Information processing apparatus, information processing method, and program

Abstract

To provide a technique for a generation model to process data in a wide range of data formats.SOLUTION: One embodiment of the present disclosure relates to an information processing apparatus which has: an input acquisition unit which acquires question data and document data; a data conversion unit which converts the document data to structured data; a search query acquisition unit which inputs a first prompt generated on the basis of the question data and the structured data to a generative model and acquires a search query from the generative model; and an answer acquisition unit which inputs a second prompt generated on the basis of the question data and a search result of the search query in the structured data, to the generative model and acquires response data from the generative model.SELECTED DRAWING: Figure 4

Description

Technical Field

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

Background Art

[0002] With the emergence of the GPT-based machine learning model known as ChatGPT, the movement to utilize generative models in business has been growing. For example, in ChatGPT, when a question sentence of text data is input, a response sentence that is the answer to the input question can be obtained.

[0003] On the other hand, the document data used in business is not limited to text data, and not only document data such as PDF (Portable Document File) files and Microsoft Word files, but also document data input into table-form data such as Microsoft Excel data is widely used. For example, document data including images, text, and tables on one page may be created. There may also be cases where such non-text-form document data is used as processing target data and input into ChatGPT to obtain a response sentence.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] An object of the present disclosure is to provide a technique for a generative model to process data in a wide range of data formats.

Means for Solving the Problems

[0006] One aspect of the present disclosure relates to an information processing apparatus, which includes an input acquisition unit that acquires question data and document data, a data conversion unit that converts the document data into structured data, a search query acquisition unit that inputs a first prompt generated based on the question data and the structured data into a generation model and acquires a search query from the generation model, and a response acquisition unit that inputs a second prompt generated based on the question data and a search result for the structured data by the search query into the generation model and acquires response data from the generation model.

Effect of the Invention

[0007] According to the present disclosure, it is possible to provide a technique for a generation model to process data in a wide range of data formats.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0010] In the following embodiments, an information processing apparatus that processes document data using a generative model such as a GPT model that realizes ChatGPT is disclosed.

[0011] [SUMMARY OF THE DISCLOSURE] In the data processing shown in FIG. 1, the response text is shown when data processing is performed using a generation model for the PDF-formatted data of the insurance manual and the text-formatted question “In unemployment insurance, which types of insurance periods can be selected?” Here, the insurance manual is PDF-formatted document data with the layout as shown, and after being converted into text-formatted table data as shown, the converted text-formatted table data and the question text are input into the generation model.

[0012] In the converted table data of the insurance manual, the correspondence between the items in the header part (for example, “types of insurance periods”, “lifetime type”, “term type”, “age maturity”, “year maturity”, etc.) and the items in the data part (for example, “○”, “-”, etc.) is disrupted, and the generation model can generate a response text such as “In unemployment insurance, the types of insurance periods of ‘term type’ and ‘age maturity’ can be selected.” This is different from the originally expected response text “In unemployment insurance, the types of insurance periods of term type (age maturity) and term type (year maturity) can be selected.”, and it is one of the cases where the generation model cannot generate an appropriate response text.

[0013] Also, in the data processing shown in FIG. 2, the response text is shown when data processing is performed using a generation model for the PDF-formatted document data of the resume and the text-formatted questions “Do you have experience in sales business in life insurance?” and “Do you have experience in sales business in a property insurance company?” Here, the resume is PDF-formatted document data with a paragraph layout as shown, and after being converted into text data as shown, the converted text data and the question text are input into the generation model.

[0014] The converted text data of the resume fails to appropriately recognize the relationship between the description about Company A, a life insurance company, and the description about Company B, a property insurance company, resulting in the confusion of the work histories of Company A and Company B. The generation model may generate response texts such as "Yes, the data frame records the experience in a life insurance company. Specifically, I have been working at 'Dai Nippon Insurance Co., Ltd.' from May 2018 until now." and "Yes, the data frame records the experience in a property insurance company. Specifically, I have been working at 'Printing Insurance Company' from April 2016 to April 2018." This is different from the originally expected response text "Yes, I have experience in sales operations in the life insurance industry." and is one example of a case where the generation model fails to generate an appropriate response text.

