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

Abstract

To support appropriate conversion into structured document data.SOLUTION: An information processing program of an embodiment causes a computer to execute division processing, creation processing, and execution processing. The division processing includes dividing document data of a conversion object into a plurality of document fragments. The creation processing includes, when requesting a large-scale language model to convert each of the divided document fragments into a document whose document structure is specified, creating a conversion request sentence that includes a document fragment immediately preceding the document fragment requested to be converted, the document fragment requested to be converted, and a converted document fragment obtained by the large-scale language model by converting the immediately preceding document fragment. The execution processing includes inputting the created conversion request sentence into the large-scale language model to execute conversion into the document fragments whose document structure is specified.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] An embodiment of the present invention relates to an information processing program, an information processing method, and an information processing device. [Background technology]

[0002] Document data intended for human reading, such as design documents, manuals, and procedure manuals, often has poor machine readability due to insufficient or poor structure in terms of document structure (chapters, sections, tables, annotations, references, sources, etc.) through tagging, etc.

[0003] For such insufficiently structured document data, there is a conventional technique for converting it into structured document data by loading it into a large language model (LLM) such as ChatGPT (registered trademark) and requesting structuring. Furthermore, since the document data to be converted, such as design documents, manuals, and procedure manuals, is lengthy and has many pages, there is a conventional technique for summarizing the document data to be converted so as to satisfy the token length constraints of the LLM before loading it into the LLM. There is also a conventional technique for converting and dividing the document data to be converted into document fragments and loading them into the LLM. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2023-73095 [Patent Document 2] Patent No. 7378003 specification [Patent Document 3] US Patent Application Publication No. 2004 / 0117734 [Patent Document 4] US Patent Application Publication No. 2022 / 0036153 [Patent Document 5] U.S. Patent No. 1,1748,577 [Patent Document 6] International Publication No. 2023 / 145051 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the above-mentioned conventional technologies have the problem that it is sometimes difficult to convert document data into appropriately structured data. For example, when document data to be converted is summarized and loaded into LLM, the summarization can result in the loss of information important for structuring, etc., which can impair the accuracy of the converted text data. Furthermore, when document data to be converted is converted and divided into document fragments, each document fragment may not contain the information necessary for structuring, and the structuring by LLM may become unstable, such as by tagging main headings instead of subheadings.

[0006] In one aspect, an object of the present invention is to provide an information processing program, an information processing method, and an information processing device that can support appropriate conversion into structured document data. [Means for solving the problem]

[0007] In one proposal, the information processing program causes a computer to execute a dividing process, a creating process, and an executing process. The dividing process divides document data to be converted into a plurality of document fragments. The creating process, when requesting a large-scale language model to convert each divided document fragment into a document whose document structure is explicitly stated, creates a conversion request statement including the document fragment immediately preceding the document fragment for which conversion is requested, the document fragment for which conversion is requested, and a document fragment resulting from conversion of the immediately preceding document fragment by the large-scale language model. The executing process inputs the created conversion request statement into the large-scale language model and executes conversion into a document fragment whose document structure is explicitly stated. [Effects of the Invention]

[0008] According to one embodiment, it is possible to support appropriate conversion into structured document data. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram illustrating an example of a functional configuration of an information processing apparatus according to an embodiment. [Figure 2] FIG. 2 is a flowchart illustrating an example of the operation of the information processing device according to the embodiment. [Figure 3] FIG. 3 is an explanatory diagram illustrating an example of a request text. [Figure 4] FIG. 4 is an explanatory diagram illustrating an example of a request text. [Figure 5] FIG. 5 is an explanatory diagram illustrating an example of the level correction process. [Figure 6] FIG. 6 is an explanatory diagram illustrating an example of a structuring result. [Figure 7A] FIG. 7A is an explanatory diagram illustrating an example of an output result. [Figure 7B] FIG. 7B is an explanatory diagram illustrating an example of the output result. [Figure 8] FIG. 8 is an explanatory diagram illustrating a comparative example. [Figure 9] FIG. 9 is an explanatory diagram illustrating a comparative example. [Figure 10] FIG. 10 is an explanatory diagram for explaining a comparative example. [Figure 11] FIG. 11 is an explanatory diagram illustrating a comparative example. [Figure 12] FIG. 12 is an explanatory diagram illustrating an example of a computer configuration. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an information processing program, an information processing method, and an information processing device according to embodiments will be described with reference to the drawings. Components having the same functions in the embodiments will be assigned the same reference numerals, and duplicated descriptions will be omitted. Note that the information processing program, the information processing method, and the information processing device described in the following embodiments are merely examples and do not limit the embodiments. Furthermore, the following embodiments may be combined as appropriate within a range that does not cause contradictions.

