Information processing device, information processing method, and program

Abstract

To provide an information processing device that makes it easy to create a detailed specification without any trouble. [Solution] The information processing device includes a receiving unit that receives as input the components of the invention for which the specification is to be created, prior art documents for the invention, and matters described in the specification; an identification unit that performs identification processing to identify components in prior art documents that are similar to the components of the invention for which the specification is to be created; an extraction unit that extracts descriptions related to the components identified by the identification processing from the prior art documents; an instruction unit that instructs a language model to describe the details based on the extracted descriptions; and an output unit that receives and outputs output in response to the instructions from the language model.

Description

[Technical Field]

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

[0002] It is known to generate documents using large-scale language models (LLMs). Patent Document 1 discloses a technology that uses LLMs to assist in generating specifications. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2024-133453 Summary of the Invention [Problem to be solved by the invention]

[0004] In related specification generation systems, when an LLM is used to write a specification, the resulting specification may be thin. This is because the LLM is not specialized in a specific domain. When trying to write a document that deals with knowledge of a specific domain, such as a patent, the LLM may lack the expertise and produce a thin description. To solve this problem, it may be possible to train the LLM to acquire expertise, but this may require retraining. In other words, when using an LLM to create a specification, it may be difficult to create a detailed specification.

[0005] An object of the present disclosure is to provide an information processing device that can solve the above-mentioned problems. [Means for solving the problem]

[0006] The present disclosure provides an information processing device comprising: a receiving means for receiving as input components of an invention for which a specification is to be prepared, a prior document for the invention, and matters described in the specification; a specifying means for performing a specifying process for identifying components of the prior document that are similar to the components of the invention for which the specification is to be prepared; an extraction means for extracting descriptions relating to the components identified by the specifying process from the prior document; an instruction means for instructing a language model to describe the matters based on the extracted description; and an output means for receiving and outputting output in response to the instruction from the language model.

[0007] The present disclosure also provides an information processing method that receives as input components of an invention for which a specification is to be prepared, a prior document for the invention, and matters described in the specification, performs a specification process to identify components of the prior document that are similar to the components of the invention for which the specification is to be prepared, extracts descriptions related to the components identified by the specification process from the prior document, instructs a language model to describe the matters based on the extracted descriptions, and receives and outputs an output in response to the instructions from the language model.

[0008] The present disclosure also provides a program that causes a computer to execute the following processes: a process of accepting as input components of an invention for which a specification is to be prepared, prior art documents for the invention, and matters described in the specification; a process of performing a specification process to identify components of the prior art documents that are similar to the components of the invention for which the specification is to be prepared; a process of extracting descriptions related to the components identified in the specification process from the prior art documents; a process of instructing a language model to describe the matters based on the extracted descriptions; and a process of receiving and outputting output in response to the instructions from the language model. [Effects of the Invention]

[0009] According to the present disclosure, detailed specifications can be easily prepared without any hassle. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a block diagram showing a hardware configuration of an information processing device according to the present disclosure. [Figure 2] FIG. 1 is a block diagram illustrating a functional configuration of an information processing device according to the present disclosure. [Figure 3] 10 is a flowchart illustrating an operation of an information processing device according to the present disclosure. [Figure 4] 1 is a schematic block diagram showing a functional configuration of an information processing device according to the present disclosure. [Figure 5] 10 is a flowchart illustrating an example of a procedure of a process performed by an information processing device according to the present disclosure. [Figure 6] FIG. 1 is a block diagram illustrating a functional configuration of an information processing device according to the present disclosure. [Figure 7] 10 is a flowchart illustrating an operation of an information processing device according to the present disclosure. [Figure 8] FIG. 1 is a block diagram illustrating a functional configuration of an information processing device according to the present disclosure. [Figure 9] 10 is a flowchart illustrating a processing procedure of an information processing device according to the present disclosure. [Figure 10] 1 is a schematic block diagram showing a functional configuration of an information processing device according to the present disclosure. [Figure 11] 10 is a flowchart illustrating an operation of an information processing device according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0011] First, the hardware configuration of an information processing device will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the hardware configuration of an information processing device according to an embodiment of the present disclosure.

[0012] 1 is a diagram showing an example of the hardware configuration of an information processing device. The information processing device is realized by a computer 10.

[0013] The computer 10 includes a processor 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage device 14 such as a hard disk for storing programs, an input / output interface 15 for inputting and outputting data, and a communication interface 16 for network connection. Each component is connected via a bus 17.

[0014] The processor 11 runs an operating system to control the entire computer. Examples of the processor 11 include a CPU (Central Processing Unit), a DSP (Digital Signal Processor), and a GPU (Graphics Processing Unit). The processor 11 loads a program stored in, for example, the ROM 12 or the storage device 14. The processor 11 then executes each process coded in the program. The processor 11 may execute the processes or instructions in the illustrated flowchart based on the program.

[0015] The ROM 12 stores application programs, programs according to the embodiments, etc. The RAM 13 is used as a work area for the processor 11.

[0016] Examples of the storage device 14 include a semiconductor memory such as a flash memory, an HDD (Hard Disk Drive), etc. The storage device 14 stores, for example, an OS (Operating System) program, application programs, programs according to each embodiment, etc.

[0017] The input / output interface 15 is connected to peripheral devices (not shown) via a wired network or a wireless network.

[0018] The communication interface 16 is connected to a communication network (not shown), such as a LAN (Local Area Network) or a WAN (Wide Area Network), via a wireless or wired network. The communication network may be configured with multiple communication networks. As a result, the computer is connected to an external large-scale language model server that configures a large-scale language model LLM via the communication network. The computer 10 may have components other than those shown in FIG. 1. For example, the computer 10 may have a drive device or the like. For example, the processor 11 may be attached to a drive device or the like and read programs and data stored in a non-transitory tangible recording medium into the RAM 13.