[0015] Also, in the data processing shown in Figure 3, the response text when using the generation model to execute data processing for the Excel-formatted data of the inquiry log and the text-formatted question sentence "What is the number of inquiries regarding natural disasters?" is shown. Here, the inquiry log is document data in a table format as shown in the figure, and the table data of the inquiry log and the question sentence are input into the generation model.

[0016] The table data of the inquiry log detects keywords such as "fire" and "water leakage", which are not classified as natural disasters in Japan, as keywords regarding natural disasters, and generates a response text "The number of inquiries regarding natural disasters is 51." This is different from the originally expected response text "The total number of inquiries regarding natural disasters is 26." and is one example of a case where the generation model fails to generate an appropriate response text.

[0017] As described above with reference to Figures 1 to 3, according to the examples, the generation model fails to generate an appropriate response text for document data having a layout such as a table format, and a technology is required to enable the generation model to process document data with a layout well.

[0018] Schematically describing the embodiments of the present disclosure to be described later, when the information processing apparatus 100 acquires question data as text data and document data having a layout (for example, PDF file, Excel file, Word file, etc.) from the user 50, it converts the document data into structured data (for example, CSV (Comma Separated Value) format, JSON (JavaScript Object Notation) format, etc.). Then, the information processing apparatus 100 first generates a search query generation prompt for generating a search query for searching structured data corresponding to the question data.

[0019] The information processing apparatus 100 inputs the generated search query generation prompt into the generation model 200, and acquires a search query for searching structured data corresponding to the question data from the generation model 200. After that, the information processing apparatus 100 searches for structured data according to the acquired search query and obtains a search result. When obtaining the search result, the information processing apparatus 100 then generates a response data generation prompt for generating response data from the question data and the search result. Then, the information processing apparatus 100 inputs the response data generation prompt into the generation model 200 and acquires response data from the generation model 200.

[0020] Specifically, as shown in FIG. 4, in step S101, the information processing apparatus 100 first acquires text-form question data and non-text-form document data from the user 50. The document data can be, for example, document data having a layout such as a PDF file, an Excel file, or a Word file.

[0021] In step S102, the information processing apparatus 100 performs a structuring process on the document data and converts the document data into structured data (for example, CSV format, JSON format, etc.).

[0022] In step S103, the information processing apparatus 100 executes a prompt generation process for generating a search query generation prompt, and generates a search query generation prompt based on the question data and the structured data. Then, the information processing apparatus 100 inputs the generated search query generation prompt into the generation model 200.

[0023] In step S104, when the generation model 200 receives the search query generation prompt as an input, it generates a search query according to the received search query generation prompt. The search query can be, for example, program code for extracting search target data from the structured data.

[0024] In step S105, the information processing apparatus 100 searches the structured data according to the search query obtained from the generation model 200.

[0025] In step S106, the information processing apparatus 100 executes a prompt generation process for generating a response data generation prompt, and generates a response data generation prompt based on the question data and the search result. Then, the information processing apparatus 100 inputs the generated response data generation prompt into the generation model 200.

[0026] In step S107, when the generation model 200 receives the response data generation prompt as an input, it generates response data according to the received response data generation prompt.

[0027] In step S108, the information processing apparatus 100 obtains the response data from the generation model 200 and provides it to the user 50.

[0028] In step S109, the user 50 obtains the response data from the information processing apparatus 100.

[0029] Here, the information processing apparatus 100 may be realized by a computing device such as a server, a personal computer (PC), a smartphone, or a tablet, and may have, for example, a hardware configuration as shown in FIG. 5. That is, the information processing apparatus 100 includes a drive device 101, a storage device 102, a memory device 103, a processor 104, a user interface (UI) device 105, and a communication device 106 that are interconnected via a bus B.