[0011] Fig. 1 is a block diagram illustrating an example of a functional configuration of an information processing device according to an embodiment, and Fig. 2 is a flowchart illustrating an example of an operation of the information processing device according to an embodiment.

[0012] As shown in Figures 1 and 2, an information processing device 1 receives input of document data 2 to be converted for structuring the document structure, requests a large-scale language model M1 such as ChatGPT to convert the received document data 2, and obtains the converted (structured) document data.

[0013] Document structure refers to the context and hierarchical structure within a document, such as chapters, sections, tables, annotations, references, and sources. Structuring a document structure refers to making the above document structure clear within the document by tagging or other methods.

[0014] For example, text data in which chapters and sections are visually expressed using indents, numbers, CSV (Comma Separated Values), etc., does not clearly indicate the document structure using tags or the like, and is therefore poorly machine-readable. In this embodiment, such text data corresponds to document data 2 to be converted. Note that the converted document data may be any data that can clearly indicate the document structure using tags or the like. In this embodiment, as an example, the document structure is converted into text data expressed using Markdown notation.

[0015] Furthermore, the document data 2 to be converted is assumed to be long and has many pages, etc., and does not satisfy the token length constraint of the large-scale language model M1 as is.

[0016] For this reason, the information processing device 1 divides the document data 2 into multiple document fragments so as to satisfy the token length constraints of the large-scale language model M1, then requests the large-scale language model M1 to convert each document fragment, and outputs a group of context-sensitive fragments 3, which are multiple structured document fragments.

[0017] The context-conscious fragment group 3 is a set of multiple document fragments (context-conscious fragments) that are created by combining multiple structured results 27 obtained by converting each document fragment using the large-scale language model M1, and then dividing them again at the position of the document structure, such as chapter or section, that is specified by tagging, etc.

[0018] Each document fragment before conversion is obtained by dividing the document data 2 in accordance with the token length constraints of the large-scale language model M1, not by dividing it according to its position in the document structure. Therefore, the multiple structuring results 27 obtained from each document fragment are divided in the middle of the document structure, such as a chapter or section, and the information processing device 1 combines the multiple structuring results 27 to obtain structured document data.

[0019] Furthermore, the document data obtained by combining (document data after structure) is long and has many pages, just like the document data 2 before conversion, and is difficult to handle as is. Therefore, the information processing device 1 divides the structured document data again at positions in the document structure, such as chapters and sections, that are specified by tagging, and outputs a context-considered fragment group 3. Note that, although the present embodiment illustrates a configuration in which the context-considered fragment group 3 is output, it goes without saying that a configuration in which the document data obtained by combining (document data after structure) is output may also be used.

[0020] The information processing device 1 has a pre-processing unit 10, a pre-segmentation unit 11, a local context information extraction unit 12, a global context information extraction unit 13, a structured fragment DB 14, a document structuring request construction unit 15, a large-scale language model execution unit 16, and a post-segmentation unit 17.

[0021] The preprocessing unit 10 is a processing unit that performs preprocessing (S1) such as text extraction on input document data 2. Through this preprocessing, the information processing device 1 obtains a text document 20 from the document data 2.

[0022] The pre-segmentation unit 11 is a processing unit that pre-segments (S2) the text-converted document 20 into segments of a predetermined number of characters, for example, in accordance with the token length constraint of the large-scale language model M1. In this embodiment, a conversion request is made to the large-scale language model M1 by combining two document fragments, and therefore, in the pre-segmentation (S2), the segmentation is performed so that the combination of the two document fragments satisfies the token length constraint.

