Information processing apparatus

Abstract

To stabilize processing using a large-scale language model.SOLUTION: An information processing device is provided with an input means for inputting a text written in a first language, a translation means for translating the text written in the first language by using a large-scale language model, a proofreading means for extracting a part of the translated text, including the part in a prompt together with the corresponding part of the text written in the first language, and proofreading the part by using the large-scale language model, and also with an output means for performing translation and proofreading of the text written in the first language by changing the part to be extracted by the proofreading means and repeating the processing by the proofreading means a plurality of times, and inversely converting the proofread translated text into the text of the first language and outputting the text of the first language.SELECTED DRAWING: Figure 12

Description

[Technical Field]

[0001] The present invention relates to an information processing device. [Background technology]

[0002] The documents to be submitted when filing a patent application are prepared manually by the applicant or their attorney. Specifically, the applicant or attorney checks the invention notification form and the contents of the invention, organizes the prior art (or background art; the same applies below), problems, solutions, effects, etc., and records them in the prescribed format of the patent application documents.

[0003] For example, Patent Document 1 below discloses a patent document preparation device, method, computer program, computer-readable recording medium, server, and system that can reduce the time required to prepare patent documents. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2024-20595 Summary of the Invention [Problem to be solved by the invention]

[0005] Manually preparing documents was extremely time-consuming and labor-intensive, especially when the invention was complex or when there was a large amount of related prior art, and it required a great deal of effort to identify the prior art and problems, and to describe the means to solve them.

[0006] Furthermore, manual document preparation is prone to human errors such as omissions and mistakes. These errors are subject to correction instructions during patent application examination, increasing the burden on applicants and their attorneys.

[0007] Furthermore, because the content of patent application documents varies in format and expression depending on the technical field of the invention, applicants and attorneys are required to have knowledge and experience of the writing styles in each technical field, which creates the problem of relying on the personal skills of applicants and attorneys.

[0008] Another major challenge in preparing patent application documents is the difficulty of accurately grasping the technical content of the invention and explaining it in an easy-to-understand manner to third parties.

[0009] The Patent Act requires that the detailed description of an invention be clear and sufficient to enable a person skilled in the art to easily carry out the invention. However, because inventors have very detailed knowledge of their inventions, they tend to explain their inventions based on the so-called "common sense of a person skilled in the art." As a result, problems have arisen in which important parts of the invention are omitted or implicit understandings are not clearly stated.

[0010] Furthermore, some inventions make extensive use of highly specialized technical terms and mathematical formulas. Explaining such inventions in simple terms that even non-specialists can understand requires highly advanced skills. Applicants and their attorneys must work closely with inventors to provide explanations that are as easy to understand as possible without losing sight of the essence of the invention.

[0011] Furthermore, in patent application documents, it is necessary to clearly demonstrate the differences from the prior art in order to assert the novelty and inventive step of the invention. For this reason, it is necessary to accurately extract the distinctive features of the invention and highlight the points that are not present in the prior art. This task requires in-depth knowledge of the technical field of the invention and extensive research into the prior art.

[0012] In addition, writing the scope of patent claims was one of the most important and difficult parts of patent application documents. The scope of patent claims defines the scope of protection for the invention and is important in future infringement lawsuits, etc. Therefore, extremely advanced writing skills were required to describe the essential parts of the invention without excess or omission, clearly distinguish it from prior art, and ensure an appropriate scope to prevent third parties from circumventing the claims.

[0013] The scope of patent claims is often difficult for the average person to understand, and determining the subject matter of examination and the scope of rights from the scope of patent claims requires skill and is a time-consuming task even for experienced practitioners.

[0014] As mentioned above, there are various difficulties in preparing patent application documents, such as accurately grasping the technical content of the invention, explaining it in a way that is easy for third parties to understand, clarifying the differences from prior art, setting the appropriate scope of patent claims, etc. There has been a demand for a system that can overcome these difficulties and efficiently prepare high-quality patent application documents.

[0015] Furthermore, language issues have been a major obstacle when preparing patent application documents for multiple countries. Based on each country's patent laws and examination standards, specifications and claims must be written in the language of that country, but applicants and attorneys are not necessarily fluent in all languages. This requires the use of translators, which adds cost and time.

[0016] Moreover, simply translating the languages ​​was not enough; it was also necessary to adjust the content of the specifications and claims to conform to each country's patent laws and examination standards. For example, Japanese patent law requires that the detailed description of the invention describe the effects of the invention, but the United States does not. Understanding the differences in each country's legal system and optimizing the content to match them was a task that required highly specialized knowledge and experience.

[0017] Furthermore, patent application procedures vary from country to country, and preparing and submitting documents to meet formal requirements can be cumbersome. For example, some countries require the signature of the inventor on certain documents, but Japan does not have such a requirement. As such, preparing the necessary documents to comply with each country's procedures places a burden on applicants and their attorneys.

[0018] As mentioned above, when filing patent applications in multiple countries, various difficulties arise, such as language issues, differences in each country's legal system, complicated procedures, etc. A system that would resolve these difficulties and enable efficient filing of patent applications in multiple countries was needed.

[0019] The complexity and difficulty of document processing is not limited to patent documents (Japanese and foreign languages), but is a common issue for many other types of documents. Difficulties exist in all aspects of document creation, translation, revision, and checking. Even when processing using large-scale language models, it is difficult to produce stable output. [Means for solving the problem]

[0020] The present invention has been made to solve the above problems, and provides an information processing device as set forth in the appended claims. [Brief explanation of the drawings]

[0021] [Figure 1] 1 is a block diagram showing a system configuration according to an embodiment of the present invention; [Figure 2] FIG. 10 is a block diagram showing a system configuration according to another embodiment of the present invention. [Figure 3] 1 is a block diagram showing a system configuration in an embodiment of the program of the present invention; [Figure 4] 3 is a flowchart showing a processing flow in the present embodiment. [Figure 5]FIG. 10 is a diagram showing the results of drawing based on code written in the Dot language output from a claim by a computer program according to this embodiment. [Figure 6] FIG. 10 is a diagram showing the results of drawing based on code written in the Mermaid language output from a claim by a computer program according to this embodiment. [Figure 7] FIG. 10 is a diagram showing the results of drawing based on code written in the Mermaid language output from a claim by a computer program according to this embodiment. [Figure 8] This is a continuation of Figure 7. [Figure 9] This is a continuation of Figure 8. [Figure 10] FIG. 10 is a diagram showing the results of drawing based on code written in the Dot language output from a claim by a computer program according to this embodiment. [Figure 11] This is a continuation of Figure 10. [Figure 12] FIG. 10 is a block diagram showing a system configuration according to a third embodiment of the present invention. [Figure 13] FIG. 10 is a diagram illustrating the operation concept of a system according to a third embodiment of the present invention. [Figure 14] 10 is a flowchart showing a process according to a third embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] The information processing device of the present invention performs the following processing using a large-scale language model based on input contents from data that may contain inventions, such as invention notifications and input forms.

[0023] Extraction or creation of prior art (or background art; the same applies below) Extracting or generating issues Extraction or generation of solutions to problems Extracting or creating the effects of an invention Extracting or generating invention titles Extracting or creating a form for implementing the invention Extract or generate drawings or drawing generation code Extracting or generating drawing descriptions Claim extraction or generation Extracting or generating abstracts If extraction is not possible, the necessary information is generated from training data, databases, or the Internet. It is not necessary to extract or generate some of the multiple elements (items) mentioned above. Finally, the generated elements are compiled and the patent application documents are output. In other words, the documents required for patent applications are automatically generated using a large-scale language model from the contents of invention notification forms, input forms, etc. This allows applicants to accurately explain the content of their inventions and significantly reduces the time and effort required to properly write patent application documents, including patent claims.

[0024] Large-scale language models (LLMs) can generate natural-sounding sentences that are close to human-like by learning from large amounts of literature data. Applying this function to the creation of patent application documents can be expected to bring the following benefits: 1. Accurately understanding and summarizing the content of the invention By inputting a draft of an invention notification or specification into LLM, the important points of the invention can be automatically extracted and concisely summarized, allowing for the detection of insufficient explanations by the inventor or redundant descriptions, and for appropriate corrections to be made. 2. Prior art search and comparison LLM allows users to perform a wide search of prior art documents related to the technical field of an invention and automatically analyze the differences between the invention and the prior art. This allows users to objectively evaluate the novelty and inventive step of an invention and efficiently create descriptions that clarify the differences from the prior art. 3. Optimization to suit each country's legal system and procedures By having LLM learn data on each country's patent laws, examination standards, procedures, etc., it is possible to automatically optimize the content of an invention to suit the requirements of each country. This enables applicants and agents to efficiently prepare high-quality patent application documents, even if they do not have in-depth knowledge of each country's legal system. 4. Multilingual support LLM is able to perform highly accurate translations by learning from multilingual text data. Utilizing this function, patent application documents written in Japanese can be automatically translated into the languages ​​of each country. This significantly reduces translation costs and time. It is also possible to automatically submit applications to patent offices in each country via the Internet or communication lines, and automatically send application documents to lawyers and patent attorneys. 5. Automatic Claim Generation By inputting a detailed description of an invention into LLM, the content can be appropriately summarized and the claims can be automatically generated. In doing so, by utilizing the prior art and knowledge of patent laws of various countries that LLM has learned, it is possible to describe the essential parts of the invention without excess or omission, while ensuring an appropriate scope of rights.

[0025] As described above, utilizing LLM can efficiently solve various issues in preparing patent application documents. Automation through LLM can significantly reduce the burden on applicants and attorneys while enabling the rapid preparation of high-quality patent application documents.

[0026] Further possibilities for using LLM to support the preparation of patent application documents include the following: 1. Automatic generation of drawings By inputting the components of an invention and their relationships into LLM, drawings can be automatically generated. By applying the various drawing patterns that LLM has learned, drawings that accurately express the content of the invention can be efficiently created. This significantly reduces the time and effort that applicants and attorneys spend on creating drawings. 2. Support for communication with examiners LLM can assist in preparing responses to office actions from examiners. LLM analyzes the reasons for refusal and the content of the invention, presenting appropriate counterarguments, allowing applicants and their attorneys to develop effective response strategies. Additionally, by using LLM to conduct online interviews with examiners, applicants can accurately grasp the content of the invention and the examiner's concerns, enabling smooth communication. 3. Patent portfolio optimization By utilizing LLM, applicants can automatically analyze their patent portfolios and support their optimization. Specifically, LLM analyzes the content of each patent within a patent portfolio and detects overlapping inventions and unnecessary patents, thereby streamlining the portfolio. LLM also analyzes competitors' patents, visualizing their relationship with the applicant's own patent portfolio and supporting strategic patent applications and enforcement. 4. Valuation of inventions By utilizing LLM, it is possible to automatically evaluate the technical value and marketability of an invention. LLM analyzes the content of the invention and takes into account relevant market and technological trends, making it possible to calculate the objective value of the invention. This allows applicants to properly understand the value of their inventions and make effective decisions about patent applications and commercialization. 5. Utilizing patent information By utilizing LLM, it is possible to efficiently analyze large amounts of patent information and obtain useful insights. For example, by analyzing patent application trends in a specific technology field, LLM can help understand the direction of research and development in that field and the trends of competitors. In addition, by combining and analyzing patent information and paper information, LLM can also help incorporate the latest research trends in academia into patent applications.

[0027] In this way, utilizing the LLM not only supports the preparation of patent application documents, but also improves the efficiency and sophistication of patent work in general. The possibilities offered by the LLM are extremely broad, and it is expected to become an indispensable tool in patent work in the future.

[0028] In addition, by linking the patent management database with the LLM, the following benefits are expected: 1. Patent portfolio visualization and analysis By analyzing your own and other companies' patent information stored in a patent management database using LLM, you can visualize the overall picture of your patent portfolio and perform strategic analysis. For example, by analyzing the distribution of your own patents in a certain technical field and the relationship with other companies' patents, you can identify your company's strengths, weaknesses, and white spaces. In addition, by analyzing changes in your patent portfolio over time using LLM, you can understand technological trends and the actions of your competitors.

[0029] Patent portfolios can also be visualized. In this case, LLM analyzes information such as patent information, designs, and trademarks from multiple patent documents and outputs code for creating drawings. The diagram is created based on the code and accepts modifications. Drawings can also be generated directly rather than outputting code (this type of drawing generation and modification method can be used in the explanation of drawing creation and visualization below). 2. Advanced patent information search LLM enables advanced searches of patent management databases. While conventional keyword-based searches have difficulty extracting all relevant patents, LLM enables more flexible and accurate searches. For example, by understanding the content of patent documents and taking synonyms and related terms into account, LLM can extract related patents that are often overlooked using keywords. Furthermore, by analyzing the semantic similarity of patent documents, LLM can efficiently discover patents that differ little from prior art or patents with a high risk of invalidity. You can also input conditions and use LLM to generate patent search queries. 3. Patent Information Summarization and Visualization LLM summarizes and visualizes the large amount of patent information stored in patent management databases, enabling efficient information understanding and sharing. For example, LLM automatically extracts the key points of each patent document and generates a concise summary, allowing users to quickly grasp the overall picture of the patent information. LLM also graphically visualizes patent information, making it possible to intuitively understand the relationships between complex patent information. This enables effective use of patent information and smooth information sharing among stakeholders. 4. Multilingualization of patent information By processing multilingual patent information stored in patent management databases with LLM, it becomes possible to utilize patent information across language barriers. For example, LLM can automatically translate foreign language patent documents, enabling searches and database registration in one's native language, facilitating the collection and analysis of global patent information. Furthermore, by unifying the analysis of multilingual patent information with LLM, it becomes possible to compare patent information across languages ​​and understand global technology trends. 5. Patent information prediction and recommendation By analyzing patent information in patent management databases, LLM can predict and recommend future patent applications and technological development directions. For example, by analyzing time-series data on patent applications in a specific technology field, LLM can predict development trends in that field and reflect them in the company's research and development strategy. In addition, by analyzing the company's own patent portfolio and external patent information, LLM can recommend promising patents that will contribute to the company's patent strategy.

[0030] As described above, linking patent management databases and LLMs enables strategic analysis and utilization of patent information. By combining the advanced natural language processing capabilities of LLMs with the abundant data in patent management databases, it is now possible to analyze patent information and gain insights that were previously difficult to obtain. This collaboration is expected to enhance corporate intellectual property strategies and accelerate innovation.

[0031] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0032] 1 is a functional block diagram of an information processing device 100 according to one embodiment of the present invention. The information processing device 100 is a device that accepts input from an invention notification form or an invention input form, and creates documents for a patent application.

[0033] The information processing device 100 includes an input unit 110, a first generation unit 120, a second generation unit 130, a third generation unit 140, a fourth generation unit 150, a fifth generation unit 160, a sixth generation unit 170, a seventh generation unit 180, an eighth generation unit 190, a ninth generation unit 190, and an output unit 200.