[0019] The information processing device may be configured to include some of the components described in Fig. 1. For example, the information processing device may be configured to include only the processor 11, ROM 12, and RAM 13 among the components described above. In this case, the storage device 14, input / output interface 15, and communication interface 16 may each be provided as a device external to the information processing device. Furthermore, some of the calculation functions of the information processing device may be realized by an external server, a cloud, or the like.

[0020] First, the first embodiment will be described.

[0021] Fig. 2 is a block diagram showing the functional configuration of an information processing device. As shown in Fig. 2, the information processing device 100 includes a reception unit 110, an identification unit 120, an extraction unit 130, an instruction unit 140, and an output unit 150. The information processing device 100 is connected to a language model 101. In the example shown in Fig. 1, the language model 101 is provided outside the information processing device 100 and is connected to the information processing device 100 via a communication network. However, the information processing device 100 may include the language model 101 internally, and the specific mode of connection is not limited.

[0022] The receiving unit 110 is a receiving means that receives as input the components of the invention for which a specification is to be prepared, prior document information, and written matters. For example, when a user attempts to prepare a specification for a patent application for an invention called an "image processing device," the receiving unit 110 receives components such as an "image processing unit," a "display control unit," and an "operation receiving unit," information on related prior patent documents, and written matters such as "problems that the invention aims to solve" and "means for solving the problems." The receiving unit 110 may also receive instructions to correct the specification output by the output unit 150.

[0023] The reception unit 110 receives the above-mentioned information through a user interface. The reception unit 110 may use a voice recognition system. In this case, the user can input the elements of the invention, prior art document information, and description items by voice. The reception unit 110 may also use a handwriting recognition system. In this case, the user can input information by handwriting on a touch screen or tablet. The reception unit 110 may also receive input via email or an API (Application Programming Interface).

[0024] The identification unit 120 is an identification means that performs an identification process to identify components in a prior document that are similar to the components of the invention received by the receiving unit 110. The identification unit 120 uses techniques such as vector search and semantic search to identify components in a prior document that are semantically similar to the input component. For example, for a component called "image processing unit," it identifies similar components such as "image processing unit" and "video processing module" from the prior document. In a vector search, for example, the components are vectorized, and the similarity is determined using cosine similarity. Alternatively, the identification unit 120 may use a general keyword search.

[0025] The identification unit 120 may perform the identification process using a method that uses a machine learning model different from the vector search or semantic search. In this case, the machine learning model is trained on a large number of patent documents and identifies components in prior documents that are similar to the input component.

[0026] The prior document is, for example, a patent document, but is not limited to a patent document and may include various technical documents.

[0027] The extraction unit 130 is an extraction means that extracts descriptions related to the components identified by the identification unit 120 from prior documents. The extraction unit 130 performs extraction processing using a language model 101. The language model 101 is, for example, a large language model (LLM). The LLM understands a given context and extracts related information. For example, the LLM extracts descriptions related to an "image processing unit," such as its function, operation, and relationships with other components. The type of language model 101 is not particularly limited. Multiple different language models may be used as the language model 101. The extraction unit 130 may also use a rule-based natural language processing system as the language model 101. In this case, related descriptions may be extracted from prior documents based on predefined rules. The extraction unit 130 may extract descriptions using multiple methods in combination.

[0028] The instruction unit 140 is an instruction means for instructing the language model 101 to write description items based on the descriptions extracted by the extraction unit 130. For example, the instruction unit 140 instructs the language model 101 to "describe the operation of the image processing unit in about 500 characters. In doing so, be sure to mention the 'image data compression processing' and 'noise removal function' extracted from the prior art document."

[0029] The instruction by the instruction unit 140 is expressed as a prompt (instruction sentence) to be input to the language model 101. The instruction by the instruction unit 140 may be one prompt or multiple prompts. Furthermore, the instruction by the instruction unit 140 is not limited to a prompt. For example, tokenized information corresponding to the above-mentioned prompt may be input to the language model 101 as an instruction. The instruction by the instruction unit 140 may be not only a text-based instruction but also a structured data format (e.g., JSON, XML).

[0030] The output unit 150 is an output means that receives and outputs the output from the language model 101. The output sentence is a patent application specification or a sentence that is a part of the specification, written in accordance with the instructions of the instruction unit 140. For example, the sentence "Problem to be solved by the invention" output from the output unit 150 may be displayed on a display or printed by a printer. The output form from the output unit 150 is not limited to this. Furthermore, based on the output content, the user can make corrections or additional instructions as necessary.

[0031] Next, the operation of the information processing device 100 will be described with reference to Fig. 3. Fig. 3 is a flowchart showing the operation (information processing method) of the information processing device 100.

[0032] First, the receiving unit 110 receives as input the components of the invention for which a specification is to be prepared, prior art documents of the invention, and matters to be described in the specification (step S101).

[0033] Next, the identifying unit 120 identifies elements of the prior document that are similar to the elements of the received invention (step S102).

[0034] Next, the extraction unit 130 extracts descriptions relating to the components of the identified prior document from the prior document (step S103).

[0035] Thereafter, the instruction unit 140 instructs the language model 101 to write the description based on the extracted description (step S104).

[0036] Finally, the output unit 150 receives and outputs the output of the language model 101 (step S105).

[0037] According to the information processing device 100 of this embodiment, since the description is identified based on past specifications, new training of the language model is unnecessary, contributing to the creation of detailed specifications without effort. For example, the identification unit 120 identifies elements in prior documents that are similar to a specified element, the extraction unit 130 extracts the description, and the instruction unit 140 provides instructions to the language model 101. This allows for the creation of detailed specifications for a specific domain without retraining. This eliminates the thinness of descriptions that tends to result from a lack of expertise in general-purpose LLMs, and facilitates the creation of detailed specifications that effectively utilize knowledge from past specifications without effort. In other words, the information processing device 100 of this embodiment facilitates the improvement of specification quality and the reduction of time required for specification creation by efficiently utilizing knowledge of prior art. Note that this embodiment will be described using the creation of patent specifications as an example, but the scope of application of the described method is not limited to patent specification creation.