[0023] By this pre-division, the pre-division unit 11 obtains a pre-division fragment group 21 including, for example, document fragments 1 to N. Note that the pre-division unit 11 assigns numbers (1 to N) corresponding to the order to each document fragment during pre-division (S2).

[0024] The information processing device 1 reads out each of the document fragments included in the pre-segmented fragment group 21 in numerical order, and requests structuring from the large-scale language model M1. Here, the document fragment to be requested to be structured among the document fragments included in the pre-segmented fragment group 21 is referred to as the target fragment 22, and its number is denoted as i. The document fragment immediately preceding this target fragment 22 is referred to as the immediately preceding fragment 23, and its number is denoted as i-1. Note that the immediately preceding fragment 23 is not limited to the single document fragment immediately preceding the target fragment 22, but may be a collection of multiple document fragments (i-1, i-2...iM) preceding the target fragment 22.

[0025] When a request is made to structure the target fragment 22 with number i, the structured result 27 of the immediately preceding fragment 23 with number i-1 has already been obtained by the large-scale language model M1 and stored in the structured fragment DB 14. The structured result 27 of this immediately preceding fragment 23 will be referred to as the immediately preceding fragment conversion result 24.

[0026] The local context information extraction unit 12 is a processing unit that extracts local context information that indicates the local document structure and context immediately adjacent to the target fragment 22 in the document data 2. Specifically, the local context information extraction unit 12 extracts the previous fragment 23 included in the pre-segmented fragment group 21 and the previous fragment conversion result 24 stored in the structured fragment DB 14 as local context information.

[0027] The global context information extraction unit 13 is a processing unit that extracts global context information 25 that indicates the global document structure and context, such as the chapter and section structure and table headers of the document data 2. Specifically, the global context information extraction unit 13 extracts, as global context information 25, the document structure, such as the chapter and section structure and table headers extracted using a rule base, or a document summary extracted using a known document summarization technique, based on the converted document fragments (structuration results 27) that have been stored in the structured fragment DB 14 (S4).

[0028] The structured fragment DB 14 is a database that stores the structuring results 27 obtained by the large-scale language model execution unit 16 performing conversion using the large-scale language model M1 for each document fragment included in the pre-divided fragment group 21.

[0029] The document structuring request construction unit 15 is a processing unit that creates a request statement 26 that requests the large-scale language model M1 to convert each document fragment included in the pre-divided fragment group 21 into a document that clearly indicates the document structure.

[0030] Specifically, the document structuring request construction unit 15 creates a request sentence 26 that includes the target fragment 22, the local context information extracted by the local context information extraction unit 12 (the immediately preceding fragment 23 and the immediately preceding fragment conversion result 24), and the global context information 25 extracted by the global context information extraction unit 13 (S3).

[0031] As an example, the document structuring request composition unit 15 uses, as the header of the request 26, a long document structuring system prompt 26a, a long design document structuring instruction 26b, and a document structure hint 26c related to structuring a long document such as document data 2. For the long document structuring system prompt 26a, the long design document structuring instruction 26b, and the document structure hint 26c, fixed phrases set in advance by the user or the like are used. Next, the document structuring request composition unit 15 creates the request 26 whose body includes global context information 25, the target fragment 22, the previous fragment 23, and the previous fragment conversion result 24.

[0032] 3 and 4 are explanatory diagrams illustrating an example of a request sentence. As shown in FIG. 3, the document structuring request composition unit 15 includes global context information (structure) 25a and global context information (summary) 25b in the body of the request sentence 26 in parts such as [Context]. Furthermore, the document structuring request composition unit 15 includes the previous fragment 23 plus the target fragment 22 in parts such as [Input] in the body of the request sentence 26. Furthermore, the document structuring request composition unit 15 includes the previous fragment conversion result 24 in the body of the request sentence 26 in parts such as [Input].<Leading Words> >Include etc.

[0033] where <<Leading Words> > is information for guiding the output from the large-scale language model M1 to a specific pattern.<Leading Words> By including the immediately preceding fragment conversion result 24 as >, it is possible to obtain an output (structuring result 27 of the target fragment 22) that inherits the contents of the immediately preceding fragment conversion result 24.