[0034] The input unit 110 accepts input from an invention notification form or a form for inputting an invention. The input unit 110 is configured to include input devices such as a keyboard, mouse, touch panel, etc. Data may be input from the Internet, a database, HTML data, text data, drawings, or multimodal data that is a combination of these. Data may be input automatically instead of manually. Data may be input from a repository or by automatic navigation.

[0035] The first generation unit 120 extracts prior art from the content of the invention notification or input received by the input unit 110 using a large-scale language model. Here, a large-scale language model is a language model constructed by machine learning a large amount of literature data, and is used for various tasks in natural language processing. The first generation unit 120 inputs the content of the invention notification or the like into the large-scale language model and extracts descriptions related to prior art from it. If prior art cannot be extracted, the first generation unit 120 generates prior art related to the technical field of the invention from training data, a database, or information sources on the Internet. In the generation, highly accurate information can be generated by using the data received by the input unit 110 (the same applies below).

[0036] The second generation unit 130 uses a large-scale language model to extract problems from the content of the invention notification or input received by the input unit 110. Specifically, it extracts descriptions of the problems that the invention is intended to solve and problems with the prior art from the content of the invention notification, etc. If a problem cannot be extracted, the second generation unit 130 generates general problems in the technical field of the invention from training data, a database, or information sources on the Internet.

[0037] The third generation unit 140 uses a large-scale language model to extract a means for solving the problem from the content of the invention notification or input received by the input unit 110. Specifically, it extracts a description of how to solve the problem from the content of the invention notification, etc. If a means for solving the problem cannot be extracted, the third generation unit 140 generates a general means for solving the problem from training data, a database, or information sources on the Internet.

[0038] The means for solving a problem generally corresponds to the scope of a patent claim. If there are multiple means for solving a problem, they will form multiple independent claims. If there are multiple means for solving a problem, from a generic concept to a specific concept, they will form an independent claim and multiple independent claims. Therefore, by extracting and generating the means for solving a problem, it is possible to convert it into the scope of a patent claim. To ensure consistency in patent documents, it is desirable to align the means for solving a problem with the scope of the patent claim in this way. Alternatively, a separate generation unit may be provided and the scope of the patent claim may be extracted using a large-scale language model from the contents of the invention notification or input received by the input unit 110. Specifically, the description of the scope of the patent claim is extracted from the contents of the invention notification, etc. If extraction is not possible, the generation unit generates the scope of the patent claim from training data, a database, or an online source. Note that the scope of the patent claim may be extracted and generated without extracting and generating the means for solving the problem. Alternatively, the scope of the patent claim may be extracted and generated first, and then the means for solving the problem may be created based on that.

[0039] The fourth generation unit 150 uses a large-scale language model to extract the effects of the invention from the content of the invention notification or input received by the input unit 110. Specifically, it extracts descriptions of the advantageous effects brought about by the invention from the content of the invention notification, etc. If the effects of the invention cannot be extracted, the fourth generation unit 150 generates general effects in the technical field of the invention from training data, a database, or information sources on the Internet.

[0040] The fifth generation unit 160 uses a large-scale language model to extract the title of the invention from the content of the invention notification form or input received by the input unit 110. Specifically, it extracts a name that succinctly expresses the content of the invention from the content of the invention notification form, etc. If the title of the invention cannot be extracted, the fifth generation unit 160 generates a name that is suitable for the content of the invention from training data, a database, or information sources on the Internet.

[0041] The sixth generation unit 170 extracts drawings from the invention notification form or input content received by the input unit 110. Specifically, it reads the configuration of the drawings from the drawings attached to the invention notification form or the like, or from the content of the invention. If the drawings cannot be extracted, the sixth generation unit 170 generates drawings suitable for expressing the content of the invention, or code for generating the drawings, from learning data, a database, or information sources on the Internet.

[0042] The seventh generation unit 180 uses a large-scale language model to extract a description of the drawings from the content of the invention notification or input received by the input unit 110. Specifically, it extracts sentences explaining what each drawing and each element of the drawing represents from the content of the invention notification or the drawings. If it is not possible to extract a description of the drawings, the seventh generation unit 180 generates sentences suitable for explaining the content of the drawings from training data, a database, or information sources on the Internet.

[0043] The eighth generation unit 190 uses a large-scale language model to extract an abstract from the content of the invention notification or input received by the input unit 110. Specifically, it extracts a summary that succinctly summarizes the outline of the invention from the content of the invention notification, etc. If an abstract cannot be extracted, the eighth generation unit 190 generates a sentence suitable for summarizing the content of the invention from training data, a database, or information sources on the Internet.

[0044] The ninth generation unit 195 uses a large-scale language model to extract a mode for implementing the invention (and / or an example) from the content of the invention notification or input received by the input unit 110. Specifically, it extracts a mode for implementing the invention (or the best mode for implementing the invention) from the content of the invention notification, etc. If extraction is not possible, the ninth generation unit 195 generates sentences suitable for the mode for implementing the invention (and / or an example) from training data, a database, or information sources on the Internet.

[0045] The output unit 200 acquires the various documents and drawings generated by the first generation unit 120 to the ninth generation unit 195, and integrates them according to a predetermined format to generate and output documents for a patent application. Ultimately, five documents are generated: a "patent application (application)," a "specification," a "claims," ​​an "abstract," and "drawings (if necessary)." The output unit 200 includes an output device such as a display or a printer. The communication unit may transmit information as an email attachment or the like. The output unit 200 may be software for filing a patent application, and a patent application may be filed with the patent offices of various countries.

[0046] The information processing device 100 described above can automatically extract or generate various elements necessary for a patent application from the contents of an invention notification document, etc., and integrate them to efficiently prepare patent application documents. Furthermore, the use of a large-scale language model makes it possible to generate high-quality text and drawings. This makes it possible to provide high-quality patent application documents while significantly reducing the time and effort required to prepare them.

[0047] As described above, the first generation unit 120 extracts prior art from the input content of an invention notification form or an invention input form using a large-scale language model, and if it cannot be extracted, generates it from training data, a database, or the Internet. Below, modified examples, other forms, and other possibilities of the first generation unit 120 will be described in detail. 1. Use of multiple large-scale language models The first generation unit 120 can use not only a single large-scale language model, but also a combination of multiple large-scale language models (the same applies below). For example, by combining a large-scale language model specialized in the technical field of the invention with a large-scale language model that is excellent for general sentence generation, it becomes possible to generate more accurate and natural descriptions of prior art. 2. Utilizing conventional technology databases The first generation unit 120 can also utilize a pre-constructed prior art database when extracting prior art. This database comprehensively records representative prior art in each technical field, making it possible to efficiently search for prior art related to the input content. This can improve the accuracy of generation using a large-scale language model while also speeding up processing. 3. Narrowing down existing technologies through user interaction The first generation unit 120 can also present the user with candidates for prior art generated by the large-scale language model and narrow down the prior art through interaction with the user. For example, by presenting the user with multiple generated candidates for prior art and allowing the user to select the most relevant prior art, a more appropriate prior art can be identified. 4. Extraction of prior art from other patent documents The first generation unit 120 not only extracts prior art from the input content, but also extracts prior art from other related patent documents. Specifically, it searches a patent database using the input content as a query and extracts prior art from the patent documents obtained as search results. This makes it possible to supplement prior art that cannot be covered by the input content alone. 5. Automatic Summarization of Prior Art The first generator 120 can also automatically summarize the extracted or generated prior art description. It uses a large-scale language model to extract important points from the prior art description and generate a concise summary. This allows the prior art description in application documents to be concise and accurate. 6. Multilingual support The first generation unit 120 can also handle invention notifications or inputs in multiple languages. By determining the language of the input content and using a large-scale language model or prior art database corresponding to that language, it is possible to extract and generate prior art in multiple languages. This allows for more efficient description of prior art in global patent applications.

[0048] The above describes modified examples, other forms, and other possibilities for the first generation unit 120. By combining various technologies with a focus on large-scale language models, the first generation unit 120 can achieve more advanced and efficient extraction and generation of prior art. This is expected to improve the quality of descriptions of prior art in patent application documents and significantly improve the efficiency of the creation process.

[0049] The second generation unit 130 extracts a problem from the input content of the invention notification form or the form for inputting an invention using a large-scale language model, and if it cannot be extracted, it generates it from training data, a database, or the Internet. Below, we will explain in detail the modified examples, other forms, and other possibilities of the second generation unit 130. 1. Creating a task hierarchy The second generation unit 130 can also represent the problems extracted or generated from the input content in a hierarchical structure. For example, it uses a large-scale language model to analyze causal relationships and inclusion relationships between problems and represent them in a hierarchical structure. This makes it possible to grasp the overall picture of the problem that the invention is trying to solve from a bird's-eye view. 2. Automatic issue classification The second generation unit 130 can also automatically classify the extracted or generated problems according to a predefined classification system. For example, a problem classification system for each technical field is prepared, and the content of the problems is analyzed using a large-scale language model to classify each problem into an appropriate category. This makes it possible to organize the problems of the invention in accordance with the context of the technical field. 3. Use of the issue database The second generation unit 130 can also utilize a pre-constructed task database when extracting tasks. This database contains typical tasks in each technical field, making it possible to efficiently search for tasks related to the input content. This allows for faster processing while improving the accuracy of task generation using a large-scale language model. 4. Time series analysis of the issue The second generation unit 130 can also analyze the extracted or generated problems in chronological order. Specifically, it analyzes related patent documents and academic literature in chronological order to understand the evolution of the problems and technological trends. This makes it possible to clarify the positioning of the problems that the invention aims to solve in the context of technological developments. 5. Narrowing down issues through user interaction The second generation unit 130 can also present the user with candidate tasks generated by a large-scale language model and narrow down the tasks through interaction with the user. For example, by presenting the user with multiple candidate tasks generated and allowing the user to select the most relevant task, a more appropriate task can be identified. 6. Automatic issue summarization The second generation unit 130 can also automatically summarize the extracted or generated problem description. It uses a large-scale language model to extract important points from the problem description and generate a concise summary. This makes it possible to make the problem description in application documents concise and accurate. 7. Multilingual support The second generation unit 130 can also handle invention notifications or inputs in multiple languages. By determining the language of the input content and using a large-scale language model or problem database corresponding to that language, it becomes possible to extract and generate problems in multiple languages. This allows for the efficient description of problems in global patent applications.

[0050] The above describes modified examples, other forms, and other possibilities for the second generation unit 130. The second generation unit 130 combines various technologies while focusing on large-scale language models, thereby enabling more advanced and efficient problem extraction and generation. This is expected to improve the quality of problem descriptions in patent application documents and significantly streamline the creation process.

[0051] The third generation unit 140 extracts a means for solving the problem from the input content of the invention notification form or the form for inputting the invention using a large-scale language model, and if it cannot be extracted, it generates it from training data, a database, or the Internet. Below, modified examples, other forms, and other possibilities of the third generation unit 140 will be described in detail. 1. Generating and ranking multiple solutions The third generation unit 140 can not only generate a single solution, but also generate multiple solutions and rank them. Using a large-scale language model, it generates multiple solutions from the input content and scores them in terms of effectiveness, feasibility, etc. This makes it possible to present multiple options to the inventor or applicant and allow them to select the most appropriate solution. 2. Use of solution database The third generation unit 140 can also utilize a pre-constructed solution database when extracting solutions. This database contains typical solutions in each technical field, making it possible to efficiently search for solutions related to the input content. This can improve the accuracy of generating solutions using a large-scale language model while also speeding up processing. 3. Combinatorial optimization of solutions The third generation unit 140 can also generate more effective solutions by combining multiple solutions. It uses a large-scale language model to analyze the characteristics of each solution and optimize their combination. This makes it possible to address problems that are difficult to achieve with a single solution by combining multiple solutions. 4. Simulation and validation of solutions The third generation unit 140 can also verify the effectiveness of the generated solution by simulation and validation. Specifically, the solution is modeled and its effectiveness is quantitatively evaluated by performing a simulation on a computer. In addition, validation using actual data can be used to empirically confirm the effectiveness of the solution. 5. Improve the solution through user interaction The third generation unit 140 can present the generated solution to the user and improve the solution based on feedback from the user. For example, the third generation unit 140 can have the user evaluate the advantages and disadvantages of the solution and modify the solution based on the evaluation. This allows the inventor's and applicant's knowledge to be fed back into the generation of the solution, resulting in a more realistic and effective solution. 6. Automatic Solution Summarization The third generation unit 140 can also automatically summarize the generated solution description. It uses a large-scale language model to extract important points from the solution description and generate a concise summary. This makes it possible to make the solution description in the application documents concise and accurate. 7. Multilingual support The third generation unit 140 can also handle invention notifications or inputs in multiple languages. By determining the language of the input content and using a large-scale language model or solution database corresponding to that language, it becomes possible to extract and generate solutions in multiple languages. This allows for the efficient description of solutions in global patent applications.

[0052] The above describes modified examples, other forms, and other possibilities of the third generation unit 140. By combining various technologies with a focus on large-scale language models, the third generation unit 140 can realize more advanced and efficient extraction and generation of solutions. This is expected to improve the quality of the descriptions of solutions in patent application documents and significantly improve the efficiency of the creation process.

[0053] The problem to be solved can also be used as a claim in a patent, meaning that claims can be automatically generated from a given document.

[0054] The fourth generation unit 150 extracts the effect of the invention from the input content of the invention notification form or the form for inputting the invention using a large-scale language model, and if it cannot be extracted, it generates it from training data, a database, or the Internet. Modifications, other forms, and other possibilities of the fourth generation unit 150 will be described in detail below. 1. Generating and ranking multiple effects The fourth generation unit 150 can not only generate a single effect, but also generate multiple effects and rank them. Using a large-scale language model, it generates multiple effects from the input content and scores them in terms of importance, uniqueness, etc. This allows inventors and applicants to select the most important effect from multiple options. 2. Utilizing the effects database The fourth generation unit 150 can also utilize a pre-constructed effect database when extracting effects. This database contains typical effects in each technical field, making it possible to efficiently search for effects related to the input content. This allows for faster processing while improving the accuracy of effect generation using a large-scale language model. 3. Quantitative evaluation of effects The fourth generation unit 150 can also quantitatively evaluate the generated effect. Specifically, the effect is converted into metrics and expressed numerically. For example, if the effect is performance improvement, the improvement rate is calculated and expressed numerically. This makes it possible to objectively and quantitatively show the magnitude of the effect, allowing the value of the invention to be more persuasively asserted. 4. Simulation and validation of effects The fourth generation unit 150 can also verify the validity of the generated effect through simulation and validation. Specifically, the effect is modeled and simulated on a computer to quantitatively evaluate the effect. In addition, validation using actual data can be used to empirically confirm the effectiveness of the effect. 5. Improved effects through user interaction The fourth generation unit 150 can present the generated effects to the user and improve the effects based on user feedback. For example, the fourth generation unit 150 can have the user evaluate whether the effect is sufficient or insufficient and then modify the effect accordingly. This allows the inventor's or applicant's knowledge to be fed back into the generation of the effects, resulting in more realistic and persuasive effects. 6. Automatic Effects Summary The fourth generation unit 150 can also automatically summarize the generated effect description. It uses a large-scale language model to extract important points from the effect description and generate a concise summary. This makes it possible to make the effect description in application documents concise and accurate. 7. Multilingual support The fourth generation unit 150 can also handle invention notifications or inputs in multiple languages. By determining the language of the input content and using a large-scale language model or effect database corresponding to that language, it becomes possible to extract and generate effects in multiple languages. This allows for the efficient description of effects in global patent applications.