[0038] The instruction unit 140 may also use a template-based sentence generation system instead of issuing an instruction to the language model 101. In this case, the extracted description can be applied to a template prepared in advance to generate a description of the description item.

[0039] The second embodiment will be described below.

[0040] 4 is a schematic block diagram showing the functional configuration of an information processing device. In the figure, the information processing device 200 includes a reception unit 210, a specification unit 220, an extraction unit 230, an instruction unit 240, an output unit 250, and a setting unit 260. The reception unit 210, the specification unit 220, the extraction unit 230, the instruction unit 240, and the output unit 250 have the same basic functions as the reception unit 110, the specification unit 120, the extraction unit 130, the instruction unit 140, and the output unit 150 of the first embodiment, respectively. Furthermore, similar to the example of the information processing device 100, the information processing device 200 is connected to a language model 201.

[0041] The receiving unit 210 receives as input the elements of the invention for which a specification is to be prepared, prior document information, and description matters. For example, if a user inputs the element "image recognition device," specifies "JP Patent Publication No. 2020-1xxxxx" as the related prior document, and selects "Problem to be solved by the invention" as the description matter, the receiving unit 210 receives this information.

[0042] The identification unit 220 performs an identification process to identify components in a prior document that are similar to components of the invention received by the reception unit 210. For example, in response to an input of "image recognition device," the identification unit 220 uses a vector search or a semantic search to identify similar components such as an "image processing unit" or an "object recognition unit" in the prior document.

[0043] The extraction unit 230 extracts from the prior document a description relating to the component identified by the identification unit 220. For example, the extraction unit 230 extracts a description relating to an "image processing unit" such as "image data captured by a camera is input, and preprocessing such as edge detection and noise removal is performed."

[0044] The instruction unit 240 instructs the language model 201 to write a description based on the description extracted by the extraction unit 230. The language model 201 is, for example, an LLM. For example, the instruction unit 240 instructs the language model 201 to generate a description of the "problem to be solved by the invention" that takes into account the features of the new invention while referring to the extracted description.

[0045] The output unit 250 receives and outputs an output corresponding to an instruction from the instruction unit 240 from the language model 201. For example, the sentence "Problem to be solved by the invention" output from the output unit 250 is displayed on a display or printed by a printer.

[0046] The setting unit 260 sets the weight of a prior document based on the type of the prior document. The weight of a prior document represents the degree of importance or relevance of the prior document that can be identified when extracting a description. The type may include, for example, the applicant, inventor, technical field, and application date. However, examples of the type are not limited to these, and any item that can be related to the weight may be used as the type.

[0047] For example, the setting unit 260 assigns a high weight (e.g., 0.8) to the company's own patent applications, a medium weight (e.g., 0.5) to other companies' patent applications, and a low weight (e.g., 0.3) to non-patent documents. The setting unit 260 may also assign weights according to the degree of agreement between the technical field of the invention and the invention. Furthermore, the setting unit 260 may assign weights according to the degree of agreement between the inventor and the invention for which the specification is being prepared. Note that weights are expressed as numbers between 0 and 1, for example, but are not limited to this and may be set as appropriate. If appropriate information about the type of prior document cannot be obtained, the setting unit 260 may assign a weight that is determined in advance.

[0048] The instruction unit 240 changes the description instructions for the description items based on the extracted descriptions based on the weights set by the setting unit 260. For example, if a high weight is set for a prior document that is a company's own patent, the instruction unit 240 adds an instruction to the language model 201 such as "Please refer to the description of the company's own patent first and use similar expressions." Furthermore, if only one prior document is identified, the instruction unit 240 may change the description instructions according to the weight set for the identified prior document. For example, if the weight is greater than a preset standard, the instruction unit 240 may change the instruction to refer to that document.

[0049] The output unit 250 may output an output in response to an instruction from the language model 201 and a weight set for the prior document.

[0050] With this configuration, the information processing device 200 can more appropriately adjust the content of the generated specification depending on the type and importance of the prior art. For example, it becomes possible to efficiently create a specification that emphasizes the prior art of the company or that uses expressions specialized for a specific technical field.

[0051] Next, the operation of the information processing device 200 in the second embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of the procedure of processing performed by the information processing device 200.

[0052] First, the receiving unit 210 receives from the user the components of the invention, prior document information, and description items (step S201). For example, the receiving unit 210 receives input from the user of the components "image recognition device," the prior document "JP Patent Publication No. 2020-1xxxxx," and the description item "Problem to be solved by the invention."

[0053] Next, the identifying unit 220 identifies components of the prior document that are similar to the components of the received invention (step S202). For example, the identifying unit 220 uses vector search or semantic search to identify components such as an "image processing unit" or an "object recognition unit" that are similar to an "image recognition device."

[0054] Next, the setting unit 260 sets a weight based on the type of prior document (step S203). For example, if "JP Patent Publication No. 2020-123456" is a patent application of the company, the setting unit 260 sets the weight to 0.8. This weighting is a numerical representation of the importance and relevance of the prior document, and is used in subsequent processing. The weights that are set are not limited to this.

[0055] Thereafter, the extraction unit 230 extracts descriptions related to the components of the identified prior document (step S204). For example, as a description related to an "image processing unit," a description such as "image data captured by a camera is input, and preprocessing such as edge detection and noise removal is performed" is extracted.

[0056] Next, the instruction unit 240 changes the instruction content to the language model 201 based on the set weight (step S205). For example, if a high weight is set for a prior document that is a company's own patent, the instruction unit 240 adds an instruction to the language model 201 such as "please refer to the description of the company's own patent with priority and use similar expressions."