[0034] In addition, the document structuring request construction unit 15 may create a request statement 26 that explicitly states [Input] as if the previous fragment 23 to be converted is the previous fragment 23 combined with the target fragment 22, as shown in Figure 4, and [Output] as if the output up to the previous fragment 23 has already been completed.

[0035] Specifically, the document structuring request composition unit 15 writes the target fragment 22 after the previous fragment 23 in [Input]. Furthermore, the document structuring request composition unit 15 writes the previous fragment conversion result 24 after a symbol indicating the output from the large-scale language model M1 in [Output]. The symbol indicating the output from the large-scale language model M1 is, for example, a symbol such as "AI:" that is used when the AI (large-scale language model M1) gives an answer. This allows the information processing device 1 to guide the large-scale language model M1 so that, in a conversion request that combines the previous fragment 23 and the target fragment 22, the previous fragment 23 has already been converted.

[0036] The large-scale language model execution unit 16 is a processing unit that inputs the request sentence 26 created by the document structuring request composition unit 15 into the large-scale language model M1 and converts it into document fragments that clearly show the document structure. The large-scale language model execution unit 16 assigns numbers (1 to N) to the structuring results 27 obtained by the large-scale language model M1 performing conversion on the document fragments (1 to N) included in the pre-divided fragment group 21, and stores them in the structured fragment DB 14.

[0037] Here, the document structuring request composition unit 15 may include in the request statement 26 an instruction to specify the structuring level of the heading / list elements not only by tagging but also by names, etc. The large-scale language model execution unit 16 may correct the level specified by names, etc., of the structuring result 27 obtained by the large-scale language model M1 performing conversion, by level correction processing using character string replacement such as regular expressions.

[0038] Fig. 5 is an explanatory diagram illustrating an example of level correction processing. As shown in Fig. 5, the request 26 includes instructions to tag headings and list elements, as well as to specify the heading levels as H1, H2, and H3, and the list element levels as L1, L2, and L3.

[0039] In an example of an output image from the large-scale language model M1 based on such a request sentence 26, the first line is a level 3 (H3) heading, and although it is given a name indicating that it is H3, it may be given a level 1 tag.

[0040] In response to this, the large-scale language model execution unit 16 performs level correction processing (S6) using regular expressions or the like to replace the tag with a correct tag by referring to the name or the like, thereby enabling correct tagging.

[0041] 1 and 2, the post-division unit 17 is a processing unit that combines the structuring results 27 contained in the structured fragment DB 14 in order, and then divides them again (S5) at document structural positions such as chapters and sections that are specified by tagging, etc. The information processing device 1 outputs the context-conscious fragment group 3 obtained by this division by the post-division unit 17.

[0042] 6 is an explanatory diagram illustrating an example of a structuring result 27. As shown in Fig. 6, the information processing device 1 sequentially obtains structuring results 27 for the input document data 2 in the order of a first structuring result for the first document fragment, a second structuring result for the second document fragment, and so on. For these structuring results 27, for example, the second structuring result does not have any unnecessary output at the beginning or end, and a consistent structuring that is directly connected to the first structuring result is achieved.

[0043] 7A and 7B are explanatory diagrams illustrating an example of the output result. As shown in Figures 7A and 7B, the context-considered fragments 3a, 3b, 3c, etc. included in the context-considered fragment group 3 are document fragments in which chapters, sections, tables, etc. are not broken.

[0044] 8 to 11 are explanatory diagrams illustrating a comparative example. As shown in Fig. 8, in case C11, the CSV data at the beginning (lines 1 to 20) is used as the target fragment 22, and a request is made to the large-scale language model M1 to structurize (convert to Markdown) the data using a request sentence that describes only this target fragment 22. This beginning portion is converted (output) without any problems.

[0045] As shown in Figure 9, in case C12, following case C11, the CSV data from lines 21 to 40 is used as target fragment 22, and a request statement describing only target fragment 22 is used to request structuring from large-scale language model M1. In case C12, the request statement does not include local context information, global context information 25, etc., resulting in an output with errors in the item names and number of items in the table header. Also, in case C12, an error occurs in which a comment outside the table is placed inside the table.