[0055] The above describes the modified examples, other forms, and other possibilities of the fourth generation unit 150. The fourth generation unit 150 is able to achieve more advanced and efficient extraction and generation of effects by combining various technologies while focusing on a large-scale language model. This is expected to improve the quality of the description of effects in patent application documents and greatly improve the efficiency of the creation work.

[0056] The fifth generation unit 160 extracts the name of the invention from the input content of the invention notification form or the form for inputting the invention using a large-scale language model, and if the name cannot be extracted, generates it from training data, a database, or the Internet. Modifications, other forms, and other possibilities of the fifth generation unit 160 will be described in detail below. 1. Generating and ranking multiple name candidates The fifth generation unit 160 can not only generate a single name, but also generate and rank multiple name candidates. Using a large-scale language model, it generates multiple name candidates from the input content and scores them in terms of appropriateness, attractiveness, and other factors. This allows inventors and applicants to choose the most suitable name from multiple options. 2. Linguistic optimization of names The fifth generation unit 160 can also linguistically optimize the generated name. Specifically, it evaluates the feel, readability, and memorability of the words in the name, and replaces them with better words or adjusts the word order. This makes it possible to generate a memorable name that accurately expresses the content of the invention. 3. Evaluation of the uniqueness of the name The fifth generation unit 160 can also evaluate the uniqueness of the generated name. Specifically, it checks the generated name against databases of existing patents and trademarks to see if there are any identical or similar names. This prevents the risk of infringement and ensures the uniqueness of the name of the invention. 4. Evaluation of image associations of names The fifth generation unit 160 can also evaluate what kind of image the generated name evokes. Specifically, it uses a large-scale language model to generate words that evoke an image from the name, and evaluates whether the image is appropriate for the content of the invention. This makes it possible to generate a name that can intuitively convey the content of the invention. 5. Refine names through user interaction The fifth generation unit 160 can present the generated name to the user and improve the name based on feedback from the user. For example, the fifth generation unit 160 can have the user rate the impression of the name and modify the name based on the user's feedback. This allows the preferences of the inventor or applicant to be fed back into the name generation, resulting in a more desirable name. 6. Name Localization The fifth generation unit 160 can also localize the generated name into other languages. Specifically, it replaces words that make up the name with words in other languages ​​and reconstructs the name according to naming rules for each language. This makes it possible to efficiently generate names suitable for each country's language in global patent applications. 7. Visual Representation of Name The fifth generation unit 160 can also visually represent the generated name. Specifically, it converts the name into a logo design or an image. This makes it possible to convey the image of the name more intuitively, thereby increasing the persuasiveness of the patent application documents.

[0057] The above describes the modified examples, other forms, and other possibilities of the fifth generation unit 160. By combining various technologies with a focus on large-scale language models, the fifth generation unit 160 can achieve more advanced and efficient extraction and generation of invention titles. This is expected to improve the quality of invention titles in patent application documents and greatly improve the efficiency of the creation process.

[0058] The sixth generation unit 170 extracts drawings from the input contents of the invention notification form or the form for inputting the invention, and if extraction is not possible, generates drawings or code for generating drawings from learning data, a database, or the Internet. Modifications, other forms, and other possibilities of the sixth generation unit 170 will be described in detail below. 1. Generating and ranking multiple drawing candidates The sixth generator 170 can generate multiple drawing candidates and rank them, rather than just generating a single drawing. Using a large-scale language model, it generates multiple drawing candidates from the input content and scores them based on clarity, aesthetics, and other factors. This allows inventors and applicants to choose the most suitable drawing from multiple options. 2. Automatic layout adjustment of drawings The sixth generation unit 170 can also automatically adjust the layout of the generated drawing. Specifically, it optimizes the placement, size, and margins of elements within the drawing to create a beautiful, easy-to-read drawing. This saves inventors and applicants the trouble of manually adjusting the drawing. 3. Automatic drawing annotation The sixth generation unit 170 can also automatically annotate the generated drawing. Specifically, it numbers and labels each element in the drawing and provides an explanation for them outside the drawing. This helps to understand the drawing and clarifies the correspondence between the drawing and the explanation. 4. Generate 3D drawings The sixth generation unit 170 can generate not only 2D drawings but also 3D drawings. Specifically, it models the 3D structure of the invention from the input content and outputs it as a 3D drawing. This makes it possible to communicate the structure of the invention more intuitively, thereby increasing the persuasiveness of patent application documents. 5. Generating animated drawings The sixth generating unit 170 can generate not only still images but also moving images. Specifically, it animates the movements and changes of the invention in chronological order and outputs them as moving images. This makes it possible to effectively communicate the dynamic aspects of the invention. 6. Automatic drawing optimization The sixth generation unit 170 can also automatically optimize the generated drawings. Specifically, it optimizes the resolution, color depth, file format, etc. of the drawings to meet the requirements of the patent application. This makes it possible to minimize the file size while ensuring the quality of the drawings. 7. User interaction to improve the drawing The sixth generation unit 170 can present the generated drawings to the user and improve the drawings based on user feedback. For example, the sixth generation unit 170 can have the user evaluate the clarity of the drawings and revise the drawings accordingly. This allows the perspectives of the inventor or applicant to be fed back into the generation of the drawings, resulting in better drawings.

[0059] The above describes the variations, other forms, and other possibilities of the sixth generation unit 170. The sixth generation unit 170 combines various technologies while focusing on a large-scale language model, thereby enabling more advanced and efficient extraction and generation of drawings. This is expected to improve the quality of drawings in patent application documents and significantly streamline the creation process.

[0060] The seventh generation unit 180 extracts a description of the drawings from the input content of the invention notification form or the form for inputting the invention using a large-scale language model, and if extraction is not possible, generates it from training data, a database, or the Internet. Modifications, other forms, and other possibilities of the seventh generation unit 180 will be described in detail below. 1. Automatically assigning descriptions to each element of a drawing The seventh generation unit 180 can also automatically provide individual descriptions for each element in the drawing. Specifically, it uses a large-scale language model to recognize the names, functions, and interrelationships of each element in the drawing and generates them as descriptions. This allows for a more detailed understanding of the drawing. 2. Adjusting the level of detail in the drawing descriptions The seventh generating unit 180 can also adjust the level of detail of the description of the drawing to be generated. Specifically, the user can specify the level of detail of the description, and the seventh generating unit 180 can generate descriptions ranging from simpler to more detailed. This makes it possible to provide descriptions with an appropriate level of detail depending on the purpose and target audience. 3. Multilingual drawing descriptions The seventh generation unit 180 can also translate the generated description of the drawing into multiple languages. Specifically, it generates the description of the drawing in multiple languages ​​by automatically translating the description into other languages. This makes it possible to efficiently create description of the drawing in the languages ​​of various countries in global patent applications. 4. Audio description of drawings The seventh generation unit 180 can also convert the generated explanation of the drawing into audio. Specifically, the explanation is converted into audio using text-to-speech technology. This makes it possible to auditorily assist in understanding the visual drawing, making it easier to understand the drawing. 5. Quiz generation for drawing descriptions The seventh generation unit 180 can also automatically generate a quiz about the content of the drawing using the generated description of the drawing. Specifically, it extracts important points from the description and converts them into questions and answer options. This allows the content of the drawing to be confirmed. 6. Summary of Drawing Description The seventh generation unit 180 can also summarize the generated description of the drawing. Specifically, it extracts important keywords and phrases from the description and combines them to generate a summary. This allows the content of the drawing to be concisely understood. 7. Improved drawing descriptions through user interaction The seventh generation unit 180 can present the generated description of the drawing to the user and improve the description based on feedback from the user. For example, the seventh generation unit 180 can have the user evaluate the clarity of the description and revise the description based on the evaluation. This allows the viewpoints of the inventor or applicant to be fed back into the generation of the description of the drawing, resulting in a better description.

[0061] The above describes modified examples, other forms, and other possibilities for the seventh generation unit 180. The seventh generation unit 180 combines various technologies while focusing on a large-scale language model, thereby enabling more advanced and efficient extraction and generation of drawing descriptions. This is expected to improve the quality of drawing descriptions in patent application documents and significantly streamline the creation process.

[0062] The eighth generation unit 190 extracts an abstract from the input contents of an invention notification form or an invention input form using a large-scale language model, and if an abstract cannot be extracted, generates one from training data, a database, or the Internet. Modifications, other forms, and other possibilities of the eighth generation unit 190 will be described in detail below. 1. Adjusting the length of the abstract The eighth generation unit 190 can also adjust the length of the abstract to be generated. Specifically, the user can specify the length of the abstract, and the eighth generation unit 190 can generate abstracts ranging from shorter to longer. This makes it possible to provide abstracts of an appropriate length depending on the purpose and requirements. 2. Diversification of Abstract Formats The eighth generation unit 190 can also generate abstracts in various formats, such as a bulleted list format, an illustrated summary, a story-style summary, etc. This makes it possible to communicate the contents of the invention in a more understandable and impressive way. 3. Adjusting the tone and voice of your abstract The eighth generation unit 190 can also adjust the tone and speaking style of the abstract it generates. Specifically, the user can specify the tone of the abstract, and the unit can generate abstracts in a range of speaking styles from more formal to more gentle. This makes it possible to provide an abstract with an appropriate tone depending on the target audience and situation. 4. Multimedia Abstracts The eighth generation unit 190 can generate a multimedia abstract that includes not only text but also images and videos. Specifically, it automatically selects or generates images and videos related to the content of the invention and embeds them in the abstract. This makes it possible to communicate the content of the invention in a more intuitive and impactful way. 5. Multilingual Abstracts The eighth generation unit 190 can also translate the generated abstract into multiple languages. Specifically, it generates abstracts in multiple languages ​​by automatically translating the abstract into other languages. This makes it possible to efficiently create abstracts that are compatible with the languages ​​of various countries in global patent applications. 6. Personalize your abstract The eighth generation unit 190 can also personalize the abstract to suit the preferences and writing style of the inventor or applicant. Specifically, it learns the unique expression style of the inventor or applicant from their past abstracts and writings, and generates an abstract that reflects that. This makes it possible to provide an abstract that is more natural and familiar to the inventor or applicant. 7. User interaction to improve the abstract The eighth generation unit 190 can present the generated abstract to the user and improve the abstract based on feedback from the user. For example, the unit 190 can have the user evaluate the clarity and appropriateness of the abstract and revise the abstract based on the evaluation. This allows the viewpoints of the inventor and applicant to be fed back into the generation of the abstract, resulting in a better abstract.

[0063] The above describes the variations, other forms, and other possibilities of the eighth generation unit 190. The eighth generation unit 190 combines various technologies while focusing on a large-scale language model, thereby enabling more advanced and efficient abstract extraction and generation. This is expected to improve the quality of abstracts in patent application documents and greatly improve the efficiency of the creation process.

[0064] The ninth generation unit 195 extracts a form (and / or embodiment) for implementing the invention from the input content of an invention notification form or a form for entering an invention using a large-scale language model, and if extraction is not possible, generates it from training data, a database, or the Internet. Modifications, other forms, and other possibilities of the ninth generation unit 195 will be described in detail below. 1. Adjusting the Length of the Detailed Description (and / or Examples) The ninth generation unit 195 can also adjust the length of the mode for carrying out the invention (and / or example) to be generated. Specifically, the user can specify the length of the mode for carrying out the invention (and / or example), and the ninth generation unit 195 generates a mode for carrying out the invention (and / or example) ranging from a shorter mode for carrying out the invention (and / or example) to a longer mode for carrying out the invention (and / or example). This makes it possible to provide a mode for carrying out the invention (and / or example) of an appropriate length depending on the application and requirements. 2. Diversification of the forms of modes (and / or embodiments) for carrying out the invention The ninth generation unit 195 can also generate various forms of modes (and / or examples) for carrying out the invention, such as a bulleted list, a summary with illustrations, and a story-like form of modes (and / or examples) for carrying out the invention, which makes it possible to communicate the contents of the invention in a more understandable and impressive manner. 3. Adjusting the Tone and Narrative of the Detailed Description (and / or Examples) The ninth generation unit 195 can also adjust the tone and voice of the mode for carrying out the invention (and / or example) to be generated. Specifically, the user can specify the tone of the mode for carrying out the invention (and / or example), and the voice can be generated in a range from a more formal voice to a softer voice. This makes it possible to provide a mode for carrying out the invention (and / or example) with an appropriate tone depending on the target audience and situation. 4. Multimedia Implementation of the Detailed Description (and / or Examples) The ninth generation unit 195 can generate a multimedia mode (and / or embodiment) for carrying out the invention that includes not only text but also images and videos. Specifically, it automatically selects or generates images and videos related to the content of the invention and embeds them in the mode (and / or embodiment) for carrying out the invention. This makes it possible to communicate the content of the invention in a more intuitive and impactful way. 5. Multilingualization of the Detailed Description (and / or Examples) The ninth generation unit 195 can also translate the generated mode for carrying out the invention (and / or examples) into multiple languages. Specifically, the mode for carrying out the invention (and / or examples) is automatically translated into other languages ​​to generate the mode for carrying out the invention (and / or examples) in multiple languages. This makes it possible to efficiently create the mode for carrying out the invention (and / or examples) corresponding to the languages ​​of each country in global patent applications. 6. Personalization of the Description (and / or Examples) The ninth generation unit 195 can also personalize the mode (and / or example) for carrying out the invention to suit the preferences and writing style of the inventor or applicant. Specifically, it learns the unique expression style of the inventor or applicant from their past mode (and / or example) for carrying out the invention and writings, and generates a mode (and / or example) for carrying out the invention that reflects that. This makes it possible to provide a mode (and / or example) for carrying out the invention that is more natural and familiar to the inventor or applicant. 7. IMPROVEMENT OF THE MODE (AND / OR EMBODIMENT) FOR CARRYING OUT THE INVENTION BY INTERACTION WITH THE USER The ninth generation unit 195 can present the generated mode for carrying out the invention (and / or embodiment) to a user and improve the mode for carrying out the invention (and / or embodiment) based on feedback from the user. For example, the ninth generation unit 195 can have the user evaluate the understandability and appropriateness of the mode for carrying out the invention (and / or embodiment) and modify the mode for carrying out the invention (and / or embodiment) based on the evaluation. This allows the perspectives of the inventor or applicant to be fed back into the generation of the mode for carrying out the invention (and / or embodiment), thereby deriving a better mode for carrying out the invention (and / or embodiment).