[0030] For example, the user 50 may operate a client terminal, communicate with the information processing apparatus 100 realized as a server communicatively connected to the user terminal, and use the generation model 200. Alternatively, the user 50 may directly operate the information processing apparatus 100 realized as a user terminal such as a smartphone, a tablet, or a personal computer, and use the generation model 200. For example, the functions of the information processing apparatus 100 described later may be provided to the user 50 via an application installed in the information processing apparatus 100 or the like.

[0031] A program or instruction for realizing various functions and processes in the information processing apparatus 100 may be stored in a removable storage medium such as a CD-ROM (Compact Disk-Read Only Memory) or a flash memory. When the storage medium is set in the drive device 101, the program or instruction is installed from the storage medium to the storage device 102 or the memory device 103 via the drive device 101. However, the program or instruction does not necessarily have to be installed from the storage medium and may be downloaded from any external device via a network or the like.

[0032] The storage device 102 is realized by a hard disk drive or the like and stores files, data, etc. used for the execution of the installed program or instruction together with the installed program or instruction.

[0033] The memory device 103 is realized by a random access memory, a static memory, etc. When a program or an instruction is activated, it reads and stores a program, an instruction, data, etc. from the storage device 102. The storage device 102, the memory device 103, and the removable storage medium may be collectively referred to as a non-transitory storage medium.

[0034] The processor 104 may be realized by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuitry, etc. that can be composed of one or more processor cores. It executes various functions and processes of the information processing device 100 according to the programs, instructions, data such as parameters necessary to execute the programs or instructions stored in the memory device 103.

[0035] The user interface (UI) device 105 may be composed of input devices such as a keyboard, a mouse, a camera, a microphone, output devices such as a display, a speaker, a headset, a printer, and input / output devices such as a touch panel, and realizes an interface between the user 50 and the information processing device 100. For example, the user may operate the information processing device 100 by operating a GUI (Graphical User Interface) displayed on the display or the touch panel with a keyboard, a mouse, etc.

[0036] The communication device 106 is realized by various communication circuits that execute wired and / or wireless communication processing with communication networks such as external devices, the Internet, a LAN (Local Area Network), and a cellular network.

[0037] However, the above-described hardware configuration is merely an example, and the information processing device 100 according to the present disclosure may be realized by any other appropriate hardware configuration.

[0038] Information Processing Apparatus Next, the information processing apparatus 100 according to an embodiment of the present disclosure will be described. The information processing apparatus 100 according to the following embodiments uses, without limitation, for example, a GPT model that realizes ChatGPT as the generation model 200.

[0039] FIG. 6 is a block diagram showing the functional configuration of the information processing apparatus 100 according to an embodiment of the present disclosure. As shown in FIG. 6, the information processing apparatus 100 includes an input acquisition unit 110, a data conversion unit 120, a search query acquisition unit 130, and a response acquisition unit 140. Each functional unit of the input acquisition unit 110, the data conversion unit 120, the search query acquisition unit 130, and the response acquisition unit 140 may be realized by a computer program stored in the memory device 103 of the information processing apparatus 100 being executed by the processor 104.

[0040] The input acquisition unit 110 acquires question data and document data. Specifically, the input acquisition unit 110 acquires a text-based question sentence from the user 50 and document data in formats such as PDF, Excel, and Word. In this case, the document data may be composed of one or more contents such as text, tables, and images, and the question sentence may include inquiries, analyses, conversions, etc. for the contents such as text, tables, and images of the document data.

[0041] The data conversion unit 120 converts the document data into structured data. For example, the data conversion unit 120 converts document data having a layout into text-formatted data with the layout saved. For example, the data conversion unit 120 uses any known conversion tool to convert document data in formats such as PDF, Excel, and Word into text data in formats such as CVS and JSON.