[0046] As shown in Figure 10, in case C13, following case C11, the CSV data from lines 21 to 40 is used as the target fragment 22, and a request is made to the large-scale language model M1 to structure the target fragment 22 along with a request statement that describes the previous fragment 23 and the previous fragment conversion result 24. In this case C12, the request statement does not include global context information 25, resulting in an output that includes unnecessary opening sentences and quotations. Furthermore, the output is inconsistent, with headers being re-output and unnecessary supplementary sentences being included.

[0047] As shown in Figure 11, in case C14, following case C11, the CSV data from lines 21 to 40 is used as target fragment 22, and structuring is performed using the same request statement as in case C13. The output example of this case C14 contains an error in which the header of the first line, which has been cut off by the fragment, is considered to be the top level.

[0048] As described above, the information processing device 1 divides the document data 2 to be converted into a plurality of document fragments. When requesting the large-scale language model M1 to convert each of the divided document fragments into a document whose document structure is explicitly stated, the information processing device 1 creates a request statement 26 that includes the previous fragment 23 that immediately precedes the target fragment 22 for which conversion is requested, the target fragment 22 for which conversion is requested, and the previous fragment conversion result 24 after conversion of the previous fragment 23 by the large-scale language model M1. The information processing device 1 inputs the created request statement 26 into the large-scale language model M1 and executes conversion into a document fragment whose document structure is explicitly stated.

[0049] As a result, the information processing device 1 can request the large-scale language model M1 to guide the structuring of the previous fragment 23 and the target fragment 22 so that the structuring up to the previous fragment 23 is completed. In the large-scale language model M1 that receives such a request, the beginning of the converted output is influenced by the conversion results up to the previous fragment 23, i.e., the context and the like included in the structuring result 27 of the previous fragment 23 (the degree of freedom of the beginning is reduced). Therefore, the large-scale language model M1 outputs a consistent structuring result that is influenced by the conversion results up to the previous fragment 23. In this way, the information processing device 1 can prevent the structuring by the large-scale language model M1 from becoming unstable and can support appropriate conversion into structured document data.

[0050] Furthermore, the information processing device 1 creates a request sentence 26 that includes a symbol indicating the output after conversion of the large-scale language model M1, followed by the immediately preceding fragment conversion result 24. This allows the information processing device 1 to create a request sentence 26 that reproduces the situation in which the large-scale language model M1 partially outputs the conversion result with respect to the structuring of the immediately preceding fragment 23 and the target fragment 22. In other words, the information processing device 1 can appropriately guide the output of the large-scale language model M1 as an application of the Leading Word prompt pattern.

[0051] Furthermore, the information processing device 1 extracts global context information 25 indicating the sentence structure related to the document data 2 based on the plurality of previous fragment conversion results 24 after conversion, and creates a request sentence 26 including the extracted global context information 25. This enables the information processing device 1 to make a conversion request to the large-scale language model M1 so as to output a consistent sentence structure based on the plurality of previous fragment conversion results 24 after conversion.

[0052] The information processing device 1 also combines the converted document fragments (structuring result 27) and divides the combined document at positions corresponding to delimiters in the document structure. This allows the information processing device 1 to obtain a context-sensitive fragment group 3 corresponding to a document structure such as a chapter or a section. By obtaining such a context-sensitive fragment group 3, the information processing device 1 can easily present document fragments corresponding to a document structure such as a chapter or a section in a subsequent task (e.g., document search).

[0053] Furthermore, the information processing device 1 extracts first context information (global context information 25) indicating the sentence structure of the document data 2 and second context information indicating the document structure of the immediately preceding fragment 23. When requesting the large-scale language model M1 to convert each divided document fragment into a document whose document structure is explicitly stated, the information processing device 1 creates a request sentence 26 including the first context information and the second context information along with the target fragment 22 for which conversion is requested. The information processing device 1 inputs the created request sentence 26 into the large-scale language model M1 and performs conversion into a document fragment whose document structure is explicitly stated.

