keyboard, a mouse, a touch pad, a microphone, or a combination of these, etc. are used as the input device 40. For example, a display, a printer, a speaker, or a combination of these, etc. are used as the output device 50.

[0197] (Physical configuration of server 200) 17, the server 200 can be configured by a computer including a bus 210, a processor 201, a main memory 202, an auxiliary memory 203, a communication interface 204, and a communication interface 205. The processor 201, the main memory 202, the auxiliary memory 203, the communication interface 204, and the communication interface 205 are connected to each other via the bus 210.

[0198] For example, a microprocessor, a digital signal processor, a microcontroller, or a combination of these, etc. is used as the processor 201. For example, a semiconductor RAM, etc. is used as the main memory 202.

[0199] As the auxiliary memory 203, for example, a flash memory, a HDD, an SSD, or a combination of these is used. A program for operating the computer as the server 200 is stored in the auxiliary memory 203. The processor 201 loads the program stored in the auxiliary memory 203 onto the main memory 202, and executes each instruction included in the loaded program. In addition, the auxiliary memory 203 stores various data referenced by the processor 201 for operating the computer as the server 200.

[0200] The communication interface 204 is an interface that connects to the network 91. The communication interface 205 is an interface that connects to the network 92.

[0201] Each of the above-mentioned programs may be recorded in an external recording medium instead of being stored in the auxiliary memory 103 or the auxiliary memory 203, and may be supplied to the corresponding computer by being read from the external recording medium. The external recording medium may be a computer-readable "non-transitory tangible medium," such as a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. Each of the above-mentioned programs may be supplied to the computer via any transmission medium (such as a communication network or a broadcast wave) that can transmit the programs. One aspect of the present invention may also be realized in the form of a data signal embedded in a carrier wave, in which each program is embodied by electronic transmission.

[0202] 〔summary〕 The subject matter described in this specification includes at least the following configurations.

[0203] (Configuration 1) a first acquisition unit for acquiring a query; a first generation unit that generates input information including one or more target chunks and a directive corresponding to a first condition included in the query; a second acquisition unit that acquires output information output by a language model to which the input information is input; An information processing device comprising:

[0204] According to the above configuration, highly accurate data can be generated while suppressing increases in costs.

[0205] (Configuration 2) a second generating unit that generates a second condition used for extracting one or more of the target chunks from a plurality of chunks; a third acquisition unit that acquires one or more of the target chunks using the second condition; 2. The information processing device according to configuration 1, further comprising:

[0206] According to the above configuration, the one or more target chunks are obtained using the second condition, so that highly accurate data can be generated while suppressing increases in costs.

[0207] (Configuration 3) The first generation unit is a condition generating unit that generates the second condition by referring to the query; a condition modification unit that modifies the second condition in response to an instruction from a user; Equipped with 3. The information processing device according to configuration 2.

[0208] According to the above configuration, the second condition can be suitably generated by referring to the query. Also, the second condition can be modified in response to a modification instruction from a user, thereby making the condition more favorable to the user.

[0209] (Configuration 4) The first generation unit is a first display data generating unit that generates display data including at least a part of the second condition; It also has 4. The information processing device according to configuration 2 or 3.

[0210] According to the above configuration, by presenting the display data, the user can easily understand the content of the second condition.

[0211] (Configuration 5) The second acquisition unit is Among the plurality of chunks, one or more chunks that match the second condition are acquired as the one or more target chunks. 5. The information processing device according to any one of configurations 2 to 4.

[0212] According to the above configuration, since a chunk that matches the second condition is acquired as a target chunk, it is possible to preferably narrow down the chunks, and therefore it is possible to preferably reduce the cost of processing that refers to the chunks.

[0213] (Configuration 6) The second acquisition unit is acquiring one or more chunks that match the second condition from among the plurality of chunks; Obtaining the one or more target chunks by applying an expansion process to the one or more chunks that meet the second condition. 5. The information processing device according to any one of configurations 2 to 4.

[0214] According to the above configuration, the second acquisition unit can prepare a suitable amount of target chunks by applying an enlargement process, thereby improving the accuracy of the data generated by referring to the chunks.

[0215] (Configuration 7) The second generation unit is an instruction generating unit for generating the instruction from the query; 7. The information processing device according to any one of configurations 2 to 6.