[0042] The query acquisition unit 130 inputs a query generation prompt generated based on the question data and the structured data into the generation model 200, and acquires a query from the generation model 200. For example, the query acquisition unit 130 may generate a query generation prompt including a question sentence, structured data, and a processing instruction, and input the generated query generation prompt into the generation model 200. Then, the query acquisition unit 130 acquires a query from the generation model 200 for the input query generation prompt.

[0043] The response acquisition unit 140 inputs a response data generation prompt generated based on the question data and the search result for the structured data by the query into the generation model 200, and acquires response data from the generation model 200. For example, the response acquisition unit 140 may acquire a search result by a query for the structured data, generate a response data generation prompt including a question sentence, the search result, and a processing instruction, and input the generated response data generation prompt into the generation model 200. Then, the response acquisition unit 140 acquires response data from the generation model 200 for the input response data generation prompt.

[0044] Figs. 7 to 11 are diagrams showing data processing by the information processing apparatus 100 according to an embodiment of the present disclosure. In the illustrated embodiment, the user 50 inputs a question sentence "What types of insurance periods can be selected for unemployment insurance?" and document data of an insurance guide in PDF format into the information processing apparatus 100, and acquires a response sentence by the generation model 200.

[0045] In step S201, the user 50 inputs a question sentence "What types of insurance periods can be selected for unemployment insurance?" and document data of an insurance guide in PDF format into the information processing apparatus 100.

[0046] In step S202, the information processing apparatus 100 performs a structuring process on the document data of the insurance guide in PDF format and converts it into structured data in CSV format. Specifically, the structuring process may include any process for realizing structure recognition and format conversion on the document data.

[0047] Specifically, when the document data includes data blocks composed of data contents related to a common theme, the data conversion unit 120 may convert the document data into structured data for each data block. For example, in the document data example shown in FIG. 8, the information processing apparatus 100 first recognizes the layout of the document data and recognizes text parts, image parts, table parts, etc. Further, when the document data has regions 1 to 3 consisting of contents related to a common theme, and each of the regions 1 to 3 is composed of text, an image, and / or a table, the information processing apparatus 100 may use any known technique to determine which region (meaning block) the recognized text part, image part, and table part can be grouped into. In the illustrated document data example, for example, the text in region 1 may explain the image in region 1, and the text in regions 2 and 3 may explain the tables in regions 2 and 3. Then, the information processing apparatus 100 may use any conversion tool to convert the text part and the table part into data in CSV format or JSON format as shown.

[0048] In this way, the information processing apparatus 100 can acquire data in the CSV format as shown in FIG. 9 from the document data of the insurance guide in the PDF format. In the illustrated CSV data, four rows of "insurance type" and "life type", "term type (age maturity)", and "term type (year maturity)" regarding the type of insurance period are extracted, and in each row, "whole life insurance, ○, -, -", "term insurance, -, ○, ○", etc. are extracted. Here, for "whole life insurance, ○, -, -", it means that the "insurance type" is "whole life insurance", and the "whole life insurance" corresponds to the "life type" and does not correspond to the "term type (age maturity)" and the "term type (year maturity)". For "term insurance, -, ○, ○", it means that the "insurance type" is "term insurance", and the "term insurance" does not correspond to the "life type" and corresponds to the "term type (age maturity)" and the "term type (year maturity)".

[0049] In step S203, the information processing apparatus 100 generates a search query generation prompt from the question sentence and the structured data. For example, the information processing apparatus 100 may generate a search query generation prompt as shown in FIG. 11A based on the question sentence "In the employment disability insurance, which type of insurance period can be selected?" and the structured data in the CSV format. Here, the question sentence is included in the search query generation prompt as "#Question In the employment disability insurance, which type of insurance period can be selected?", the structured data is included in the search query generation prompt as "#Data content *Insurance type *Life type *Term type (age maturity) *Term type (year maturity)", and the processing instruction is included in the search query generation prompt as "You handle the data format of Python named df. Please understand the data content and generate a search code so that the answer to the question can be derived." The information processing apparatus 100 inputs the generated search query generation prompt into the generation model 200.