[0054] In this way, the information processing device 1 issues a conversion request including the first context information and the second context information to the large-scale language model M1 for each divided document fragment, thereby enabling the large-scale language model M1 to obtain a consistent structuring result in line with the context information. Therefore, the information processing device 1 can prevent structuring by the large-scale language model M1 from becoming unstable and can support appropriate conversion into structured document data.

[0055] Note that the components of each device shown in the figure do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

[0056] Furthermore, the various processing functions of the preprocessing unit 10, local context information extraction unit 12, global context information extraction unit 13, structured fragment DB 14, document structuring request construction unit 15, large-scale language model execution unit 16, and post-segmentation unit 17 of the information processing device 1 may be executed in whole or in part on a CPU (or a microcomputer such as an MPU or MCU (Micro Controller Unit)). It goes without saying that the various processing functions may be executed in whole or in part on a program analyzed and executed by the CPU (or a microcomputer such as an MPU or MCU), or on hardware using wired logic. Furthermore, the various processing functions performed by the information processing device 1 may be executed by multiple computers working together using cloud computing.

[0057] The various processes described in the above embodiments can be realized by executing a program prepared in advance on a computer. Therefore, an example of a computer configuration (hardware) that executes a program having the same functions as those of the above embodiments will be described below. Fig. 12 is an explanatory diagram illustrating an example of a computer configuration.

[0058] 12, computer 200 includes CPU 201 for executing various types of arithmetic processing, input device 202 for receiving data input, monitor 203, and speaker 204. Computer 200 also includes medium reading device 205 for reading programs and the like from a storage medium, interface device 206 for connecting with various devices, and communication device 207 for connecting with external devices via wired or wireless communication. Computer 200 also includes RAM 208 for temporarily storing various types of information, and hard disk drive 209. Each unit (201 to 209) within computer 200 is connected to bus 210.

[0059] The hard disk drive 209 stores a program 211 for executing various processes in the functional configuration described in the above embodiment (e.g., the preprocessing unit 10, the local context information extraction unit 12, the global context information extraction unit 13, the structured fragment DB 14, the document structuration request construction unit 15, the large-scale language model execution unit 16, and the post-segmentation unit 17). The hard disk drive 209 also stores various data 212 referenced by the program 211. The input device 202, for example, accepts input of operation information from an operator. The monitor 203, for example, displays various screens operated by the operator. The interface device 206 is connected to, for example, a printing device. The communication device 207 is connected to a communication network such as a LAN (Local Area Network) and exchanges various information with external devices via the communication network.

[0060] The CPU 201 reads out the program 211 stored in the hard disk drive 209, expands it into the RAM 208, and executes it to perform various processes related to the above-described functional configuration (e.g., the preprocessing unit 10, the local context information extraction unit 12, the global context information extraction unit 13, the structured fragment DB 14, the document structuration request construction unit 15, the large-scale language model execution unit 16, and the post-segmentation unit 17). In other words, the CPU 201 is an example of a control unit. The program 211 does not have to be stored in the hard disk drive 209. For example, the program 211 stored in a storage medium readable by the computer 200 may be read out and executed. Examples of the storage medium readable by the computer 200 include portable storage media such as CD-ROMs, DVDs, USB (Universal Serial Bus) memory, semiconductor memories such as flash memory, and hard disk drives. The program 211 may also be stored in a device connected to a public line, the Internet, a LAN, or the like, and the computer 200 may read out and execute the program 211 from the device.

[0061] The following additional notes are provided regarding the above-described embodiments.

[0062] (Appendix 1) Divide the document data to be converted into multiple document fragments, When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes a document fragment immediately before the document fragment for which conversion is requested, the document fragment for which conversion is requested, and a converted document fragment obtained by converting the immediately before document fragment by the large-scale language model; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing program that causes a computer to execute a process.