[0216] According to the above configuration, it is possible to preferably generate an instruction sentence to be included in the input information so that the instruction sentence corresponds to the query.

[0217] (Configuration 8) The second generation unit is A second display data generating unit that generates display data including at least a part of the input information. 8. The information processing device according to any one of configurations 2 to 7.

[0218] According to the above configuration, by presenting the display data, the user can easily understand the content of the input information.

[0219] (Configuration 9) an answer information generating unit that generates answer information by referring to the output information; It also has 9. The information processing device according to any one of configurations 2 to 8.

[0220] According to the above configuration, it is possible to suitably generate answer information that is an answer to a user in response to the query, so that the answer information corresponds to the output information.

[0221] (Configuration 10) The answer information generation unit, a response information display data generating unit for generating display data including the response information; 10. The information processing device according to configuration 9.

[0222] According to the above configuration, by presenting the display data, the user can easily understand the content of the answer information.

[0223] (Configuration 11) The query may include: At least one of the following is included: physical property name, substance name, composition name, physical property value, unit, experimental condition, and experimental apparatus; The response information includes: The document contains at least one of the following: literature information, property name, substance name, composition name, property value, unit, experimental condition, and experimental apparatus. 11. The information processing device according to configuration 9 or 10.

[0224] According to the above configuration, in materials informatics and related fields, highly accurate data can be generated while suppressing increases in costs.

[0225] (Configuration 12) The second generation unit is a first input including a first portion of the one or more target chunks and a directive resulting from the query; second input information including a second portion of the one or more target chunks and a directive resulting from the query; Generate The third acquisition unit is first output information output by the language model to which the first input information is input; and second output information output by the language model to which the second input information is input; and Get The answer information generation unit, The answer information is generated by integrating at least a part of the first output information and at least a part of the second output information. 12. An information processing device according to any one of configurations 9 to 11.

[0226] According to the above configuration, it is possible to suitably input input information and obtain output information without placing an excessive load on the language model LM or in a manner that conforms to the specifications of the language model LM.

[0227] (Configuration 13) The answer information generation unit, With respect to at least a portion of one or more items included in the query, Scoring at least a part of one or more elements included in the output information acquired by the third acquisition unit. 13. An information processing device according to any one of configurations 9 to 12.

[0228] According to the above configuration, the user can easily identify information that is important or highly reliable.

[0229] (Configuration 14) a condition generating unit that generates a condition used to extract one or more target chunks from a plurality of chunks; a chunk acquisition unit that acquires one or more of the target chunks using the condition; An information processing device comprising:

[0230] The above configuration contributes to generating highly accurate data while suppressing increases in costs.

[0231] (Configuration 15) A query acquisition unit that acquires a query, The condition generating unit generates the condition by referring to the query. 15. An information processing device according to configuration 14.

[0232] The above configuration contributes to generating highly accurate data while suppressing increases in costs.

[0233] (Configuration 16) An information processing method executed by an information processing device, a first retrieval step of retrieving a query; a first generation step of generating input information including one or more target chunks and a directive corresponding to a first condition included in the query; A second acquisition step of acquiring output information output by a language model to which the input information is input; An information processing method comprising the steps of:

[0234] According to the above configuration, the same effects as those of the information processing device according to the first configuration are achieved.

[0235] (Configuration 17) An information processing method executed by an information processing device, a condition generating step of generating a condition used for extracting one or more target chunks from the plurality of chunks; a chunk acquisition step of acquiring one or more of the target chunks using the condition; An information processing method comprising the steps of:

[0236] (Configuration 18) A program for causing a computer to function as the information processing device according to configuration 1, the program causing a computer to function as the first acquisition unit, the first generation unit, and the second acquisition unit.

[0237] According to the above configuration, the same effects as those of the information processing device according to the first configuration are achieved.

[0238] (Configuration 19) A program for causing a computer to function as the information processing device according to configuration 14, the program causing a computer to function as the condition generating unit and the chunk acquiring unit.

[0239] According to the above configuration, the same effects as those of the information processing device according to the fourteenth configuration are achieved.

[0240] (Configuration 20) A computer-readable recording medium having the program according to configuration 18 recorded thereon.

[0241] According to the above configuration, the same effects as those of the information processing device according to the first configuration are achieved.

[0242] (Configuration 21) A computer-readable recording medium having the program according to configuration 19 recorded thereon.