[0050] In step S204, the generation model 200 generates a search query based on the search query generation prompt and returns the generated search query to the information processing device 100. For example, the generation model 200 can generate the python code "df[df['Insurance Type'] == 'Disability Insurance']" to obtain the row where the "Insurance Type" is "Disability Insurance" for the search query generation prompt as shown in Fig. 11A.

[0051] In step S205, when the information processing device 100 obtains a search query from the generation model 200, it searches for structured data according to the obtained search query and obtains the search result "Disability Insurance, -, ○, ○" as shown in Fig. 10. Then, based on the question sentence "For Disability Insurance, which types of insurance periods can be selected?" and the search result "Disability Insurance, -, ○, ○", the information processing device 100 may generate a response sentence generation prompt as shown in Fig. 11B.

[0052] In step S206, the generation model 200 generates a response sentence based on the response sentence generation prompt and returns the generated response sentence to the information processing device 100. For example, the generation model 200 can generate a response sentence such as "For Disability Insurance, the types of insurance periods of 'Fixed-term (age maturity)' and 'Fixed-term (year maturity)' can be selected." for the response sentence generation prompt as shown in Fig. 11B.

[0053] In step S207, the information processing device 100 provides the response sentence obtained from the generation model 200 to the user 50.

[0054] In step S208, the user 50 can check the response sentence provided by the information processing device 100.

[0055] Next, FIGS. 12 to 15 are diagrams showing data processing by the information processing apparatus 100 according to an embodiment of the present disclosure. In the illustrated embodiment, a user 50 inputs a question sentence "Do you have experience in sales operations in life insurance?" and document data of a resume in PDF format to the information processing apparatus 100, and desires generation of a response sentence by the generation model 200.

[0056] In step S301, the user 50 inputs a question sentence "Do you have experience in sales operations in life insurance?" and document data of a resume in PDF format to the information processing apparatus 100.

[0057] In step S302, the information processing apparatus 100 executes a structuring process on the document data of the resume in PDF format and converts it into structured data in JSON format. The structuring process here may include any process for realizing structure recognition and format conversion for the document data, as described above with reference to FIG. 8.

[0058] Specifically, the information processing apparatus 100 can obtain JSON format data as shown in FIG. 13 from the document data of the resume in PDF format. In the illustrated JSON format data, work experience related to "Dainichi Insurance Co., Ltd." is described in paragraph 1, and work experience related to "Printing Insurance Co., Ltd." is described in paragraph 2. Specifically, in paragraph 1, the company name "Dainichi Insurance Co., Ltd.", industry type "life insurance industry", work period "May 2018 - present" and job content "Regular visits to existing customers,..." are described, and in paragraph 2, the company name "Printing Insurance Co., Ltd.", industry type "non-life insurance industry", work period "April 2016 - April 2018" and job content "Customer support team (about 50 people)..." are described.

[0059] In step S303, the information processing apparatus 100 generates a search query generation prompt from the question text and the structured data. For example, the information processing apparatus 100 may generate a search query generation prompt as shown in FIG. 15A based on the question text “Do you have experience in sales business in life insurance?” and the structured data in JSON format. Here, the question text is included in the search query generation prompt as “#Question Do you have experience in sales business in life insurance?”, the structured data is included in the search query generation prompt as “#Data content *Company name *Business content *Employment period *Industry type”, and the processing instruction is included in the search query generation prompt as “You handle Python data formats named df. Please understand the data content and generate a search code so that an answer to the question can be derived.” The information processing apparatus 100 inputs the generated search query generation prompt into the generation model 200.

[0060] In step S304, the generation model 200 generates a search query based on the search query generation prompt and returns the generated search query to the information processing apparatus 100. For example, the generation model 200 may generate Python code “df[df[‘Industry type’]==‘Life insurance’]&(df[‘Business content’].str.contains(‘Sales’))]” for obtaining rows where the “Industry type” is “Life insurance” and the “Business content” contains “Sales” for the search query generation prompt as shown in FIG. 15A.