[0065] The above describes the modified examples, other forms, and other possibilities of the ninth generation unit 195. The ninth generation unit 195 is able to realize more advanced and efficient abstract extraction and generation by combining various technologies while focusing on a large-scale language model. This is expected to improve the quality of abstracts in patent application documents and greatly improve the efficiency of the creation work.

[0066] It may also generate a patent application form from the given information, and the International Patent Classification may be output by LLM and included in the application.

[0067] The following describes in detail a specific example of automatically sending patent application documents to a patent office (or a superior, a checker, or the Patent Office) after they have been completed. 1. Automatic formatting of application documents Completed patent application documents are automatically adjusted to the format required by the patent office (or patent office) reception system. Specifically, the document layout, font, size, page number, etc. are automatically corrected to comply with the patent office's regulations. This prevents rejection of applications due to incomplete documents. This may also include converting word processing data, text data, drawings, etc. into HTML documents. 2. Automatic PDF conversion of application documents After formatting, patent application documents may be automatically converted to PDF format. At this time, PDF security settings (e.g., no editing, password protection, etc.) are automatically applied to prevent document tampering. PDF properties (e.g., title, author, keywords, etc.) are also automatically set to facilitate document management. A timestamp may be added to certify the creation date and time. 3. Automatic destination selection The patent office, etc. to which patent application documents should be sent is automatically selected from a pre-registered contact list. The contact list contains the name, email address, and reception system URL of each patent office. It is also possible to automatically select the most suitable patent office based on the attributes of the applicant and invention (e.g., technical field, importance, etc.). 4. Automatic selection of sending method Depending on the patent office selected, the system automatically selects the method of sending the patent application documents, such as sending by email, uploading to a dedicated reception system, or sharing via online storage, according to the method specified by each patent office. 5. Automatic tracking of delivery status The system automatically tracks the transmission status of patent application documents and checks whether they were sent successfully. It automatically checks the status of each transmission method, such as confirming delivery after sending via email, notifying users that uploading to a reception system is complete, or notifying users that they have downloaded the documents via online storage. If transmission fails, the system automatically attempts to resend the documents and notifies the user. 6. Automatic confirmation of receipt completion Automatically confirms that application documents have been received by the patent office. Specifically, it automatically detects the receipt completion notification email from the patent office or periodically checks the status change of the receipt system. Once receipt completion is confirmed, the applicant is notified and the application information is automatically registered in the company's patent management system. 7. Automated Processing of Post-filing Notifications After a patent application is filed, various notifications from patent offices and patent offices (e.g., notices of rejection from examiners, notices of allowance for patents, etc.) are automatically received and processed appropriately. Specifically, the system automatically identifies the type of notification and notifies the applicant, inventor, and relevant parties within the company. It also automatically reminds users of necessary actions based on the notification (e.g., submitting an argument regarding the reasons for rejection, paying patent fees in response to the allowance for patents, etc.).

[0068] Above, we have explained the process of automatically sending patent application documents to patent offices and automating the subsequent processing. These automations can significantly improve the efficiency of administrative work related to patent applications and reduce the burden on applicants, inventors, and patent offices. Automation can also prevent document deficiencies and transmission errors, improving the quality and reliability of patent applications.

[0069] The following describes in detail a configuration in which a timestamp is added to patent application documents, etc., so that the date of invention, etc. can be later proven. 1. Automatic generation of timestamps Once a patent application document is completed, a timestamp is automatically generated based on its contents. The timestamp includes the creation date and time of the application document, information about the applicant and inventor, and a hash value of the application document. A trusted timestamp authority (TSA) is used to generate the timestamp, ensuring tamper-proofing and time legitimacy. 2. Embedding timestamps in application documents The generated timestamp is automatically embedded in patent application documents. Specifically, the timestamp is added to the PDF of the application document as a digital signature, and the timestamp information is recorded in the metadata of the application document. This firmly links the application document to the timestamp, preventing tampering with the timestamp. 3. Storage of time-stamped application documents Patent application documents with embedded timestamps are automatically stored in secure storage. Possible storage locations include a dedicated archive system with tamper-proofing capabilities or distributed storage on a blockchain. The stored application documents with embedded timestamps are stored long-term in preparation for future requests for discovery of evidence, etc. 4. Timestamp Verification When verifying the authenticity of a time-stamped patent application document, the validity of the timestamp is first verified. Specifically, the digital signature included in the timestamp is verified with the TSA's public key to confirm that the timestamp has not been tampered with. Next, the hash value of the application document included in the timestamp is compared with the hash value of the actual application document to confirm that the application document has not been tampered with. 5. Proving the Date of Invention If you need to prove the date of invention after filing a patent application, submit the time-stamped patent application documents as evidence. By verifying the timestamp, you can prove the authenticity of the application documents' creation date and contents, which can then be used to prove the date of invention. This allows you to advantageously claim that your company's invention is a prior invention, for example, in a prior invention defense. 6. Identification of the inventor Even if the identification of an inventor becomes an issue after a patent application is filed, time-stamped patent application documents can serve as valid evidence. The inventor at a specific point in time can be identified from the inventor information written in the application documents and the timestamp. This makes it possible to prove the legitimate inventor in cases such as the succession of a work-related invention or a claim for compensation. 7. Update Timestamp To preserve evidence over the long term, timestamps are updated periodically. Specifically, a new timestamp based on the old timestamp is generated every certain period (e.g., every few years) and attached to the application documents. This allows the authenticity of the application documents to be maintained by the new timestamp, even if the encryption algorithm of the old timestamp is compromised.

[0070] We have explained the process for using timestamps to prove the creation date and content of patent application documents. Proof using timestamps is an important issue that directly affects the validity and scope of patent rights. This configuration allows applicants to reliably and efficiently prove the date of invention, etc., which will be useful in protecting and enforcing patent rights.

[0071] The following describes in detail the configuration for verifying the time when an invention notification or invention was entered using a timestamp.

[0072] 1. Real-time saving of input contents This system saves the input contents in real time in invention notification forms and forms where invention details are entered. Specifically, every time a user enters content into an input field, the content is automatically sent to a server and saved. The saved content includes not only the input string, but also the input date and time, user information, etc. 2. Hashing of input A hash value is calculated from the stored input content. A highly secure algorithm such as SHA-256 is used as the hash algorithm. The calculated hash value is used to ensure the integrity of the input content. 3. Generating a timestamp When input is completed (e.g., when the submit button for an invention notification form is pressed) or after a certain period of time has passed, a timestamp is generated based on the input content at that time. The timestamp includes a hash value of the input content, the date and time of generation, user information, etc. A trusted timestamp authority (TSA) is used to generate the timestamp. 4. Storage of timestamps The generated timestamp is stored in secure storage along with the input content. Possible storage locations include a database with tamper-proof functionality or distributed storage on a blockchain. This firmly links the input content and the timestamp, preventing their tampering. 5. Proof of time of input When the novelty or inventive step of an invention is disputed and it becomes necessary to prove the time of input of an invention notification or invention, the stored timestamp can be submitted as evidence. By verifying the timestamp, the authenticity of the hash value of the input content and the generation date and time can be proven, which can be used to prove the time of input. 6. Verifying the authenticity of input Even if the timestamp proves the time of input, there is a possibility that the input itself may have been tampered with after the fact. Therefore, the authenticity of the input is proven by comparing the hash value of the stored input with the hash value contained in the timestamp. If the two match, it is guaranteed that the input has not been tampered with from the time the timestamp was generated to the present. 7. Third-Party Verification To further ensure the verification of the time of input and the contents of the input, third-party verification is enabled. Specifically, the timestamp and hash value of the input contents are recorded in a public ledger such as a blockchain, which is highly resistant to tampering. This allows anyone to verify the consistency of the timestamp and the input contents, improving the reliability of the verification.

[0073] We have explained the process for using timestamps to prove the time of inputting an invention or invention notification. This configuration makes it possible to reliably prove important points related to the novelty and inventive step of an invention in the process from the idea of ​​the invention to the patent application. This is extremely important in obtaining patent rights and determining the scope of rights, and will greatly contribute to protecting the rights of inventors.

[0074] We will provide a detailed explanation of how to automatically submit completed patent application documents to the Patent Office, including how to use online application software and APIs. 1. Automated application verification The system automatically verifies whether completed patent application documents comply with the format and content required by the Patent Office. Specifically, it checks the document's items, character count, whether or not attached documents are included, and issues an alert if there are any deficiencies. This verification utilizes validation tools provided by the Patent Office and checklists created from past application data. 2. Encryption of applicant information Application documents containing personal information of applicants and inventors are encrypted for secure handling. Public key cryptography is used for encryption, and by encrypting with the Patent Office's public key, only the Patent Office can decrypt the data. This reduces the risk of application data being leaked to external parties. 3. Automatic payment of application fees Automatically pay the fees required for patent applications via online banking or credit card. The fee amount is automatically calculated based on the type of application and the number of claims. Once payment is complete, the payment number is automatically recorded on the application documents. 4. Automated online applications The application procedure is automated using the internet application software provided by the Japan Patent Office. The detailed steps are shown below. a. Application software version check It automatically checks whether the latest online application software is installed and updates it if necessary, allowing you to maintain the latest application environment at all times. b. Logging in to the application software You will automatically log in to the online application software using your digital certificate and password. The digital certificate must be obtained in advance in a secure manner and registered in the system. c. Attachment of application documents The encrypted application documents and required attachments are automatically attached to the online application software, with the file format and size verified again to ensure they are appropriate. d. Submitting application data Application data, including application documents, attachments, payment information, etc., are automatically sent to the Patent Office via internet application software. When sending, care must be taken to encrypt communications and handle errors. e. Automatically save receipts Automatically download receipts from the patent office and save them in a designated location. Receipt documents contain important information such as the application number and the date and time of receipt, so they must be stored securely. 5. Automating API Applications The application process is automated using the REST API provided by the Japan Patent Office. The application procedure using the API is shown below. a. Obtaining an access token To use the Patent Office's API, an access token is obtained using an authentication protocol such as OAuth 2.0. Access tokens are valid for a limited time and must be updated periodically. b. Preparation of application data Prepare the encrypted application documents, attachments, applicant information, etc. as the payload for the API request. The payload format is described in structured data such as JSON or XML according to the API documentation. c. Calling the application API Use the prepared payload to call the Patent Office's application API with a POST request. Set the API URL, parameters, headers, etc. according to the API documentation. d. Response analysis Analyze the response returned from the application API and determine whether the application was accepted successfully. If accepted, extract information such as the application number and save it in the database. If an error occurs, analyze the error code and message and perform appropriate error handling. 6. Automatic notification of application status Once an application is completed, the applicant, inventor, and relevant parties within the company will be automatically notified. The notification will include information such as the application number, filing date, and expected examination schedule. Users can choose the notification method based on their preferences, such as email, chatbot, or posting to the company's internal system. 7. Automatic backup of application data Automatically back up application data and store it in a safe location. Backups should be made not only immediately after application, but also periodically. Use cloud storage or an in-house backup server as the backup destination. This allows application data to be restored in the unlikely event of an emergency.

[0075] Above, we have explained the detailed configuration for automatically submitting completed patent application documents to the Patent Office. By using online application software in conjunction with APIs, the automation of applications can be further advanced. Furthermore, by automating security measures, error handling, backups, and other aspects, it is possible to significantly improve the reliability and efficiency of application procedures.

[0076] In the case of employee inventions, it is necessary to properly manage the payment of rewards to inventors. This article explains in detail the structure for managing the payment of rewards at each stage: at the time of application, at the time of patent acquisition, and at the time of patent maintenance. 1. Inventor information management Information on inventors of employee inventions is managed in conjunction with the company's internal personnel database. Specifically, information such as the inventor's department, position, and contact information is automatically retrieved from the personnel database and registered in the patent management system. This makes it possible to properly manage the recipients of bonus payments even if the inventor is transferred or resigns. 2. Digitalization of reward regulations The company's internal regulations regarding bonuses for employee inventions are managed as digital data. The regulations clearly state the timing and amount of bonus payments, as well as the calculation method. The digitalized regulations are registered in a patent management system and used to automatically calculate bonuses. 3. Reward Management at the Time of Application Once a patent application is completed, the inventor will be automatically paid a bonus at the time of filing. The bonus amount will be calculated automatically based on company regulations. The calculation will take into account the inventor's contribution and the value of the invention. The payment will be automatically transferred to a designated bank account, separate from the inventor's salary. 4. Management of rewards at the time of patent grant Once a patent is registered, the inventor will automatically receive a bonus upon patent acquisition. The amount of the bonus is calculated automatically based on company regulations. The calculation takes into account the number of patent claims and the implementation status of the invention. Payment will be made by automatic bank transfer, just as it was at the time of application. 5. Reward Management for Maintaining Rights Every time an annuity payment required to maintain a patent is made, a bonus for maintaining the patent is automatically paid to the inventor. The amount of the bonus is calculated automatically based on company regulations. The calculation takes into account factors such as the remaining term of the patent and the profits from the exploitation of the invention. Payments are made by automatic bank transfer, just as they are at the time of filing an application or when the patent is granted. 6. Bounty Payment Notification Once the payment of the reward has been completed, the inventor will be automatically notified. The notification will include information such as the payment amount, calculation basis, and bank details. The notification method will be selectable according to the inventor's preference, such as email or posting to the company's internal system. 7. Tax Treatment of Bonuses The system automatically processes tax procedures related to bonus payments. Specifically, it withholds income tax and local resident tax depending on the payment amount and timing. It also automatically generates the necessary tax documents and submits them to the tax office. This reduces the tax burden on inventors and ensures tax compliance for the company. 8. Accounting for incentive payments The system automatically processes accounting for bonus payments. Specifically, it records bonus payments in the appropriate account and in the accounting books. It also automatically generates and stores supporting documents for bonus payments. This ensures the accuracy of the company's financial statements and reduces the burden of responding to audits. 9. Bounty Data Analysis Analyzing the data on bonus payments will help promote employee inventions. Specifically, we visualize the trends in bonus payments by department and by inventor, and analyze the correlation between bonus payments and the quality and quantity of patents. This allows us to quantitatively evaluate the effectiveness of the bonus system and revise it as necessary.

[0077] Above, we have explained the configuration for managing the payment of employee invention incentives. By automatically calculating and paying incentives at each stage of patent application, patent acquisition, and patent maintenance, inventors can be appropriately provided with incentives. Furthermore, by automating tax processing, accounting processing, data analysis, etc., it is possible to reduce the operating costs of the incentive system and effectively promote employee inventions.