[0063] (Supplementary Note 2) The creating process creates the conversion request sentence including a symbol indicating the converted output of the large-scale language model followed by the converted document fragment. 2. The information processing program according to claim 1,

[0064] (Appendix 3) The computer further executes a process of extracting a sentence structure related to the document data based on the plurality of converted document fragments; the creating process creates the conversion request sentence including the extracted sentence structure. 2. The information processing program according to claim 1,

[0065] (Appendix 4) The computer is further caused to execute a process of combining the converted document fragments and dividing the combined document at positions corresponding to the delimiters of the document structure. 2. The information processing program according to claim 1,

[0066] (Appendix 5) Divide the document data to be converted into multiple document fragments, extracting first context information indicating a sentence structure related to the document data and second context information indicating a document structure related to a document fragment immediately before the document fragment for which conversion is requested; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes the document fragments to be converted, as well as the first context information and the second context information; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing program that causes a computer to execute a process.

[0067] (Supplementary Note 6) In the extracting process, a document structure extracted based on the plurality of converted document fragments is used as the first context information. 6. The information processing program according to claim 5,

[0068] (Supplementary Note 7) In the extracting process, the immediately preceding document fragment and a document fragment obtained by converting the immediately preceding document fragment using the large-scale language model are used as the second context information. 6. The information processing program according to claim 5,

[0069] (Appendix 8) Divide the document data to be converted into multiple document fragments, When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes a document fragment immediately before the document fragment for which conversion is requested, the document fragment for which conversion is requested, and a converted document fragment obtained by converting the immediately before document fragment by the large-scale language model; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing method characterized in that the processing is executed by a computer.

[0070] (Appendix 9) The creating process creates the conversion request sentence including a symbol indicating the converted output of the large-scale language model followed by the converted document fragment. 9. The information processing method according to claim 8,

[0071] (Appendix 10) The computer further executes a process of extracting a sentence structure related to the document data based on the plurality of document fragments after conversion; the creating process creates the conversion request sentence including the extracted sentence structure. 9. The information processing method according to claim 8,

[0072] (Appendix 11) The computer further executes a process of combining the converted document fragments and dividing the combined document at positions corresponding to the delimiters of the document structure. 9. The information processing method according to claim 8,

[0073] (Appendix 12) Divide the document data to be converted into multiple document fragments, extracting first context information indicating a sentence structure related to the document data and second context information indicating a document structure related to a document fragment immediately before the document fragment for which conversion is requested; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes the document fragments to be converted, as well as the first context information and the second context information; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing method characterized in that the processing is executed by a computer.

[0074] (Supplementary Note 13) In the extracting process, a document structure extracted based on the plurality of converted document fragments is used as the first context information. 13. The information processing method according to claim 12,

[0075] (Supplementary Note 14) In the extracting process, the immediately preceding document fragment and a document fragment obtained by converting the immediately preceding document fragment using the large-scale language model are used as the second context information. 13. The information processing method according to claim 12,

[0076] (Appendix 15) Divide the document data to be converted into multiple document fragments, When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes a document fragment immediately before the document fragment for which conversion is requested, the document fragment for which conversion is requested, and a converted document fragment obtained by converting the immediately before document fragment by the large-scale language model; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing device comprising: a control unit that executes processing.

[0077] (Appendix 16) The creating process creates the conversion request sentence including a symbol indicating the converted output of the large-scale language model followed by the converted document fragment. 16. The information processing device according to claim 15,

[0078] (Supplementary Note 17) The control unit further executes a process of extracting a sentence structure related to the document data based on the plurality of document fragments after conversion; the creating process creates the conversion request sentence including the extracted sentence structure. 16. The information processing device according to claim 15,

[0079] (Appendix 18) The control unit further executes a process of combining the converted document fragments and dividing the combined document at positions corresponding to the delimiters of the document structure. 16. The information processing device according to claim 15,

[0080] (Appendix 19) Divide the document data to be converted into multiple document fragments, extracting first context information indicating a sentence structure related to the document data and second context information indicating a document structure related to a document fragment immediately before the document fragment for which conversion is requested; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes the document fragments to be converted, as well as the first context information and the second context information; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing device comprising: a control unit that executes processing.