[0061] In step S305, when the information processing apparatus 100 obtains a search query from the generation model 200, it searches the structured data according to the obtained search query and obtains the search result of paragraph 1 as shown in FIG. 14. Then, the information processing apparatus 100 may generate a response text generation prompt as shown in FIG. 15B based on the question text “Do you have experience in sales business in life insurance?” and the search result of paragraph 1.

[0062] In step S306, the generation model 200 generates a response sentence based on the prompt for response sentence generation, and returns the generated response sentence to the information processing apparatus 100. For example, the generation model 200 can generate a response sentence such as "Yes, I have experience in sales operations in the life insurance industry" for the prompt for response sentence generation as shown in FIG. 15B.

[0063] In step S307, the information processing apparatus 100 provides the response sentence obtained from the generation model 200 to the user 50.

[0064] In step S308, the user 50 can check the response sentence provided by the information processing apparatus 100.

[0065] Next, FIGS. 16 to 20 are diagrams showing data processing by the information processing apparatus 100 according to an embodiment of the present disclosure. In the illustrated embodiment, the user 50 inputs a question sentence "What is the number of inquiries regarding natural disasters?" and document data of the inquiry log for October in Excel format to the information processing apparatus 100, and desires to generate a response sentence by the generation model 200.

[0066] In step S401, the user 50 inputs a question sentence "What is the number of inquiries regarding natural disasters?" and document data of the inquiry log for October in Excel format to the information processing apparatus 100.

[0067] In step S402, the information processing apparatus 100 performs structuring processing on the document data of the inquiry log in Excel format and converts it into structured data in CSV format. The structuring processing here may include any processing for realizing structure recognition and format conversion on the document data, as described above with reference to FIG. 8.

[0068] For example, the information processing apparatus 100 can obtain structured data in CSV format as shown in FIG. 17 from the document data of the inquiry log for October in Excel format. For example, the log of each row of the inquiry log for October in Excel format is converted into the record of each row of the structured data in CSV format. Specifically, since the cause of the log in the first row is "A fire occurred in the bathroom at home, and some walls and floors were damaged. Can the repair cost be covered by insurance?", the factor of the record in the first row of the structured data in CSV format becomes "fire", and "○" is entered for human-caused fire. Also, since the cause of the log in the second row is "Due to strong winds, a tree in the garden fell and damaged my car. Can the repair cost be covered by insurance?", the factor of the record in the second row of the structured data in CSV format becomes "fall of a tree due to strong winds", and "○" is entered for strong winds.

[0069] In step S403, the information processing apparatus 100 generates a search query generation prompt from the question text and the structured data. For example, the information processing apparatus 100 may generate a search query generation prompt as shown in FIG. 19A based on the question text "What is the number of inquiries regarding natural disasters?" and the structured data in CSV format. Here, the question text is included in the search query generation prompt as "#Question What is the number of inquiries regarding natural disasters?", the structured data is included in the search query generation prompt as "#Data content *ID *Strong winds...", and the processing instruction is included in the search query generation prompt as "You are named df and handle Python data formats. Understand the data content and generate a search code so that the answer to the question can be derived." The information processing apparatus 100 inputs the generated search query generation prompt into the generation model 200.

[0070] That is, the information processing apparatus 100 extracts keywords from the structured data based on a dictionary regarding the search target keywords in the question data. Specifically, for the search target keyword "natural disaster", a group of keywords related to "natural disaster" (in this example, "strong wind", "snow / hail", "lightning strike", "flood / heavy rain", and "earthquake") is defined in advance, and this group of keywords is registered as a dictionary of keywords corresponding to the search target keyword "natural disaster". Thereby, the group of keywords corresponding to the search target keyword can be appropriately extracted.