[0057] Then, the instruction unit 240 instructs the language model 201 to write the description items based on the changed instruction content (step S206). For example, the instruction unit 240 instructs the language model 201 to generate a description of the "problem to be solved by the invention" by referring to the extracted description and taking into account the features of the new invention.

[0058] When the language model 201 generates a description based on the instruction, the output unit 250 receives and outputs the output of the language model 201 (step S207). For example, the generated sentence "Problem to be solved by the invention" is displayed on a display or printed by a printer.

[0059] In this way, the information processing device 200 of the second embodiment can efficiently generate more appropriate description of the specification by weighting the prior art according to the importance of the prior art. The information processing device 200 can efficiently create a description that emphasizes the prior art of the company or a specification that uses expressions specialized for a specific technical field, for example.

[0060] According to the information processing device 200 of this embodiment, by setting weights based on the type of prior document and adjusting the instructions to the language model 201 according to the weights, it is possible to efficiently generate more appropriate descriptions in the specification. For example, by setting high weights for the company's own patents and prior documents with high relevance, it is possible to create a specification that places more emphasis on the descriptions in those documents. This makes it possible to generate a higher quality specification that not only references the descriptions in the prior documents but also takes into account the importance and relevance of the prior documents.

[0061] The third embodiment will be described below.

[0062] Fig. 6 is a block diagram showing the functional configuration of an information processing device. As shown in Fig. 6, the information processing device 300 of this embodiment includes a reception unit 310, a specification unit 320, an extraction unit 330, an instruction unit 340, and an output unit 350. The reception unit 310, the specification unit 320, the extraction unit 330, the instruction unit 340, and the output unit 350 have the same basic functions as the reception unit 110, the specification unit 120, the extraction unit 130, the instruction unit 140, and the output unit 150 of the first embodiment, respectively. Furthermore, similar to the example of the information processing device 100, the information processing device 300 is connected to a language model 301.

[0063] The receiving unit 310 receives as input the components of the invention for which a specification is to be prepared, prior document information, and description matters. For example, if a user inputs the component "face recognition camera," specifies "high-precision face recognition system" and "personal authentication method using image processing" as related prior documents, and selects "modes for carrying out the invention" as a description matter, this information is received.

[0064] The identification unit 320 performs an identification process to identify components in prior documents that are similar to components of the invention received by the receiving unit 310. In this embodiment, the identification unit 320 performs the identification process multiple times using different parameters. In this embodiment, the parameters are parameters related to the search method used by the identification unit 320 for identification, such as parameters related to the algorithm selected by the identification unit 320. In a more detailed example, when the identification unit 320 identifies components in prior documents using a semantic search and a keyword search, the parameters may be values ​​indicating the weights of the two search methods. For example, the identification unit 320 performs three searches using different parameters and identifies related components from all the obtained prior documents. In this way, the identification unit 320 aims to identify more related components in prior documents by performing the identification process multiple times using different parameters.

[0065] Furthermore, an increase in the number of specific processes may result in longer processing times and increased waiting times for users, especially when there is a large amount of prior art or when the invention has complex components.

[0066] In this case, the identifying unit 320 may set an upper limit on the processing time, and when the upper limit is reached, identify related components from the prior documents obtained at that time. For example, if a time limit of 30 seconds is set for each identifying process, and the total time taken for three identifying processes exceeds 90 seconds, the identifying unit 320 may identify related components from the prior documents obtained at that time.

[0067] The extraction unit 330 extracts descriptions related to the components of the prior documents identified by the identification unit 320. In this embodiment, when there are multiple prior documents, the extraction unit 330 extracts descriptions common to those documents. Furthermore, the extraction unit 330 excludes descriptions related to specific data or numerical values ​​from the extraction target.

[0068] Descriptions common to multiple prior documents are expected to be general or important technical content in the technical field to which the specification is being written. Therefore, the extraction unit 330 extracts descriptions common to multiple prior documents. Furthermore, specific data or numerical values ​​may be inappropriate for explaining the invention for which the specification is being written, since the underlying conditions differ from document to document. Therefore, the extraction unit 330 excludes descriptions related to specific data or numerical values ​​from the extraction target.

[0069] For example, for an invention of a "face recognition camera" for which a specification is to be prepared, the extraction unit 330 extracts the description "extracting facial feature points and comparing them with registered data," which is found in both a prior document related to a "high-precision face recognition system" and a prior document related to a "personal authentication method using image processing." On the other hand, for a similar invention for which a specification is to be prepared, the extraction unit 330 does not extract descriptions related to specific numerical values, such as "the camera has a resolution of 12 million pixels or more," which are found in prior documents.

[0070] In addition, when the extraction unit 330 extracts descriptions common to multiple prior documents, if only one highly relevant prior document is found or if there are very few common descriptions, the extraction unit 330 may not be able to extract sufficient descriptions.

[0071] In this case, if the common description is less than a certain amount, the extraction unit 330 may extract descriptions that are considered important from individual prior documents. For example, if the common description is less than 100 characters, the extraction unit 330 may extract additional descriptions from the prior document with the highest similarity, extracting a total of approximately 300 characters. Note that the number of characters to be extracted is not limited to this and may be determined appropriately depending on the required amount of description and the similarity of the prior documents.

[0072] Furthermore, important information may be lost if the extraction unit 330 excludes descriptions related to specific data or numerical values ​​from the extraction target. For example, qualitative expressions such as "high precision" or "high speed" may not be enough to fully explain the characteristics of an invention.

[0073] Therefore, when excluding specific numerical values ​​from the extraction target, the extraction unit 330 may extract expressions that indicate the tendency or relative magnitude relationship of the numerical values. For example, instead of excluding the statement "The camera resolution is 12 million pixels or more" from the extraction target, the extraction unit 330 may extract the expression "A high-resolution camera is used."

[0074] The instruction unit 340 instructs the language model 301 to write description items based on the descriptions extracted by the extraction unit 330. For example, the instruction unit 340 gives an instruction such as "Regarding 'face recognition camera,' please generate a general description of 'Mode for carrying out the invention' that does not include specific numerical values ​​or data, based on descriptions commonly extracted from multiple prior documents."