[0078] The above configuration can also be applied when creating patent application documents by inputting slides and papers from academic conference presentations. The process is explained in detail below. 1. Input data mapping Each section of a conference presentation slide or paper is automatically matched to the corresponding item in a patent application. For example, the "Introduction" and "Introduction" sections of a paper are matched to the "Background Art" section of a patent application. Similarly, the "Experimental Method" and "Results" sections of a paper are matched to the "Mode of Invention" section of a patent application. This matching is achieved using natural language processing technology and machine learning. 2. Recognition and extraction of inventions The system automatically recognizes and extracts invention-related descriptions from input slides and papers. Specifically, it finds novel technical features and solutions to problems. This process involves comparing with patent databases and analyzing technical terms. The extracted invention content is used to create patent claims. 3. Automatic generation of drawings The system automatically generates the drawings required for patent application documents from figures and charts in papers and illustrations in slides. For example, it reconstructs schematic diagrams of experimental equipment and flow charts to fit the format of patent drawings. This process utilizes image recognition technology and automated graphic design technology. 4. Identification of the inventor The system automatically identifies inventors from the author information of papers and the presenter information of slides. At the same time, the system also extracts the affiliations and contact information of the authors and presenters, and automatically enters them into the inventor field of the application documents. In the case of co-inventions, the system automatically determines the order of inventors based on the contributions of the authors and presenters. 5. Prior Art Search The system automatically searches for relevant prior art from references cited in input slides and papers. The search utilizes not only patent databases but also academic literature databases. If the search finds any highly relevant prior art, it automatically adds that content to the background art section. 6. Automatic Claim Generation The system automatically generates patent claims from the extracted invention content. The system verbalizes the essential features of the invention while taking into account patent law requirements (novelty, inventive step, industrial applicability, etc.). The system also automatically constructs a claim set according to the structure of independent and dependent claims. 7. Automatic statement generation The system automatically generates a patent specification by integrating the extracted invention details, generated drawings, and prior art search results. The specification is generated by logically constructing a detailed description of the invention in accordance with the requirements of patent law and the guidelines for writing specifications. Explanations of technical terms and descriptions of examples are also automatically added. 8. Automatic application document checking The generated patent application documents are automatically checked against patent law and examination guidelines. The checks include both formal requirements (document format, writing method, etc.) and substantive requirements (unity of invention, support requirements, etc.). If any deficiencies are found as a result of the check, they are automatically corrected or the user is notified.

[0079] Above, we have explained the process of creating patent application documents by inputting conference presentation slides and papers. Automating this process can significantly lower the barrier to patent applications for researchers. It also leads to improved quality of patent applications and reduced application costs. However, the automatically generated application documents are merely drafts, and ultimately require review and polishing by a patent attorney or other expert.

[0080] Instead of an invention proposal, you can input slides and papers from academic presentations and use the structure shown in Figure 1 to create patent application documents.

[0081] The following configurations of the invention are also possible. [Claim 1] An information processing device for preparing documents for patent applications, which is equipped with a checking means for accepting slides for academic presentations, papers, online posts, and emails, and checking whether or not they contain inventions, a first generation means for extracting prior art from slides from the conference presentation, papers, online posts, and emails using a large-scale language model, and generating prior art from training data, a database, or the Internet if prior art cannot be extracted; A second generation means for extracting issues from slides at the time of the academic conference presentation, papers, online posts, and emails using a large-scale language model, and generating issues from training data, a database, or the Internet if the issues cannot be extracted; A third generation means extracts a solution to the problem from the slides of the conference presentation, a paper, a post on the Internet, or an email using a large-scale language model, and if the solution cannot be extracted, generates it from training data, a database, or the Internet; a fourth generation means for extracting the effect of the invention from the slides of the academic conference presentation, papers, online posts, and emails using a large-scale language model, and generating the effect from training data, a database, or the Internet if the effect cannot be extracted; a fifth generating means for extracting the name of the invention from the slides of the academic presentation, the paper, the online post, or the email using a large-scale language model, and generating the name of the invention from training data, a database, or the Internet if the name cannot be extracted; A sixth generation means for extracting a drawing from the slides of the conference presentation, a paper, a post on the Internet, or an email, and if the drawing cannot be extracted, generating a drawing or a code for generating a drawing from learning data, a database, or the Internet; a seventh generation means for extracting a description of the drawing from slides, papers, online posts, or emails from the conference presentation using a large-scale language model, and generating the description from training data, a database, or the Internet if the description cannot be extracted; an eighth generation means for extracting an abstract from the slides of the academic conference presentation, a paper, a post on the Internet, or an email using a large-scale language model, and if the abstract cannot be extracted, generating the abstract from training data, a database, or the Internet; and output means for outputting documents for a patent application by combining the generated outputs of said first to eighth generation means.

[0082] 2 is a functional block diagram of an information processing device 100 according to an embodiment of the present invention. The information processing device 100 is a device that accepts slides and papers for academic presentations, online posts, and emails, and creates documents for patent applications. It may also accept URLs of sites to be made public on the Internet.

[0083] The information processing device 100 includes a check unit 110, a first generation unit 120, a second generation unit 130, a third generation unit 140, a fourth generation unit 150, a fifth generation unit 160, a sixth generation unit 170, a seventh generation unit 180, an eighth generation unit 190, a ninth generation unit 195, and an output unit 200.

[0084] The checking unit 110 checks whether input slides from academic presentations, papers, online posts, and emails contain inventions. Specifically, it uses natural language processing technology and machine learning to extract technically novel content from the input data and determines whether it is patentable.

[0085] The first generation unit 120 uses a large-scale language model to extract prior art from input data that has been determined to contain an invention by the check unit 110. If prior art cannot be extracted, the first generation unit 120 generates related prior art from training data, a database, or the Internet.

[0086] The second generation unit 130 extracts tasks from the input data using a large-scale language model, and if a task cannot be extracted, generates related tasks from training data, a database, or the Internet.

[0087] The third generation unit 140 extracts a solution to the problem from the input data using a large-scale language model. If a solution cannot be extracted, the third generation unit 140 generates a related solution from training data, a database, or the Internet.

[0088] The fourth generation unit 150 extracts the effects of the invention from the input data using a large-scale language model. If the effects cannot be extracted, it generates related effects from training data, a database, or the Internet.

[0089] The fifth generation unit 160 extracts the name of the invention from the input data using a large-scale language model. If the name cannot be extracted, it generates an appropriate name from training data, a database, or the Internet.

[0090] The sixth generation unit 170 extracts a drawing from the input data. If a drawing cannot be extracted, it generates a related drawing from learning data, a database, or the Internet, or a code for generating the drawing.

[0091] The seventh generation unit 180 extracts a description of the drawing from the input data using a large-scale language model. If a description cannot be extracted, it generates an appropriate description from training data, a database, or the Internet.

[0092] The eighth generation unit 190 extracts an abstract from the input data using a large-scale language model, or if an abstract cannot be extracted, generates an appropriate abstract from training data, a database, or the Internet.

[0093] The ninth generation unit 195 extracts modes (and / or examples) for carrying out the invention from the input data using a large-scale language model. If modes (and / or examples) for carrying out the invention cannot be extracted, the ninth generation unit 195 generates appropriate modes (and / or examples) for carrying out the invention from training data, databases, or the Internet.

[0094] The output unit 200 integrates the various data generated by the first generation unit 120 to the ninth generation unit 195, and generates and outputs documents for patent applications. The output format is automatically adjusted to fit the patent office's acceptance system.

[0095] The information processing device 100 described above can automatically extract or generate various components required for patent applications from academic presentation slides and papers, online posts, and emails, and then integrate them to efficiently prepare patent application documents. This enables smooth patent application of inventions by researchers and engineers, promoting the protection and utilization of intellectual property.

[0096] The following describes in detail the configuration for determining whether to continue or terminate a patent right after it has been granted. 1. Tracking patent enforcement Track whether granted patents are actually being implemented by regularly monitoring the use of patents in your own products and services, the existence and content of license agreements, and the use of patents by competitors. This information is an important indicator for assessing the value of patents. 2. Patent invalidity risk assessment We evaluate the risk of an invalidation trial being requested for a patent that has been granted a right. Specifically, we re-examine the validity of the patent by investigating prior art and examining the patent's novelty, inventive step, and description requirements. We also analyze the relationship with competitors' patents and products to predict the possibility of an invalidation trial being requested. Patents with a high risk of invalidation may not be worth the cost of maintaining the rights. 3. Evaluating the patent's technological superiority The technical superiority of patents is evaluated in light of the latest technological trends. Specifically, we analyze the degree of superiority of the technical features described in the patent claims in relation to the current state of the art. We also take into account the existence of alternative technologies and the speed at which technologies become obsolete. Patents whose technological superiority has declined may no longer need to be maintained. 4. Assessing the market value of patents The market value of a patent is evaluated based on the sales and profit margins of related products and services. Specifically, the economic value of a patent is calculated by multiplying the sales of the product or service in which the patent is implemented by the patent's contribution. It also takes into account licensing income and the future potential of businesses that utilize the patent. Patents with low market value may not justify the cost of maintaining the rights. 5. Assessing the Defensive Value of Patents We evaluate whether granted patents act as barriers to entry for competitors. Specifically, we analyze the extent to which patents affect competitors' research and development and business development. We also consider the risk of patent disputes with competitors and the possibility of cross-licensing. Patents with high defensive value may be worth maintaining even if they are not being used. 6. Optimizing your entire patent portfolio The evaluation results of each individual patent are combined to optimize the entire patent portfolio. Specifically, the priority of maintaining rights is determined according to the value and importance of the patent. It is also effective to group related patents and decide whether to maintain rights on a group-by-group basis. Optimizing the entire patent portfolio makes it possible to maximize the value of patents while minimizing the cost of maintaining rights. 7. Patent Term Considerations Consider the duration of the patent and compare the remaining patent term with the cost of maintaining the patent. Generally, the value of a patent tends to decline towards the end of the patent term, so patents with a short remaining patent term may be given a lower priority for maintaining the patent. However, as there are some patents, such as pharmaceutical patents, that are only implemented towards the end of the patent term, a blanket judgment should be avoided. 8. Automating the patent maintenance decision-making process The above evaluation items are quantified to automate the decision-making process for maintaining patent rights. Specifically, the importance of each evaluation item is set and the evaluation results are converted into a score to calculate a rights maintenance score for each patent. Based on this score, it is possible to automatically determine whether or not to maintain rights and present possible decisions. By automating the decision-making process, it is possible to significantly reduce the time and effort required for decision-making while ensuring objectivity and consistency in evaluation.

[0097] Above, we have explained the structure for determining whether to continue or discontinue patent rights after they have been granted. It is important to conduct a multifaceted evaluation, including the patent's implementation status, invalidity risk, technological superiority, market value, and defensive value. It is also essential to optimize the entire patent portfolio and consider the duration of the rights. Automating these evaluation processes as much as possible will enable efficient and effective patent management.

[0098] Next, we will explain in detail the module that accepts invention consultations from researchers, checks patent requirements, and determines whether or not to file an application. 1. Invention consultation reception It provides an interface for accepting invention consultations from researchers. Specifically, it allows researchers to enter details of their inventions through an online invention consultation form or an in-house invention reporting system. Input items include the name and summary of the invention, the problem and solution, and its effects. It also accepts attachments such as related papers, presentation materials, and experimental data. 2. Automatic classification of invention content The content of the invention consultation received is automatically classified using natural language processing technology. Specifically, the content of the invention is analyzed from the perspective of keywords, technical field, problems and solutions, etc., and classified according to a pre-defined classification system. The classification results are used in subsequent patent requirement checks and prior art searches. 3. Check patentability requirements The system automatically checks patent requirements for received invention consultations. Specifically, it uses AI to evaluate the following requirements: Novelty: Evaluate whether the invention is new compared to existing technology. This evaluation utilizes the results of prior art searches using patent databases and paper databases. b. Inventive step: Evaluate whether the invention is something that could easily be conceived from existing technology. This evaluation involves a comprehensive analysis of the problem, solution, and effects of the invention, and determines whether it would be non-obvious from the perspective of a person skilled in the art. c. Industrial Applicability: Evaluate whether the invention can be applied industrially. This evaluation takes into account the feasibility of implementing the invention, marketability, and legal restrictions. 4. Automating prior art searches The system automatically performs prior art research necessary for checking patent requirements. Specifically, it searches patent databases, research paper databases, web databases, etc. using keywords and classification codes related to the content of the invention. From the search results, it automatically extracts prior art documents highly relevant to the invention and uses them to evaluate novelty and inventive step. 5. Judgment on whether or not to accept the application The system automatically determines whether an invention should be patented by combining the results of patent requirement checks and prior art searches. It uses preset thresholds to make the decision. For example, if the evaluations of novelty and inventive step are both above a certain level, and no similar inventions are found in the prior art search, it determines that the invention can be patented. 6. Explaining and visualizing the results of decisions The results of the application approval / disapproval decision are explained to researchers in an easy-to-understand manner. Specifically, the evaluation results of patent requirements and prior art search results are visualized using graphs and charts. The evaluation criteria and prior art documents that formed the basis of the decision are also explained in detail. This allows researchers to deepen their understanding of the patentability of their inventions. 7. Requests for Additional Information and Reassessment If the decision on whether to grant an application is unclear or if the researcher provides additional information, a reevaluation will be conducted. Specifically, the researcher will be asked to submit additional information and materials, and the patent requirements will be checked again, including these, and a prior art search will be conducted again. The decision may be overturned as a result of the reevaluation. 8. Application procedure support For inventions that are deemed eligible for patent application, the system supports the patent application process. Specifically, it automatically generates the scope of claims, specifications, drawings, etc. from the content of the invention, and supports the preparation of documents required for application. It also supports post-application intermediate processing, handling of rejections, and schedule management until rights are granted.

[0099] Above, we have explained the module that accepts invention consultations from researchers, checks patent requirements, and determines whether or not to apply for patents. By effectively operating this module, researchers' inventions can be properly evaluated, leading to patent applications. Furthermore, by utilizing AI, it is possible to improve the accuracy and efficiency of evaluations and reduce the burden on researchers. However, it is important to note that AI evaluations are only for reference, and the final decision must be made by a patent expert.