[0071] In step S404, the generation model 200 generates a search query based on the search query generation prompt, and returns the generated search query to the information processing apparatus 100. For example, the generation model 200 can generate a Python code of "natural_disasters=['strong wind','snow / hail', 'lightning strike', 'flood / heavy rain', 'earthquake'] df[natural_disasters].apply(lambda x:x=='○').sum()" for the search query generation prompt as shown in FIG. 19A. The search query defines natural disasters as'strong wind','snow / hail', 'lightning strike', 'flood / heavy rain', and 'earthquake', and counts the number of rows in which these items are "○" in the structured data.

[0072] In step S405, when the information processing apparatus 100 acquires a search query from the generation model 200, it searches the structured data according to the acquired search query, and as shown in FIG. 16, obtains a search result of "26 items". Then, the information processing apparatus 100 may generate a response text generation prompt as shown in FIG. 19B based on the question text "What is the number of inquiries regarding natural disasters?" and the search result of "26 items".

[0073] In step S406, the generation model 200 generates a response sentence based on the response sentence generation prompt and returns the generated response sentence to the information processing apparatus 100. For example, the generation model 200 can generate a response sentence such as "The total number of inquiries regarding natural disasters is 26." as shown in FIG. 18 for the response sentence generation prompt as shown in FIG. 19B.

[0074] In step S407, the information processing apparatus 100 provides the response sentence obtained from the generation model 200 to the user 50.

[0075] In step S408, the user 50 can check the response sentence provided by the information processing apparatus 100.

[0076] Note that the above-described structured data in CSV format may be as shown in FIG. 20. In this case, for example, for the question sentence "What is the number of inquiries that may be related to typhoons?", the generation model 200 can similarly generate a search query generation prompt of "typhoon=[‘strong wind’,‘lightning’,‘flood·heavy rain’] df[typhoon].apply(lambda x:x==‘○’).sum()", and further generate a response sentence "The total number of cases that may be related to typhoons is 18." based on the search result of "18 cases" and the question sentence.

[0077] As described above in detail with respect to the embodiments of the present disclosure, the present disclosure is not limited to the specific embodiments described above, and various modifications and changes are possible within the scope of the gist of the present disclosure described in the claims.

Description of Reference Numerals

[0078] 50 User 100 Information Processing Apparatus 110 Input Acquisition Unit 120 Data Conversion Unit 130 Search Query Acquisition Unit 140 Response Acquisition Unit

Claims

1. An input acquisition unit that acquires question data and document data, A data conversion unit that converts the document data into structured data, A search query acquisition unit that inputs a first prompt generated based on the question data and the structured data into a generation model and acquires a search query from the generation model, A response acquisition unit that inputs a second prompt generated based on the question data and the search result for the structured data by the search query into the generation model and acquires response data from the generation model, An information processing apparatus comprising:

2. The information processing apparatus according to claim 1, wherein the structured data is data in CSV (Comma Separated Value) format or JSON (JavaScript Object Notation) format.

3. The information processing apparatus according to claim 1, wherein the search query is program code for the structured data.

4. The information processing apparatus according to claim 1, wherein the data conversion unit extracts keywords from the structured data based on a dictionary regarding search target keywords in the question data.

5. The document data includes data blocks composed of data contents regarding a common theme, The information processing apparatus according to claim 1, wherein the data conversion unit converts the document data into the structured data for each data block.

6. Acquiring question data and document data; Converting the document data into structured data; Inputting a first prompt generated based on the question data and the structured data into a generation model and acquiring a search query from the generation model; Inputting a second prompt generated based on the question data and the search result for the structured data by the search query into the generation model and acquiring response data from the generation model; An information processing method executed by a computer.

7. Acquiring question data and document data; Converting the document data into structured data; Inputting a first prompt generated based on the question data and the structured data into a generation model and acquiring a search query from the generation model; Input a second prompt generated based on the question data and the search results for the structured data by the search query into the generation model, and obtain response data from the generation model; A program for causing a computer to execute.

Images