[0075] The output unit 350 receives and outputs an output in response to an instruction from the language model 301. For example, the output unit 350 displays the generated description of the "Description of the Preferred Embodiments" on a display.

[0076] Next, a description will be given of the operation of the information processing device 300 of this embodiment. Fig. 7 is a flowchart showing the operation of the information processing device 300 according to this embodiment.

[0077] First, the receiving unit 310 receives the components of the invention for which a specification is to be prepared, prior art document information, and description items (step S301).

[0078] Next, the identifying unit 320 performs the identifying process multiple times with different parameters to identify similar components of prior documents (step S302).

[0079] Next, the extraction unit 330 extracts descriptions related to the components of the identified prior document from the prior document (step S303). In step S303, if there are multiple prior documents, the extraction unit 330 may extract common descriptions. In addition, in step S303, the extraction unit 330 may exclude descriptions related to specific data and numerical values ​​from the extraction target.

[0080] Thereafter, the instruction unit 340 instructs the language model 301 to write the description based on the extracted description (step S304).

[0081] The output unit 350 receives and outputs the output of the description generated by the language model 301 based on the instruction (step S305).

[0082] As described above, the information processing device 300 of this embodiment can generate a more general-purpose and highly reusable specification by extracting common descriptions from multiple prior documents and excluding specific data and numerical values. This eliminates the thin description that tends to be caused by a lack of specialization in general-purpose LLMs, while avoiding overly specific descriptions, making it possible to create a specification that covers a wide range of embodiments.

[0083] In addition, by performing the identification process multiple times with different parameters, it is possible to identify more relevant components of prior documents and improve extraction accuracy.Furthermore, by extracting only common descriptions, it is possible to focus on important information that is consistent across multiple prior documents, resulting in the generation of more reliable descriptions.

[0084] These measures enable the information processing device 300 of this embodiment to generate specification descriptions more efficiently and effectively. By limiting the processing time, it is possible to collect as much relevant information as possible while appropriately managing user waiting time. Furthermore, measures taken when there are few common descriptions ensure that a certain amount of useful description is always available. Furthermore, by replacing numerical data with expressions that show trends, it is possible to appropriately express the features of the invention while avoiding specific numerical values.

[0085] The fourth embodiment will be described below.

[0086] Fig. 8 is a block diagram showing the functional configuration of an information processing device. As shown in Fig. 8, the information processing device 400 includes a reception unit 410, a specification unit 420, an extraction unit 430, an instruction unit 440, and an output unit 450. Similarly to the example of the information processing device 100, the information processing device 400 is connected to a language model 401.

[0087] The receiving unit 410 receives as input the components of the invention for which a specification is to be prepared, prior document information, and description matters. For example, if a user is trying to prepare a specification for an invention called an "image processing device," the receiving unit 410 receives components such as an "image processing unit," "display control unit," and "operation reception unit," information on related prior patent documents, and description matters such as "problems that the invention aims to solve" and "means for solving the problems."

[0088] The identification unit 420 performs an identification process to identify components in a prior document that are similar to the components of the invention received by the reception unit 410. The identification unit 420 uses techniques such as vector search and semantic search to identify components in a prior document that are semantically similar to the input component. For example, for a component called an "image processing unit," the identification unit 420 identifies similar components such as an "image processing unit" and a "video processing module" from the prior document.

[0089] The extraction unit 430 extracts from the prior document descriptions relating to the constituent elements identified by the identification unit 420. For example, the language model 401 may be used in this extraction process.

[0090] The instruction unit 440 instructs the language model 401 to write the description based on the description extracted by the extraction unit 430.

[0091] The instruction unit 440 instructs the language model 401 to determine whether adding a description based on the description extracted by the extraction unit 430 will cause a contradiction with the content of the invention to be described, and to change the description of the description based on the extracted description based on the result of the determination. That is, the instruction unit 440 determines whether adding the extracted description will cause a contradiction with the content of the invention, and instructs the language model 401 to change the description as necessary.

[0092] For example, the instruction unit 440 may give an instruction to the language model 401 such as, "If the description of the extracted 'image processing unit' is added to the description of the 'image processing unit' of the present invention, please determine whether there is any contradiction with the purpose of the invention, 'realizing real-time processing.' If there is a contradiction, please revise the description to eliminate the contradiction."

[0093] The instruction unit 440 also issues instructions to determine whether the components of the prior document corresponding to the components of the invention for which the specification is being written have a function that is not included in the components of the invention for which the specification is being written, and if it is determined that the components have this function, to determine whether adding this function will cause a contradiction in the invention for which the specification is being written, and if it is determined that no contradiction will occur, to instruct the language model 401 to add this function. That is, the instruction unit 440 determines whether an additional function that a component of a prior document has can be added to the invention, and if there is no contradiction, to instruct the language model 401 to add it.

[0094] In this case, the instruction unit 440 may instruct the language model 401, for example, "If the 'noise removal function' of the 'image processing unit' in the prior document is added to the 'image processing unit' of the present invention, please determine whether there is any problem with the purpose of the invention and consistency with other components. If there is no problem, please generate a description of the 'image processing unit' with this function added."

[0095] The output unit 450 receives and outputs an output corresponding to an instruction from the language model 401.

[0096] Next, the operation of the information processing device 400 will be described with reference to Fig. 9. Fig. 9 is a flowchart showing an example of the operation of the information processing device 400.

[0097] First, the receiving unit 410 receives as input the components of the invention for which a specification is to be prepared, prior art documents of the invention, and matters to be described in the specification (step S401).

[0098] Next, the identifying unit 420 identifies elements of the prior document that are similar to the elements of the received invention (step S402).

[0099] Next, the extraction unit 430 extracts descriptions relating to the components of the identified prior document from the prior document (step S403).