[0100] Next, we will provide a detailed explanation of the module that accepts invention consultations from researchers, checks patent requirements, and suggests research directions. 1. Reception and classification of invention consultations It provides an interface for researchers to request invention consultations and automatically classifies the content of the consultations received. The classification is based on aspects such as the technical field of the invention, the problem, and the means of solving it. The results of this classification are used in subsequent patent requirement checks and research direction proposals. 2. Checking patentability requirements The system automatically checks patent requirements (novelty, inventive step, industrial applicability) for received invention consultations. This involves conducting prior art searches using AI and analyzing the invention content. If the check shows that the patent requirements are not met, the reasons are analyzed and feedback is provided to the researcher. 3. Research and analysis of related technologies The system automatically searches patent databases, research paper databases, web databases, etc. for technologies related to the content of the invention consultation. The results of the search are analyzed to provide information on trends, issues, and solutions in related technologies to researchers. This information is used to propose research directions. 4. Marketability Analysis The system automatically analyzes the marketability of products and services related to the invention consultation. Specifically, it investigates the size and growth rate of the relevant market, as well as the trends of competing companies, to evaluate the commercial viability of the invention. It also analyzes the applicable uses and customer segments of the invention and presents this information to researchers. 5. Technological Trend Forecasting We predict future trends in technological fields related to the content of invention consultations. Specifically, we use AI to analyze time-series data on patent applications and paper publications to predict technological development directions and promising research themes. These prediction results are reflected in proposals for research directions. 6. Proposal of research direction We will synthesize the results of patent requirement checks, analysis of related technologies, marketability analysis, and technological trend forecasts, and propose research directions to researchers. The proposals will include the following: a. Invention improvement proposals: Specific proposals are made on which parts of the invention should be improved to meet patent requirements. Improvement proposals utilize the results of research into related technologies and the results of AI analysis of the invention content. b. Proposal for additional experiments and evaluations: Propose additional experiments and evaluations to verify the effectiveness and feasibility of the invention. The proposal should include specific experimental methods and data acquisition methods. c. Proposal for fusion with related technologies: Propose the possibility of creating new value by combining the invention with related technologies. The proposal should include specific methods for combining the technologies and the expected effects. d. Proposal for new uses and markets: If the possibility of commercialization can be increased by expanding the applicable uses and markets of the invention, propose a direction for doing so. The proposal should include specific uses, target markets, and entry strategies. 7. Visualization and explanation of proposals We visualize and explain proposed research directions in an easy-to-understand way for researchers. Specifically, we use charts and graphs to show the data and analysis results that form the basis of the proposal. We also present a roadmap for realizing the proposal and the expected results. 8. Gather feedback and learn We collect feedback from researchers on the proposals and use it to train the modules. Specifically, we collect researchers' evaluations of the usefulness and feasibility of the proposals. We use this evaluation data to retrain the proposed generative model and improve its accuracy.

[0101] Above, we have explained the module that accepts invention consultations from researchers, checks patent requirements, and proposes research directions. Utilizing this module not only helps researchers patent their inventions, but also promotes research itself. Furthermore, utilizing AI is expected to maximize researchers' creativity and accelerate innovation. However, it is important to note that the proposals are for reference only, and the final decision on research direction must be made by the researcher themselves.

[0102] Next, we will provide a detailed explanation of the module that analyzes a company's past, including patent information, and the development status of other companies, and proposes research directions, etc. 1. Analysis of your own patents Analyze your company's past patent applications and granted patents to understand your company's technological strengths and weaknesses, as well as research and development trends. Specifically, analyze the patent's IPC classification, application date, inventor, citation relationships, etc. to visualize your company's technology portfolio. Also, analyze the number of citations and duration of each patent to identify important patents. 2. Analysis of other companies' patents Analyze the patent application status of competitors and companies in related industries to understand their R&D trends and technological advantages. Specifically, analyze the patent IPC classification, application period, inventors, citation relationships, etc. to visualize other companies' technology portfolios. Also, analyze the relationship between your company's patents and those of other companies to identify competitive technological areas and areas where collaboration is possible. 3. Analysis of technological trends We analyze trends in the overall technology field and future trends based on patent information. Specifically, we analyze the changes in the number of patent applications and registrations over time to identify technological areas that are attracting attention and areas that are in decline. We also analyze the technical content of patents using natural language processing technology to extract new technological keywords and promising research themes. 4. Analysis of blank maps Analyze the distribution of your own patents and those of other companies to identify technology areas with few patent applications (blank maps). Blank maps are areas with little competition and room for new entrants, making them candidates for research and development. Patent mapping and landscape analysis methods are used to analyze blank maps. 5. Patent Citation Network Analysis We analyze patent citation relationships as a network structure and extract important patents and technologies. Specifically, we extract patents with high citation counts and hub patents that connect citation sources and citation destinations, visualizing the lineage and development path of technology. We also analyze citation relationships between our own patents and those of other companies to understand the inflow and outflow of technology. 6. Proposal of research and development direction We will propose the direction of research and development by integrating the results of our own patent analysis, the results of the analysis of other companies' patents, the results of the analysis of technology trends, the results of the analysis of blank maps, and the results of the analysis of patent citation networks. The proposal will include the following: a. Proposing technology areas to focus on: Proposing areas where the company can utilize its strengths, areas where it can differentiate itself from other companies, and areas where future growth is expected. b. Proposal of a new research topic: Based on the results of the blank map analysis and the analysis of technological trends, propose a new research topic. The proposal should also include the purpose of the research and expected results. c. Proposals for collaboration and partnerships with other companies: Based on the results of analysis of other companies' patents and patent citation networks, we propose companies with complementary technical relationships and companies that are candidates for joint research. d. Proposing patent strategies: Based on the analysis of the relationship between our own patents and those of other companies and the results of technological trends, we propose patent strategies such as patent applications, rights acquisition, and licensing. 7. Visualization and explanation of proposals We visualize and explain proposed R&D directions to managers and researchers in an easy-to-understand manner. Specifically, we use graphical representations such as patent maps, landscapes, and citation networks to show the data and analysis results that form the basis of the proposal. We also show a roadmap for realizing the proposal and the necessary resources. 8. Gather feedback and learn Feedback on the proposals from managers and researchers is collected and used to train the modules. Specifically, evaluations from stakeholders are collected on the feasibility of the proposals and their expected effects. This evaluation data is used to retrain the generative model of the proposals and improve their accuracy.

[0103] We have explained above the module that analyzes a company's past, including patent information, and the development status of other companies, and proposes research directions. By utilizing this module, companies can leverage their technological strengths, differentiate themselves from others, and advance research and development that anticipates future technological trends. Furthermore, by utilizing AI, it is possible to efficiently analyze vast amounts of patent information and gain insights and ideas that would be difficult for humans to notice. However, it is important to note that the proposals are merely for reference, and the final decision on research and development policy must be made as a management decision.

[0104] In addition, when generating an embodiment of a patent document using a large-scale language model, a configuration in which multiple major items are output in response to a prompt and text is output sequentially for each major item in response to the prompt is described in detail below. By dividing the embodiment of a patent document into major items and explaining it sequentially, an easy-to-understand embodiment of a patent document can be created. It is also possible to create a hierarchical description, such as major items, medium items, and minor items, and extract and generate explanations for each item. 1. Enter the outline of the invention An outline of the invention to be patented is input into a large-scale language model. The input information includes the name of the invention, the problem, the solution, and the effects. This information is used as context for generating embodiments. As mentioned above, various inputs, such as an invention proposal, can be used. 2. Creation of major items of the embodiment After inputting the outline of the invention, a prompt is given to the large-scale language model to generate the major items (or hierarchical items such as medium items and minor items) of the embodiment. Examples of the prompt include the following: Please suggest five major items to clearly explain the mode of carrying out the following invention. In response to this prompt, the large-scale language model proposes multiple major categories of embodiments based on the outline of the invention. The proposed major categories are selected appropriately depending on the technical field, problem, and solution of the invention. Major categories may also be those relating to the configuration or operation of the device. 3. Generate detailed descriptions for each major topic For each of the major categories of the generated embodiment, detailed descriptions are generated in sequence using a large-scale language model. Specifically, the following prompts are used to generate the description of each major category: Please provide a detailed explanation of "Major item 1." Please provide a detailed explanation of "Major item 2." ... In response to this prompt, the large-scale language model generates a detailed description of each major topic based on the outline of the invention, including the technical features of the invention, the means by which it solves the problem, and its effects. 4. Formatting and editing the generated description The descriptions of each major topic generated by the large-scale language model are formatted and edited to fit the format of patent documents. Specifically, the following process is performed. a. Proofreading: Automatically correct spelling and grammatical errors. b. Standardization of terminology: Automatically replace technical terms used in the description of inventions with standardized expressions. c. Summarizing and elaborating the text: Summarize and / or elaborate the generated explanations as necessary. d. Insertion of diagrams and tables: Automatically generate and insert appropriate diagrams and tables according to the content of the explanation. 5. Check overall consistency The consistency of the content is checked throughout the generated embodiment. Specifically, the following points are checked: a. Consistency with the summary of the invention: Check whether the description of the embodiment is consistent with the problems, solutions, and effects stated in the summary of the invention. b. Consistency among major categories: Check whether there are any contradictions or overlaps in the explanations of each major category. c. Technical accuracy: Check whether the content of the description is technically correct. If there are any inconsistencies, fix or regenerate the relevant parts. 6. User Verification and Correction The generated description of the embodiment is presented to the user (applicant, inventor, patent officer, etc.) for confirmation. The user examines the content of the description and makes corrections or additions as necessary. The user's corrections are fed back to improve the quality of the results generated by the large-scale language model. 7. Output of the final embodiment The description of the embodiment that has been confirmed and corrected by the user is output as part of the patent document. The output embodiment, together with other documents required for the patent application (abstract, claims, drawings, etc.), constitutes a complete set of application documents.

[0105] We have described a configuration for generating embodiments of patent documents using a large-scale language model, in which multiple items are output in response to prompts, and the content of each item is output in sequence in response to prompts. By adopting this configuration, comprehensive and detailed descriptions of embodiments that are in line with the outline of the invention can be efficiently generated. Furthermore, by incorporating user confirmation and correction, it is possible to ensure the quality of the generated descriptions. However, it is important to note that the generated descriptions are merely drafts, and ultimately require review and refinement by patent experts.

[0106] In addition, because patent specifications tend to be lengthy, it is preferable to first define the necessary documents (application, specification, drawings, claims, abstract) and then output each separately (in order), rather than immediately creating patent application documents from input data. It is also preferable to output the contents of each document item separately (in order).

[0107] For example, in the case of a specification, it is desirable to generate the "Title of the Invention," "Technical Field," "Background Art," "Prior Art Documents," "Summary of the Invention," "Problem to be Solved by the Invention," "Means for Solving the Problem," "Effects of the Invention," "Brief Description of the Drawings," "Form for Implementing the Invention," etc. separately and separately using LLM, and then finally combine them.

[0108] It is also desirable to generate the scope of claims not all at once, but for each claim, category (product, method), corresponding embodiment, independent claim, and dependent claim.

[0109] For the description of the embodiments of the invention, the LLM may first generate the major categories to be described, and then output the contents of each category sequentially. For example, a major category for each embodiment, such as "First embodiment," "Second embodiment," etc., may be generated, and each embodiment may be output in sequence. Alternatively, major categories such as "Device configuration," "Device operation," and "Effects of the device" may be generated, and the contents of each category may be output in sequence. Furthermore, the explanation for one drawing may be considered as one major category, and the explanation may be output sequentially for each drawing (item).

[0110] As described above, the information processing device comprises an input receiving unit, first to ninth generation units, and an output unit. The input receiving unit receives input from an invention notification form or a form for inputting an invention (receives information about a specific invention). The first to ninth generation units use a large-scale language model to extract or generate prior art, problems, solutions, effects, invention titles, drawings, drawing descriptions, and abstracts. The output unit compiles each generated element and outputs patent application documents. The information processing device of the present invention can significantly reduce the burden on applicants by automatically generating documents necessary for patent applications from the contents of invention notification forms and input forms.

[0111] In this example, all documents required for a patent application are created using multiple generation methods utilizing large-scale language models, starting with the initial input of the invention notification form and form input provided by the inventor. The process is divided into multiple stages, from extracting information from the initial input, completing the necessary information, and outputting the final document. The stages are as follows:

[0112] Initial input acceptance: Receive input from inventors via invention notification forms or online forms.

[0113] Extraction of prior art (first generation means): A large-scale language model is used to extract descriptions of prior art from the input information. If there is missing information, it is supplemented through external databases or internet searches.

[0114] Extracting problems (second generation method): Identify and clearly document the problems that the invention aims to solve from the input. If there is still insufficient information in this process, supplement it.

[0115] Proposing a solution (third generation step): Clarify how the proposed invention will solve the problem. If necessary, gather additional information.

[0116] Specifying the effects of the invention (fourth generation means): Identify and describe the effects and advantages obtained by the invention.

[0117] Determining the name of the invention (fifth generation means): Generate a name that is appropriate for the invention. This name must succinctly reflect the content of the invention.

[0118] Drawing generation (sixth generation means): Extract or generate drawings related to the invention. If drawings are missing, generate a drawing code and create a new drawing.

[0119] Extracting Drawing Descriptions (Seventh Generation Method): Generate detailed descriptions for the drawings, including descriptions of what each part of the drawing shows.

[0120] Preparation of abstract (eighth generation means): An abstract is prepared that briefly explains the outline of the invention as a whole.

[0121] Final document output: All of the above generated results are integrated and output as a set of patent application documents.

[0122] The information processing device of this embodiment may combine a large-scale language model with an external database. This allows for the use of abundant information related to inventions to quickly and accurately generate documents necessary for patent applications. This system significantly reduces the burden on inventors and patent attorneys when preparing patent applications.

[0123] Modifications will be described below.

[0124] In a modified example, the above-mentioned information processing device is applied to provide an extended function that not only prepares patent application documents but also performs patent validity analysis and competitive analysis. This system aims to automate preliminary evaluation of patentability, comparative analysis with competing patents, and even patent strategy proposals based on information collected from invention notification documents or form input. The system further comprises the following means: 1. Preliminary evaluation of patentability (9th generation means): - Using large-scale language models, a preliminary assessment of the patentability of an invention is performed based on the information input, including searching patent databases for similar and prior art, and analyzing the invention's novelty and inventive step. 2. Comparative analysis with competing patents (10th generation method): - Identifying existing patents that may compete with the proposed invention and conducting a comparative analysis to clarify their differences, which will provide useful information for developing a patent application strategy. 3. Patent strategy proposal (11th generation method): - Based on the results of the preliminary evaluation and competitive analysis, we will propose the best strategy for patenting your invention, including optimizing the patent claim structure, target geographic filing strategy, and even ways to avoid potential patent infringement risks. 4. Generate and output the final report: - Compile the information obtained through all the above processes and output a patent feasibility assessment report, a competitive analysis report, and a patent strategy proposal in addition to the patent application documents.

[0125] The information processing system of this modified example not only prepares patent application documents but also evaluates the market position of an invention and supports more strategic patent application. This system allows inventors and patent agents to gain a deeper understanding of the commercial value of an invention and the possibility of obtaining a patent prior to filing a patent application, enabling them to gain important insights in formulating patent strategies.