[0243] According to the above configuration, the same effects as those of the information processing device according to the fourteenth configuration are achieved.

[0244] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0245] 100, 100A, 100B, 100C ··· Information processing device 110, 110A, 110B... Control section 11 First acquisition section 21 First generation part 211 ··· Condition generator 212 · Condition correction section 213 First display data generating unit 12 Second acquisition section 22 Second generation part 221 ··· Directive sentence generation unit 222 Second display data generating unit 13. Third Acquisition Section 23 ··· Answer information generation unit 231 Answer information display data generation unit 35 Registration Department 41 Query acquisition section 42 ··· Condition generator 43 Chunk acquisition section 200 Server

Claims

1. a first acquisition unit that acquires a query; a first generation unit that generates input information including one or more target chunks and a directive corresponding to a first condition included in the query; a second acquisition unit that acquires output information output by a language model to which the input information is input; An information processing device comprising:

2. a second generating unit configured to generate a second condition used for extracting one or more target chunks from a plurality of chunks; a third acquisition unit that acquires one or more of the target chunks using the second condition; The information processing apparatus according to claim 1 , further comprising:

3. The second generation unit is a condition generating unit that generates the second condition by referring to the query; The information processing apparatus according to claim 2 ,

4. a condition modification unit that modifies the second condition in response to an instruction from a user; Equipped with The information processing device according to claim 3 .

5. The second generation unit is a second display data generating unit for generating display data including at least a part of the second condition; It also has The information processing device according to claim 4.

6. The third acquisition unit is Among the plurality of chunks, one or more chunks that match the second condition are acquired as the one or more target chunks. The information processing device according to claim 5 .

7. The third acquisition unit is acquiring one or more chunks that match the second condition from among the plurality of chunks; Obtaining the one or more target chunks by applying an expansion process to the one or more chunks that match the second condition. The information processing device according to claim 5 .

8. The first generation unit is an instruction generating unit for generating the instruction from the query; The information processing device according to claim 6.

9. an answer information generating unit that generates answer information by referring to the output information; It also has The information processing device according to claim 1 .

10. The answer information generation unit, a response information display data generating unit for generating display data including the response information; The information processing device according to claim 9.

11. The query may include: At least one of the following is included: physical property name, substance name, composition name, physical property value, unit, experimental condition, and experimental apparatus; The response information includes: The document contains at least one of the following: literature information, property name, substance name, composition name, property value, unit, experimental condition, and experimental apparatus. The information processing device according to claim 10.

12. The first generation unit is a first input including a first portion of the one or more target chunks and a directive resulting from the query; second input information including a second portion of the one or more target chunks and a directive resulting from the query; Generate The second acquisition unit is first output information output by the language model to which the first input information is input; and second output information output by the language model to which the second input information is input; and Get The answer information generation unit, The answer information is generated by integrating at least a part of the first output information and at least a part of the second output information. The information processing device according to claim 9.

13. The answer information generation unit, With respect to at least a portion of one or more items included in the query, Scoring at least a part of one or more elements included in the output information acquired by the second acquisition unit. The information processing device according to claim 9.

14. a condition generating unit that generates a condition used for extracting one or more target chunks from a plurality of chunks; a chunk acquisition unit that acquires one or more of the target chunks using the condition; An information processing device comprising:

15. A query acquisition unit that acquires a query, The condition generating unit generates the condition by referring to the query. The information processing device according to claim 14.

16. An information processing method executed by an information processing device, A first obtaining step of obtaining a query; a first generation step of generating input information including one or more target chunks and a directive statement according to a first condition included in the query; A second acquisition step of acquiring output information output by a language model to which the input information is input; An information processing method comprising the steps of:

17. An information processing method executed by an information processing device, a condition generating step of generating a condition used for extracting one or more target chunks from the plurality of chunks; a chunk acquisition step of acquiring one or more of the target chunks using the condition; An information processing method comprising the steps of:

18. 2. A program for causing a computer to function as the information processing device according to claim 1, the program causing a computer to function as the first acquisition unit, the first generation unit, and the second acquisition unit.

19. 15. A program for causing a computer to function as the information processing device according to claim 14, the program causing a computer to function as the condition generating unit and the chunk acquiring unit.

20. A computer-readable recording medium having the program according to claim 18 recorded thereon.

21. A computer-readable recording medium having the program according to claim 19 recorded thereon.

Images