[0100] Thereafter, the instruction unit 440 instructs the language model 401 to write the description matter based on the extracted description (step S404). In step S404, the instruction unit 440 gives instructions to determine whether writing the description matter based on the description extracted by the extraction unit 430 will cause a contradiction with the content of the invention for which the specification is being written, and to change the description of the description matter based on the extracted description based on the determination result. The instruction unit 440 also gives instructions to determine whether the components of the prior document corresponding to the components of the invention for which the specification is being written have a function that is not included in the components of the invention for which the specification is being written, and if it is determined that the components have this function, gives instructions to determine whether adding this function will cause a contradiction with the invention for which the specification is being written, and if it is determined that no contradiction will occur, gives instructions to add this function.

[0101] Finally, the output unit 450 receives and outputs the output from the language model 401 (step S405).

[0102] With this configuration, the information processing device 400 of this embodiment can effectively utilize the descriptions in prior art documents and support the writing of detailed specifications in a manner that does not impair the essence of the invention. In particular, the information processing device 400 can determine inconsistencies and evaluate additional functions, thereby enabling the writing of higher quality specifications.

[0103] The inconsistency determination function and additional function evaluation function, which are features of this embodiment, eliminate the need for users to check the consistency between prior art documents and the content of the invention, or the appropriateness of additional functions one by one, significantly reducing the burden of specification drafting. At the same time, by having the LLM make these judgments automatically, it is possible to prevent minor inconsistencies that users tend to overlook, and to prevent the omission of useful additional functions, resulting in the efficient drafting of higher quality specifications.

[0104] The fifth embodiment will be described below.

[0105] Fig. 10 is a schematic block diagram showing the functional configuration of an information processing device. In Fig. 10, the information processing device 500 includes a reception unit 510, an identification unit 520, an extraction unit 530, an instruction unit 540, and an output unit 550. The reception unit 510, the identification unit 520, the extraction unit 530, the instruction unit 540, and the output unit 550 have the same basic functions as the reception unit 110, the identification unit 120, the extraction unit 130, the instruction unit 140, and the output unit 150 of the first embodiment, respectively. Furthermore, similar to the example of the information processing device 100, the information processing device 500 is connected to a language model 501.

[0106] The information processing device 500 uses a language model 501 to create a specification, and iteratively corrects and improves the specification.

[0107] The receiving unit 510 receives as input the components of the invention for which a specification is to be prepared, prior document information, and written matters. For example, if a user wants to prepare a specification for an invention called an "image processing device," the receiving unit 510 receives components such as an "image processing unit" and a "display control unit," related prior patent documents, and written matters such as "problems to be solved by the invention" and "effects of the invention." The receiving unit 510 may also receive instructions to correct the specification output by the output unit 550.

[0108] The identification unit 520 performs an identification process to identify components in a prior document that are similar to the components of the invention received by the receiving unit 510. The identification unit 520 uses techniques such as vector search and semantic search to identify components in a prior document that are semantically similar to the input component. For example, for a component called an "image processing unit," the identification unit 520 identifies "image processing unit" and "video processing module" in the prior document as similar components. As with the example of the identification unit 120 described above, the specific identification method of the identification unit 520 is not limited to this.

[0109] The extraction unit 530 extracts from the prior literature a description relating to the component identified by the identification unit 520. For example, the extraction unit 530 extracts a description relating to an "image processing unit" such as "image data captured by a camera is input, and preprocessing such as edge detection and noise removal is performed."

[0110] The instruction unit 540 instructs the language model 501 to write description items based on the descriptions extracted by the extraction unit 530. For example, the instruction may be, "Please write a description of the operation of the image processing unit in about 200 characters. In doing so, be sure to mention the 'image data compression processing' and 'noise removal function' extracted from the prior art document."

[0111] Furthermore, the instruction unit 540 may repeatedly instruct the creation and revision of the specification based on the content of the specification generated by the language model 501. For example, the instruction unit 540 performs the following process.

[0112] The instruction unit 540 instructs the language model 501 to create an initial specification based on the output received from the language model 501. For the created specification, the identification unit 520 again identifies elements of the prior document that are similar to the elements of the created specification. The extraction unit 530 extracts descriptions related to the newly identified elements of the prior document. Based on the extracted new descriptions, the instruction unit 540 instructs the language model 501 to modify the descriptions of the previously created specification.

[0113] For example, if the explanation of the "image processing unit" was insufficient in the initial specification, the instruction unit 540 gives the language model 501 an instruction such as, "Please provide a more detailed explanation of the image processing unit. In particular, please mention the 'real-time processing function' newly extracted from the prior art and explain its advantages in the present invention."

[0114] The output unit 550 receives and outputs an output corresponding to an instruction from the language model 501. For example, the content of the specification generated by the language model 501, details of any corrections, etc. are output by the output unit 550.

[0115] With this configuration, the information processing device 500 can repeatedly revise the content of a specification after the initial specification creation. This makes it possible to efficiently create a more detailed and comprehensive specification. This can be expected to reduce the time required to create a specification and improve the quality of the created specification.

[0116] Next, a description will be given of the operation of the information processing device 500. FIG.

[0117] The receiving unit 510 receives components of the invention, prior art document information, and description items from the user (step S501).

[0118] The identifying unit 520 identifies elements of the prior document that are similar to elements of the received invention (step S502).

[0119] The extraction unit 530 extracts descriptions relating to the constituent elements of the identified prior document (step S503).

[0120] The instruction unit 540 instructs the language model 501 to write the description based on the description extracted in step S503 (step S504).

[0121] The output unit 550 receives and outputs the output from the language model 501 (step S505).

[0122] The instruction unit 540 instructs the language model 501 to create a specification (step S506).

[0123] The identifying unit 520 identifies elements in the prior document that are similar to elements described in the created specification (step S507).

[0124] The extraction unit 530 extracts a description relating to the newly identified component (step S508).