[0126] As a modified example, an interactive patent application support system can also be provided. In this modified example, an interactive patent application support system is provided that dynamically responds to various scenarios that inventors and patent agents may face. This system responds immediately to inputs and questions from users, and aims to make the patent application process more flexible and efficient. The configuration is described below. 1. Interactive input interface (first function): - Provides an interactive guide that asks questions and helps users enter information about their invention, helping them collect more detailed and accurate information. 2. Real-time feedback (2nd function): - Provides real-time feedback based on the information you provide, including identifying deficiencies and providing suggestions to improve the success of your patent application. 3. Question-answering system (third function): - Utilizing large-scale language models, the system provides instant answers to users' questions about patent applications, helping to quickly resolve doubts in the patent application process. 4. Customized Document Generation (Fourth Function): - Generate customized patent application documents by combining user input with additional information obtained through interactive sessions, providing documents in the format and content best suited to each individual invention. 5. Progress Management and Notifications (Function 5): - Track progress at each stage of the patent application process and notify users of important milestones and required actions, helping to prevent delays and streamline the process.

[0127] This interactive patent application support system dynamically responds based on user input, making the patent application process more user-friendly and effective. Real-time feedback and question-and-answer functionality also allow users to immediately resolve any questions about their patent application, resulting in higher-quality application documents. Progress management functionality also improves visibility into the entire process, ensuring a smoother path to patent acquisition.

[0128] Alternatively, an AI-driven system can be provided to identify, analyze, and manage various risks during the patent application process. The system aims to increase the probability of successful patent applications by utilizing large-scale language models, data analysis, and predictive modeling techniques. The system has the following capabilities: 1. Risk Identification and Assessment (Function 1): - Automatically identify risk factors associated with patent applications and assess the severity and probability of those risks, including lack of technical novelty, oversight of prior art, and incomplete application documents. 2. Proposing risk response strategies (secondary function): - Based on identified risks, propose specific strategies and action plans to mitigate or avoid the risks, which may include conducting additional technical investigations, detailed documentation reviews, or restructuring claims. 3. Progress and Risk Monitoring (Function 3): - Track progress through the patent application process and identify emerging risks in real time, as well as assess the effectiveness of existing risk response strategies and adjust them as needed. 4. Predictive Analytics and Risk Prevention (Function 4): - Analyze past patent application data and performance to predict the risks that a particular application may face, allowing you to take preventative measures before problems arise. 5. User Interface and Report Generation (5th Function): - Provides an interactive user interface and generates detailed reports on risk management progress, proposed strategies, and the probability of successful filing.

[0129] In conclusion, an AI-driven patent application risk management system can help inventors and patent agents proactively identify and effectively address potential pitfalls in the patent application process. By leveraging real-time risk monitoring and predictive analytics, the probability of successful patent applications can be significantly improved. The system also provides a more data-driven and insightful approach to patent strategy development and implementation.

[0130] As a variant, an information processing system can be provided that automatically generates variants of inventions and variants of embodiments from input data on inventions using a large-scale language model (LLM). This system analyzes the outline, purpose, background, and technical details of the invention, and proposes possible variants and implementation methods based on this information. The system has the following configuration. 1. Input acceptance module: - Receive detailed information about the invention (summary, technical background, objectives, specific embodiments, etc.) in text format (which may also include other data such as images). 2. Analysis module: - Analyze the input data and extract the key points of the invention, its technical features, and the problems it is trying to solve. 3. Variation generation module (using LLM): - Based on the information extracted by the analysis module, LLM is used to generate variants of the invention and variants of the embodiment, including variants that can be adapted to different uses and conditions, as well as proposals for forms that expand the technical scope. 4. Output Shaping Module: - Organize the generated variations and variations of the embodiment and output them in a format that is easy for the user to understand (text, diagrams, flowcharts, etc.). 5. Feedback Collection Module: - Gather user feedback and identify improvements to improve the accuracy and effectiveness of the system.

[0131] Such a system operates as follows. 1. Data Entry: - The user provides information about the invention to an input acceptance module. 2. Information analysis: - The input data is processed by the analysis module to extract important features and problems of the invention. 3. Automatic generation of variants: - Based on the analyzed data, the variant generation module automatically suggests potential variants and improvements to the invention. 4. Result output: - Proposed variations and embodiment variations are presented to the user through an output shaping module. 5. Feedback and Improvement: - Collect user feedback and encourage continuous improvement of the system through the feedback collection module.

[0132] This system helps inventors and researchers to broaden the scope of their inventions and explore various application possibilities, while reducing the effort required for technical documentation and maximizing the potential of their inventions.

[0133] It is also possible to create a patent document creation system that utilizes web information and past patent publications.

[0134] In this example, a system is provided that automatically generates documents required for patent applications by integrating and utilizing information on the web and past patent publications. This system collects and analyzes information to enhance the novelty and inventive step of an invention during the patent application preparation process, and generates higher quality patent documents.

[0135] The main components of the system are as follows: 1. Information collection module: - Collect information from the web, such as related technical articles, research papers, industry news, etc. Also, search and collect relevant past patent publications from patent databases around the world. 2. Analysis module: - Analyze the collected information using natural language processing (NLP) techniques to extract keywords and phrases related to the background, purpose, prior art, and details of the invention. 3. Document Generation Module: - Based on the analysis results, a large-scale language model (LLM) is used to generate a draft patent document that includes a description of the background art, differences from the prior art, a detailed description of the invention, and its effects. 4. Documentation Improvement Module: - The resulting draft is then refined under the supervision of patent legal experts to enhance the clarity, novelty, and inventive step of the document, and, if necessary, further information is gathered and analyzed. 5. Review & Feedback Module: - Provide the improved documentation to inventors and patent agents for review and feedback, which will be used to further improve the documentation.

[0136] The operation flow of the system is as follows: 1. Information Collection: - Based on the keywords and description of the invention provided by the inventor, the information gathering module collects related technical information and patent publications from the web. 2. Information analysis: - The analysis module analyzes the collected information and extracts important information related to the invention. 3. Automatic document generation: - The document generation module generates a draft document in a format suitable for patent application based on the analysis results. 4. Documentation Improvements: - The document improvement module improves the quality of the generated draft under the supervision of experts. 5. Review and Finalization: - Collect feedback from inventors and patent agents through the Review & Feedback module and finalize the document.

[0137] The above system has the following features: - Strengthening novelty and inventive step: By extensively collecting and analyzing online information and past patent publications, we provide detailed information that supports the novelty and inventive step of an invention, which is expected to reduce the rejection rate at the time of patent application. - Time and effort savings: Automated information gathering and document generation processes significantly reduce the time and effort required for inventors and patent agents to manually research and prepare documents. - Improved quality: The combination of automated generation using large-scale language models and expert review results in high-quality patent documentation, optimizing the scope of patent protection and maximizing the value of patent rights. - Flexibility and scalability: The system is adaptable to inventions in a wide range of technical fields and can be easily expanded by integrating new sources of information and analytical tools.

[0138] The above-mentioned device and system can be applied to any organization that frequently files patent applications, such as start-up companies, research institutes, and the R&D departments of large corporations. In particular, for small and medium-sized enterprises with limited resources and individual inventors, this system will be a powerful tool that simplifies the patent application process and lowers the hurdles to obtaining a patent.

[0139] In particular, the patent document preparation system that utilizes web information and past patent publications offers an innovative approach that can simultaneously improve the quality and efficiency of patent applications, accelerating technological innovation and more effectively protecting intellectual property rights, thereby further strengthening an organization's competitiveness.

[0140] For example, an information processing device creates documents for a patent application through the following process. 1. Input acceptance: Accepts input from inventors regarding their inventions through invention notification forms or specific forms, which may include information on the outline, purpose, drawings, etc. of the invention. 2. Prior art extraction: Using large-scale language models, the system extracts relevant prior art information from the input information. If extraction is not possible, it searches training data, databases, or the Internet to generate relevant prior art information. 3. Identifying the problem: Extract the problem that the invention is trying to solve from the input. If the problem cannot be clearly extracted, generate additional information to clarify the problem. 4. Solution extraction: Extract the solutions and technical features provided by the invention. If the solutions cannot be extracted directly, they are inferred and generated based on related information. 5. Effects of the invention: Extract the effects and advantages of the invention. If the effects cannot be extracted directly, generate them based on related information. 6. Invention Title: Extract or generate a suitable name for the invention. 7. Extract or generate drawings: Extract drawings from the input, or if drawings are not available, generate drawings or code to generate drawings based on information about the invention. 8. Drawing Description: Extract or generate a description for a drawing. 9. Abstract Creation: Extract or generate an abstract of the invention based on the input information. 10. Document Output: Integrate the above information and complete and output the documents required for the patent application. This process includes documents containing sections such as claims, background art, and detailed description of the invention based on the extracted information and generated content.

[0141] This example shows a series of processes that utilize technology centered on a large-scale language model to automatically create documents required for patent applications from information provided by inventors. Such a system is expected to help inventors efficiently proceed through the patent application process and accelerate preparations for obtaining a patent.

[0142] The information processing device may also provide the following additional functions and processes during the patent application preparation process: 11. Automated Feedback Loop: Based on the information entered and the document generated, the system accepts feedback from the inventor and uses it to correct and improve the document. This process allows the inventor to provide additional information for specific sections or make corrections to the suggested content. 12. Search and analysis of relevant patent literature: Using large-scale language models, we automatically search patents and literature relevant to an invention and provide the results to inventors, providing them with reference information for assessing the novelty and inventive step of the invention. 13. Automatic Claim Generation: Automatically generate patent claims based on the core features of the invention. This process proposes draft claims to define the scope of protection for the invention, which the inventor can review and modify as needed. 14. Legal Check: Automatically checks generated documents for proper format and content based on the legal requirements of a specific patent office, including document formatting, presence of required sections, and accuracy of language used. 15. Translation Service: Supports translation of generated documents in case inventors file patent applications in different countries. This function automatically translates patent application documents into different languages, facilitating the preparation of international patent applications. In addition to automatic translation, patent application documents may be converted into a written format appropriate for each country. 16. Filing Support: To further simplify the patent application process, generated documents can be output in a format suitable for online submission directly to the patent office, and calculation and payment instructions for required filing fees can also be provided.

[0143] These information processing devices are intended to help inventors navigate the complex procedures for patent applications efficiently and accurately. Compared to traditional manual processes, they are expected to significantly reduce time and costs, allowing inventors to focus more on their inventions. The introduction of these systems will help improve the quality of patent applications, expedite the process, and ultimately accelerate innovation.

[0144] Another embodiment of the apparatus for preparing patent application documents using an invention notification form as input will be described below.

[0145] The apparatus for preparing patent application documents using an invention notification form in this embodiment is implemented as a computer system that includes a CPU, RAM, ROM, a hard disk drive, a display, a keyboard, a mouse, and a network interface.

[0146] The input means is implemented as a web interface that allows users to upload invention notifications via a web browser. Invention notifications are accepted in text or PDF format.

[0147] The invention notification form analysis means analyzes the invention notification form using natural language processing technology and extracts the title of the invention, information about the inventor, an outline of the invention, a detailed description of the invention, drawings, and related prior art documents. The analysis results are saved in a structured data format.

[0148] The patent specification generator uses a GPT-3 language model pre-trained with a large amount of patent specification data. Using the extracted components of the invention notification form as input, GPT-3 generates each section of the patent specification (title of the invention, summary of the invention, detailed description of the invention, scope of the claims, and description of drawings).

[0149] The consistency assurance unit compares each section of the generated patent specification to check the consistency of the contents, and automatically corrects any inconsistencies or discrepancies between the summary and detailed description of the invention, the claims and detailed description of the invention, and the drawing descriptions and detailed description of the invention based on predefined rules.

[0150] The output means outputs the consistent patent specification in Word or PDF format, and users can download the generated patent specification via a web interface.

[0151] The feedback learning mechanism allows users to provide feedback on the generated patent specification. The feedback consists of a quality assessment of the patent specification and suggestions for improvement. The collected feedback is used to continuously improve the GPT-3 language model.

[0152] The device for creating patent application documents using an invention notification form as input in this embodiment can significantly improve the efficiency of the patent application process by automatically generating a patent specification from the invention notification form. Furthermore, by using the GPT-3 language model, the quality of the generated patent specification can be improved. Note that GPT-3 is a specific example and is not intended to be limiting.

[0153] This example describes the specific configuration and operation of a device that uses an invention notification form as input to create patent application documents. The example describes the configuration of a computer system, how each means is implemented, and the use of the GPT-3 language model. The example also emphasizes the efficiency of the patent application process and the improvement of the quality of patent specifications as benefits of the present invention.

[0154] The following may be provided as the building blocks of the invention, and may serve as the main components of a device that uses an invention notification form as input to create patent application documents. 1. Input means: Receive the invention notification form and input it into the system. 2. Invention notification analysis means: Analyzes the input invention notification and breaks it down into components. 3. Patent specification generator: Uses a language model such as GPT-3 to generate sections of the patent specification from the analyzed components. 4. Consistency assurance measures: Check the consistency between each section of the generated patent specification and make corrections if necessary. 5. Output means: Output a consistent patent specification. 6. Feedback learning: Receive user feedback and use it to continuously improve the GPT-3 language model.

[0155] These components work together to realize the process of automatically creating patent application documents from invention notification documents.

[0156] The following steps can be considered to create a program using LLM that generates a patent specification for a patent application from an invention notification form. 1. Data collection and preprocessing: - Collect a large amount of sample data on invention notifications. - Label and tokenize the collected data to match the format of patent specifications. - Split the data into training, validation, and test sets. 2. LLM Selection and Fine-Tuning: - Select an LLM suitable for patent specification generation (e.g., GPT-3, T5, etc.). - Fine-tune your LLM using the collected invention notification and patent specification pairs. - During fine-tuning, train LLMs to adapt to the style and structure of patent specifications. 3. User Interface Development: - Develop an interface that allows users to enter invention notifications. - Use LLM to generate patent specifications from input invention declarations. - Allow users to view and edit the generated patent specification. 4. Rating and Feedback: - Establish indicators to assess the quality of generated patent specifications (e.g. BLEU, ROUGE, expert evaluation, etc.). - Collecting user feedback and continuously improving the LLM based on it. 5. Legal and Ethical Considerations: - Ensure that the patent specification produced complies with the requirements of patent law. - Ensure that the patent specifications produced do not infringe the intellectual property rights of the original inventors. - Establish and adhere to ethical guidelines for the use of LLM. 6. Deployment and Maintenance: - Deploy the program in a secure and scalable manner. - Regularly monitor and evaluate the performance of LLMs and provide retraining or improvement where necessary. - Establish a user support and maintenance system.

[0157] These steps allow the development of an LLM-based program that generates patent specifications from invention notifications. However, because generating patent specifications is a complex intellectual task with significant legal implications, development efforts must involve collaboration with individuals with expertise in patent law and patent specification drafting.