[0081] (Supplementary Note 20) In the extracting process, a document structure extracted based on a plurality of converted document fragments is used as the first context information. 20. The information processing device according to claim 19,

[0082] (Supplementary Note 21) In the extracting process, the immediately preceding document fragment and a document fragment obtained by converting the immediately preceding document fragment using the large-scale language model are used as the second context information. 20. The information processing device according to claim 19, [Explanation of symbols]

[0083] 1...Information processing device 2. Document data 3…Context-sensitive fragments 3a~3c…Context-sensitive fragments 10...Pre-processing section 11...Pre-splitting section 12...Local context information extraction unit 13...Global context information extraction unit 14…Fragment DB after structuring 15...Document structuring request composition section 16...Large-scale language model execution unit 17…Post division part 20...Text document 21...Pre-split fragments 22...Target fragment 23...Fragment just before 24...Result of previous fragment conversion 25...Global context information 25a...Global context information (structure) 25b...Global context information (summary) 26...Request 26a…Long document structuring system prompt 26b…Long design document structuring instructions 26c…Document Structure Hints 27…Structuring result 200...Computer 201...CPU 202...input device 203...Monitor 204...Speaker 205...Media reader 206...Interface device 207...Communication equipment 208...RAM 209...Hard disk drive 210...bus 211…Program 212...Various data C11~C14...Case M1: Large-scale language model

Claims

1. Dividing the document data to be converted into multiple document fragments; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes a document fragment immediately before the document fragment for which conversion is requested, the document fragment for which conversion is requested, and a converted document fragment obtained by converting the immediately before document fragment by the large-scale language model; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing program that causes a computer to execute a process.

2. the creating process creates the conversion request sentence including a symbol indicating the converted output of the large-scale language model followed by the converted document fragment; 2. The information processing program according to claim 1, wherein:

3. further causing the computer to execute a process of extracting a sentence structure related to the document data based on the plurality of converted document fragments; the creating process creates the conversion request sentence including the extracted sentence structure.

2. The information processing program according to claim 1, wherein:

4. further causing the computer to execute a process of combining the converted document fragments and dividing the combined document at positions corresponding to the delimiters of the document structure; 2. The information processing program according to claim 1, wherein:

5. Dividing the document data to be converted into multiple document fragments; extracting first context information indicating a sentence structure related to the document data and second context information indicating a document structure related to a document fragment immediately before the document fragment for which conversion is requested; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes the document fragments to be converted, as well as the first context information and the second context information; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing program that causes a computer to execute a process.

6. the extracting process extracts a document structure based on the plurality of converted document fragments as the first context information; 6. The information processing program according to claim 5,

7. the extracting process uses the immediately preceding document fragment and a document fragment obtained by converting the immediately preceding document fragment using the large-scale language model as the second context information; 6. The information processing program according to claim 5,

8. Dividing the document data to be converted into multiple document fragments; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes a document fragment immediately before the document fragment for which conversion is requested, the document fragment for which conversion is requested, and a converted document fragment obtained by converting the immediately before document fragment by the large-scale language model; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing method characterized in that the processing is executed by a computer.

9. Dividing the document data to be converted into multiple document fragments; extracting first context information indicating a sentence structure related to the document data and second context information indicating a document structure related to a document fragment immediately before the document fragment for which conversion is requested; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes the document fragments to be converted, as well as the first context information and the second context information; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing method characterized in that the processing is executed by a computer.

10. Dividing the document data to be converted into multiple document fragments; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes a document fragment immediately before the document fragment for which conversion is requested, the document fragment for which conversion is requested, and a converted document fragment obtained by converting the immediately before document fragment by the large-scale language model; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing device comprising: a control unit that executes processing.

11. Dividing the document data to be converted into multiple document fragments; extracting first context information indicating a sentence structure related to the document data and second context information indicating a document structure related to a document fragment immediately before the document fragment for which conversion is requested; When requesting a large-scale language model to convert each of the divided document fragments into a document that explicitly indicates the document structure, a conversion request sentence is created that includes the document fragments to be converted, as well as the first context information and the second context information; The created conversion request sentence is input to the large-scale language model to convert it into a document fragment that clearly indicates the document structure. An information processing device comprising: a control unit that executes processing.

Images