[0125] The instruction unit 540 instructs the language model 501 to revise the specification based on the newly extracted description (step S509).

[0126] The output unit 550 receives and outputs the revised specification, which is the output of the language model 501 (step S510).

[0127] The user checks the output statement and determines whether further corrections are necessary. The need for corrections is determined, for example, based on the completeness of the statement and the importance of newly extracted information. The reception unit 510 receives a correction instruction for the output statement (step S511). If corrections are necessary (step S511: Yes), the process returns to step S507, for example, and is repeated. If corrections are not necessary (step S511: No), the process ends. Note that cases where corrections are not necessary include, for example, when the reception unit 510 receives an input indicating that corrections are not necessary, or when an appropriately determined predetermined time has passed without the reception unit 510 receiving a correction instruction.

[0128] As described above, the information processing device 500 of this embodiment can improve the specification created using the language model 501, making it possible to efficiently create a higher quality specification. The information processing device 500 can solve the problem of thin descriptions caused by a tendency for general-purpose LLMs to lack expertise. In other words, it is possible to create a higher quality specification by supplementing important information that was overlooked during the initial specification creation or more appropriate expressions in subsequent processing.

[0129] According to this embodiment, the specification created using the language model 501 can be iteratively improved. This makes it possible to repeatedly improve the content of the specification even after the initial creation, thereby efficiently creating a higher quality specification. For example, the identifying unit 520 identifies similar components in prior documents again based on the components of the created specification, the extracting unit 530 extracts descriptions related to the newly identified components, and the instructing unit 540 instructs the language model 501 to revise the specification based on the new descriptions. This process can improve the quality of the specification.

[0130] Furthermore, this iterative process can improve the thinness of the description, which tends to result from a lack of general LLM expertise. Subsequent processing can supplement important information overlooked in the initial specification drafting and more appropriate wording, resulting in a higher-quality specification. For example, it becomes easier to add or revise technical details not mentioned in the initial specification drafting or the use of more appropriate terminology through the iterative process.

[0131] Furthermore, the information processing device 500 makes it easy to shorten the time required to prepare a specification and improve the quality of the specifications that are prepared. By using the language model 501, it becomes possible to extract relevant information from prior art documents and prepare and improve specifications based on that information, thereby reducing the time required to prepare a specification. In addition, by using the language model 501, it is possible to ensure consistency in the writing style and use of technical terminology throughout the specification, making it easy to standardize quality.

[0132] In this way, the information processing device 500 of this embodiment can realize efficient specification creation by utilizing the language model 501 and correcting the specification through repeated processing.

[0133] In this embodiment, the information processing device 500 creates a specification and repeatedly modifies it using a language model 501. This makes it possible to extract appropriate similar components and useful descriptions even when the prior document information provided by the user is insufficient or when there are few prior documents related to a specific technical field.

[0134] The prompting unit 540 may generate a message prompting the user to provide additional prior document information. For example, the message may say, "Relevant prior documents are insufficient. Please provide more prior document information." The output unit 550 outputs this message.

[0135] Furthermore, if no similar components are found for a specific component, the instruction unit 540 may inquire of the language model 501 about general technical information about the component and instruct the language model 501 to supplement the description in the specification based on that information. For example, the instruction unit 540 instructs the language model 501 to say, "No similar components are found for the image processing unit. Please generate a general description of the image processing technology in about 200 characters and incorporate it into the specification."

[0136] The identifying unit 520 may also gradually widen the search range for similar components. For example, the identifying unit 520 may initially identify components in prior documents by seeking stricter similarity, and if sufficient results are not obtained, may identify components in prior documents by lowering the threshold for similarity. This allows the identifying unit 520 to identify a wide range of potentially related components even if no directly similar components are found.

[0137] In this way, the information processing device 500 of this embodiment can easily create a specification that maintains a certain level of quality even when prior art information is insufficient. Also, by prompting the user to provide additional information, it is possible to support information gathering for creating a more comprehensive specification.

[0138] The present disclosure may be realized by recording a program for realizing the above-described functions on a computer-readable recording medium, and loading and executing the program recorded on the recording medium into a computer system. Note that the term "computer system" here includes hardware such as an OS and peripheral devices.

[0139] Additionally, "computer-readable recording media" refers to portable media such as flexible disks, optical magnetic disks, ROMs, CD-ROMs, etc., and storage devices such as hard disks built into computer systems. Furthermore, "computer-readable recording media" may also include devices that dynamically store programs for a short period of time, such as communication lines when transmitting programs via networks such as the Internet or communication lines such as telephone lines, and devices that store programs for a certain period of time, such as volatile memory within computer systems that serve as servers or clients in such cases.

[0140] The program may also be a program for implementing some of the functions described above, or may be a so-called differential file (differential program) that can implement the functions described above in combination with a program already stored in the computer system.

[0141] Furthermore, the above program may be one that realizes part of the above-mentioned functions, or may be one that can realize the above-mentioned functions in combination with a program already recorded in a computer system, or may be one that is realized using a programmable logic device such as an FPGA (Field Programmable Gate Array).

[0142] Although the embodiments of the present disclosure have been described above in detail with reference to the drawings, the specific configurations are not limited to these embodiments and include designs within the scope of the present disclosure. Furthermore, the contents described in each embodiment can be used in combination with each other.

[0143] Some or all of the above embodiments can be described as, but are not limited to, the following supplementary notes.

[0144] (Appendix 1) a receiving means for receiving as input the elements of the invention to be described, prior art documents of the invention, and matters described in the specification; a means for performing a process of identifying elements in the prior document that are similar to elements of the invention for which the specification is to be written; an extraction means for extracting descriptions relating to the components identified by the identification process from the prior document; an instruction means for instructing a language model to write the description based on the extracted description; an output means for receiving an output corresponding to the instruction from the language model and outputting the output; An information processing device comprising: (Appendix 2) a setting means for setting a weight of the prior document based on the type of the prior document; the instruction means changes an instruction for the description of the description item based on the extracted description based on the weight. 2. The information processing device according to claim 1. (Appendix 3) the specifying means performs the specifying process a plurality of times with different parameters; 3. The information processing device according to claim 1 or 2. (Appendix 4) When the receiving means receives a plurality of prior documents, the extracting means extracts a description common to the plurality of prior documents. 3. The information processing device according to claim 1 or 2. (Appendix 5) The extraction means excludes descriptions relating to specific data or numerical values ​​from the extraction target. 3. The information processing device according to claim 1 or 2. (Appendix 6) The instruction means An instruction to determine whether the description of the matters based on the extracted description will cause a contradiction with the content of the invention to be described; and issuing an instruction to change the description of the description item based on the extracted description based on the result of the determination. 3. The information processing device according to claim 1 or 2. (Appendix 7) The instruction means An instruction to determine whether the elements of the prior document corresponding to the elements of the invention to be written have a function not present in the elements of the invention to be written; If it is determined that the function is present, an instruction to determine whether adding the function will cause a contradiction in the invention for which the specification is to be written; and if it is determined that no contradiction occurs, issuing an instruction to add the function. 7. The information processing device according to claim 6. (Appendix 8) the instruction means instructs the creation of a statement based on the received output; The identifying means identifies elements in the prior document that are similar to elements described in the written specification, The extraction means extracts descriptions relating to the newly identified components of the prior document, the instruction means instructs the language model to modify the description of the specification based on the newly extracted description. 2. The information processing device according to claim 1. (Appendix 9) The system accepts as input the elements of the invention to be written, the prior art documents of the invention, and the matters described in the specification, performing a process of identifying elements in the prior art document that are similar to elements of the invention that is the subject of the specification; extracting descriptions related to the components identified by the identification process from the prior art documents; instructing a language model to write the description based on the extracted description; receiving and outputting an output for the instruction from the language model; Information processing methods. (Appendix 10) On the computer, A process of receiving as input the elements of the invention to be described, prior art documents of the invention, and matters described in the specification; A process of performing a process of identifying elements in the prior document that are similar to elements of the invention that is the subject of the specification; A process of extracting descriptions related to the components identified by the identification process from the prior literature; a process of instructing a language model to write the description based on the extracted description; receiving and outputting an output from the language model in response to the instruction; A program that executes the following.

[0145] Note that some or all of the configurations described in Supplementary Notes 2 to 8, which are dependent on Supplementary Note 1, may also be dependent on Supplementary Notes 9 and 10 in the same dependent relationship as Supplementary Notes 2 to 8. Furthermore, not limited to Supplementary Notes 1, 9, and 10, but within the scope of each of the above-mentioned embodiments, some or all of the configurations described as Supplements may also be dependent on various hardware, software, various recording means for recording software, or systems. [Explanation of symbols]

[0146] 100 Information processing device 110 Reception 120 Specific section 130 Extraction part 140 Instruction section 150 Output section 160 Setting section

Claims

1. a receiving means for receiving as input the elements of the invention to be described, prior art documents of the invention, and matters described in the specification; a means for performing a process of identifying elements in the prior document that are similar to elements of the invention for which the specification is to be written; an extraction means for extracting descriptions relating to the components identified by the identification process from the prior document, excluding descriptions relating to specific data or numerical values; an instruction means for instructing a language model to write the description based on the extracted description; and an output unit that receives an output corresponding to the instruction from the language model and outputs the output.

2. a setting means for setting a weight of the prior document based on the type of the prior document; The information processing apparatus according to claim 1 , wherein the instruction means changes an instruction for the description of the description item based on the extracted description, based on the weight.

3. the identifying means performs the identifying process using semantic search and keyword search multiple times with different parameters; The information processing device according to claim 1 , wherein the parameter is a weight of the semantic search and the keyword search in the specific process.

4. The information processing apparatus according to claim 1 , wherein when the receiving means receives a plurality of prior documents, the extracting means extracts a description common to the plurality of prior documents.

5. The instruction means An instruction to determine whether the description of the matters based on the extracted description will cause a contradiction with the content of the invention to be described; The information processing device according to claim 1 , further comprising: an instruction to change the description of the description item based on the extracted description, based on a result of the determination.

6. The instruction means An instruction to determine whether the elements of the prior document corresponding to the elements of the invention to be written have a function not present in the elements of the invention to be written; If it is determined that the function is present, an instruction to determine whether adding the function will cause a contradiction in the invention for which the specification is to be written; The information processing apparatus according to claim 5 , further comprising: a command to add the function when it is determined that no contradiction occurs.

7. the instruction means instructs the creation of a statement based on the received output; The identifying means identifies elements in the prior document that are similar to elements described in the written specification, The extraction means extracts descriptions relating to the newly identified components of the prior document, The information processing apparatus according to claim 1 , wherein the instruction means instructs the language model to correct the description of the specification based on the newly extracted description.

8. A computer comprising: The system accepts as input the elements of the invention to be written, the prior art documents of the invention, and the matters described in the specification, performing a process of identifying elements in the prior art document that are similar to elements of the invention that is the subject of the specification; extracting descriptions relating to the components identified by the identification process from the prior art documents, excluding descriptions relating to specific data or numerical values; instructing a language model to write the description based on the extracted description; An information processing method that receives and outputs an output corresponding to the instruction from the language model.

9. On the computer, A process of receiving as input the components of the invention to be written, prior art documents of the invention, and matters described in the specification; A process of performing a process of identifying elements in the prior document that are similar to elements of the invention that is the subject of the specification; a process of extracting descriptions relating to the components identified by the identification process from the prior document, excluding descriptions relating to specific data or numerical values; a process of instructing a language model to write the description based on the extracted description; and receiving and outputting an output corresponding to the instruction from the language model.

Images