[0158] A computer system that breaks down an invention notification document into its components and generates a patent specification for a patent application using LLM has the following configuration. 1. Input module: - An interface for importing invention notifications into the system. - Convert invention notifications into text format and apply OCR where necessary. - Extracts components of invention notification forms (invention title, inventor information, invention summary, detailed description, etc.). 2. Data Preprocessing Module: - Organize and structure the extracted components of the invention notification form. - Uses natural language processing techniques to clean, normalize, and tokenize text. - Prepare input data for each section of the patent specification. 3. LLM-based patent specification generation module: - Generate each section of a patent specification from input data using a pre-trained LLM. - Implement logic to maintain consistency between sections. - Combine the generated sections to create a complete patent specification. 4. Post-processing module: - Format and adjust the generated patent specifications as needed. - Checks for compliance with patent law requirements and provides warnings and suggestions. - Provides an interface that allows users to edit and modify. 5. Output module: - Output the generated patent specification in the appropriate file format (Word, PDF, HTML, XML, etc.) - Allow the user to download or print the file. 6. Feedback and Learning Module: - Gather user feedback and evaluate LLM performance. - Continuously retrain and improve your LLM based on collected feedback. 7. Data Storage and Management Module: - Securely store data such as invention notifications, generated patent specifications, and user feedback. - Provides data versioning and tracking. 8. User authentication and authorization module: - Control access to the system and manage user authentication. - Enforce access control based on user roles and permissions.

[0159] These modules work together to create a computer system that efficiently and effectively generates patent specifications from invention notifications. System development requires personnel with specialized knowledge of natural language processing, machine learning, patent law, and other areas. Furthermore, sufficient consideration must be given to data security and privacy protection when operating the system.

[0160] The prompts used when generating a patent specification from an invention notification document should correspond to each component of the invention notification document and appropriately instruct the generation of each section of the patent specification. Some examples of prompts are shown below. 1. Title of invention: - "Generate the title of the invention in the patent specification from the title of the invention below: [title of the invention in the invention notification form]" 2. Summary of the Invention: - "Generate the Summary of Invention section of a patent specification from the following Summary of Invention: [Summary of Invention from Invention Notification Form]" - "Please create a summary of the invention for the patent specification that explains the purpose, structure, and effect of the invention using the following information: [Purpose, structure, and effect of the invention in the invention notification form]" 3. Detailed Description of the Invention: - "Generate the Description of Invention section of the patent specification from the following Description of Invention: [Description of Invention from Invention Notification Form]" - "Based on the following information, please prepare a detailed description of the invention for a patent specification, including a description of the prior art, the problem, the means for solving the problem, and examples: [Description of the prior art, the problem, the means for solving the problem, and examples in the invention notification form]" 4. Claims: - "Generate claims for a patent specification from the detailed description of the invention below. [Detailed description of the invention from the invention notification form]" - "Draw up a claim that complies with the requirements of the Patent Act based on the following invention structure and effect: [structure and effect of the invention in the invention notification form]" 5. Description of the drawings: - "Generate the drawing description section of the patent specification from the following brief description of the drawing: [Brief Description of Drawing in Invention Notification Form]" - "Based on the drawings and detailed description provided, please create a drawing description for the patent specification: [Drawings and detailed description from the invention notification form]" These prompts provide guidance for translating each component of the Invention Notification into the corresponding section of the patent specification. The prompts are designed to properly convey the information in the Invention Notification to the LLM and generate the required information for each section of the patent specification.

[0161] The actual prompts will need to be tailored to the characteristics of the LLM being used and the formal requirements of the patent specification, and additional prompts and logic may be required to ensure consistency between sections of the generated patent specification.

[0162] To ensure consistency between sections of the generated patent specification, additional prompts can be used, such as: 1. Check consistency between the summary and description of the invention: - "Compare the generated Summary and Detailed Description to ensure consistency. If there are any discrepancies, revise the Summary to match the Detailed Description." - "Please confirm whether the purpose, configuration, and effects of the invention as described in the Summary of the Invention are adequately explained in the Detailed Description. If necessary, please revise the Summary of the Invention or the Detailed Description." 2. Checking the consistency of claims and description: - "Make sure that the elements of the invention described in the claims are adequately supported in the detailed description of the invention. If they are not adequately supported, please complete the detailed description or amend the claims." - "Please confirm whether the effects of the invention as described in the detailed description of the invention are properly reflected in the claims. If necessary, please amend the claims or the detailed description." 3. Check consistency between the drawing description and the detailed description of the invention: - "Please ensure that the drawing reference numbers listed in the drawing description are properly cited in the detailed description of the invention. If there is a discrepancy, please correct the drawing description or detailed description." - "Check that the examples described in the detailed description of the invention are consistent with the drawing description and drawings. If necessary, amend the detailed description, drawing description, or drawings." 4. Overall consistency check: - "Read through all sections of the generated patent specification to identify any inconsistencies or discrepancies between the title, summary, description, claims, and drawing descriptions. If any inconsistencies or discrepancies are found, correct the appropriate sections." - "Make sure that the technical problem of the invention and its solution are consistently described throughout the patent specification. If there are any inconsistencies, revise the relevant sections." These additional prompts provide guidance for comparing sections of LLM-generated patent specifications to ensure consistency of content. The prompts are designed to highlight the relationships between sections and identify and correct any inconsistencies or discrepancies.

[0163] The actual prompts can be tailored to include more specific instructions depending on the type and technical field of the patent specification, and the consistency check process can be repeated multiple times to improve the overall quality of the patent specification.

[0164] In addition, the following configurations of the invention are also possible. 1. A device for creating patent application documents using an invention notification form as input, - input means for receiving an invention notification form as input; - invention notification analysis means for decomposing said invention notification into constituent elements; - a patent specification generation means for generating each section of the patent specification using a language model based on the data consisting of the components; - a consistency assurance means for checking consistency between sections of the generated patent specification and correcting it if necessary; - an output means for outputting the patent specification with the consistency ensured; 1. An apparatus for preparing patent application documents using an invention notification form as an input, comprising: 2. In the above device, The invention notification document analysis means extracts the name of the invention, inventor information, an outline of the invention, a detailed description of the invention, drawings, and related prior art documents from the invention notification document, and is a device for creating patent application documents using an invention notification document as input. 3. In the above device, An apparatus for creating patent application documents using an invention notification form as input, characterized in that the patent specification generation means generates sections including the title of the invention, an outline of the invention, a detailed description of the invention, claims, and descriptions of drawings. 4. In the above device, The consistency assurance means is a device for creating patent application documents using an invention notification form as input, characterized in that it checks the consistency between the outline of the invention and the detailed description, the consistency between the claims and the detailed description of the invention, the consistency between the description of the drawings and the detailed description of the invention, and the consistency of the entire patent specification, and makes corrections as necessary. 5. In the above device, 1. An apparatus for creating patent application documents using an invention notification document as input, wherein the language model is pre-trained to match the style and structure of a patent specification. 6. In the above device, 1. An apparatus for preparing patent application documents using an invention notification document as an input, further comprising: a feedback learning means for receiving feedback from a user and continuously improving the language model.

[0165] In the above configuration, the main features of the device that creates patent application documents using an invention notification form as input are captured in the independent claims, and the details of its components and processing are described in the dependent claims. When asserting the patentability of this invention, it is possible to emphasize the efficiency gained by automatically generating a patent specification from an invention notification form and the improvement in the quality of the patent specification by using a language model.

[0166] The format of the invention notification form varies depending on the company, but it generally includes the following items: 1. Title of the invention 2. Inventor information (name, department, contact information, etc.) 3. Summary of the Invention - Object of the invention - Structure of the invention - Effect of the invention 4. Detailed Description of the Invention - Description of conventional technology and issues - Means to solve the problem - Example - Brief description of the drawing 5. Claims 6. Drawings 7. Desired country / region of application 8. Related Prior Art Documents 9. Date of completion and notification of invention 10. Signature section (inventor, department head, etc.) It is common to summarize these items concisely on about 4 to 5 pages of A4 paper. Some companies may have their own formats or templates. This type of information can be input and converted into a statement in this embodiment. Of these, 3. Summary of the Invention - Object of the invention - Structure of the invention - Effect of the invention 4. Detailed Description of the Invention - Description of conventional technology and issues - Means to solve the problem - Example - Brief description of the drawing 5. Claims 6. Drawings 7. Desired country / region of application 8. Related Prior Art Documents At least one of the above items should be left to the device, and the invention notification form may be left blank, or the user may select "Leave it to the device." When leaving it to the device, the device will search the web or a database if necessary from the given information to obtain related information and automatically generate it.

[0167] The invention notification form does not need to be divided into these categories. In this case, the LLM will automatically obtain the data corresponding to the categories from the input.

[0168] Prior art documents may be searched by the device, and the device may automatically determine patent classifications such as IPC.

[0169] Drawings for patent application documents may be created using the following systems:

[0170] FIG. 3 is a block diagram showing a system configuration in one embodiment of the program of the present invention.

[0171] This system includes an input unit 10 for inputting claims (or even just claims), a conversion unit 20 for inputting the input claims as prompts into the LLM, analyzing them, and converting them into code, a drawing unit 30 for generating drawings from the converted code, and a display unit 40 for displaying the generated drawings. Some of these components may be executed by an external computer. In other words, the processing of this embodiment may be distributed across multiple computers.

[0172] The input unit 10 for inputting claims may input a prompt sentence for generating an image (or a code for generating an image) along with the claims. The input unit 10 may also automatically add an appropriate prompt sentence to the input claims, and send the input claims to the conversion unit 20, which analyzes the input claims using an LLM and converts them into code. The claims may also be input as a prompt and sent to the conversion unit 20.

[0173] The conversion unit 20 may generate not only the code for creating the drawing but also text to explain the drawing. In this way, embodiments that cover the scope of the claims can be automatically created. The conversion unit 20 may also directly generate image data instead of the code for creating the drawing.

[0174] The conversion unit 20 may write the names of the elements of the claims within the blocks (boxes represented by solid or dotted lines) in the drawing, or may also write descriptions of those elements. Only descriptions may be written. Descriptions other than the elements may be written outside the blocks. If elements (tangible or intangible) are exchanged between blocks, the flow may be represented by arrows. The names of the elements in that flow may be written next to the arrows.

[0175] If there are main components (for example, the hardware configuration of a device) and sub-components (such as signals exchanged between hardware), the representation of the blocks may be changed to make it clear whether they are main or sub. For example, this may be done by changing the line type. Components may also be color-coded by type. Main components may be drawn as software modules, and sub-components may be drawn as information exchanged between modules. Blocks may be nested, with blocks written within blocks. Explanations may also be written within blocks. The top-level configuration may be represented by a block, and the configurations within it may be illustrated as blocks within that block.

[0176] When the entered claims are written in a first language, the drawings may be written in a second language (machine translation may be performed).

[0177] The conversion unit 20 may process only independent claims, only claim 1, or dependent claims as well. Independent and dependent claims may be represented on a single sheet. The dependent relationships may be expressed in a way that makes them clear (for example, the dependent relationships between claims represented by each block may be expressed by connecting the blocks with arrows pointing from the dependent to the dependent or vice versa, or by lines connecting the two). Each claim may be defined as a subgraph, and the dependent relationships between them may be indicated by arrows. When drawing a process such as a flowchart based on the scope of a claim, normal processes may be drawn with rectangular blocks, and conditional branches may be represented with diamonds. The configuration and means of a device or system may be represented with rectangular blocks, and the processes performed within them may be represented as flows within those blocks.

[0178] The procedure for creating drawings according to this embodiment will be described below. First, a user inputs claims through the input unit 10. These may be input automatically from the Internet, a repository, a database, or the like. The input claims are sent (as prompts or together with other prompts) to the conversion unit 20, where they are linguistically analyzed by the LLM. The LLM analyzes the claim text and extracts the elements of the invention and their relationships. The extracted information is then converted into code for creating drawings. This code defines the type of drawing, the layout of the elements, the connections, and so on. The conversion unit 20 may be an external computer connected via the Internet or the like.

[0179] The code generated by the conversion unit 20 is passed to the drawing unit 30, which then generates the actual diagram. The drawing unit 30 uses the information contained in the code to draw a block diagram or flowchart. The layout of components and the drawing of connecting lines are automatically optimized. Arrows are also included on the diagram to represent process flow, time series, and signal flow. The generated diagram can be output in various formats, including vector graphics, bitmap, JPEG, and GIF. It can also be output as a script in Mermaid or Dot language. This makes it easy to use the generated diagram in other software or embed it in a web page.

[0180] The generated drawing is sent to the display unit 40 and presented to the user. The user can then modify the drawing as necessary to obtain the final drawing. The drawing can be modified using the editing functions provided in the display unit 40. For example, it is possible to change the position and size of blocks, add or delete connecting lines, edit text, etc. It is also possible to convert the drawing format or output it to other software.

[0181] FIG. 4 is a flowchart showing the processing flow in this embodiment. First, claims are input (S10). Next, the input claims are analyzed by LLM and converted into code for creating drawings (S20). Based on this code, a block diagram or flowchart is drawn (S30). When drawing, the drawing format (vector graphics, bitmap, JPEG, GIF, etc.) and output format (image file, Mermaid notation, Dot language, etc.) are specified (S35). Arrows are also drawn on the drawing to represent process flow, time series, and signal flow. Finally, the generated drawing is displayed (S40), and the user can make modifications as needed (S50). FIG. 2 shows arrows representing the time flow between each process.

[0182] This embodiment makes it possible to largely automate the process of creating drawings from patent claims, significantly reducing the time and effort required to prepare a patent application. Furthermore, by using linguistic analysis by LLM, it becomes possible to more accurately extract the structure of an invention from the description of claims. Furthermore, the generated drawings can be output in various formats, making it easy to integrate with other software.

[0183] The LLM algorithm and drawing generation method can be replaced with other methods. The system configuration can also take a form other than that shown in Figure 1. For example, the input unit 10 and the display unit 40 can be integrated to simplify the user interface.

[0184] Conversely to the above explanation, an image or a code describing an image may be input and converted by the LLM into character data for explaining the claims or embodiments.

[0185] A computer is used to implement the present invention. Specific examples of computers include personal computers (desktop, laptop), smartphones, tablets, servers, game consoles, smart watches, home appliances (smart TVs, smart refrigerators, etc.), and control systems. Here, for example, a personal computer, smartphone, tablet, etc. is used as a client computer (client). The client accesses the server using a web browser and sends information, which the server processes. The processed information (HTML document, JSON format data, etc.) is sent to the client, and the information is displayed on the client's web browser. Furthermore, processing may be performed only within a single computer, and there may be no need to send or receive information to or from an external device.

[0186] The processed information may be sent to the client via e-mail or messenger software and displayed there. Information may also be sent from the client to the server via e-mail or messenger software.

[0187] The server and the client are connected to the Internet. The server sends data to the client that sent the request. Both the server and the client are computers and have the following components:

[0188] Central Processing Unit (CPU): Executes program instructions and processes data. A CPU has multiple cores, each capable of processing tasks independently.

[0189] Memory (RAM): A high-speed storage device that temporarily stores programs and data while the computer is running. The CPU directly accesses it and reads and writes data. RAM is volatile memory, meaning that data is lost when the power is turned off.

[0190] Storage Device: A device that provides long-term data storage, such as a hard disk drive (HDD) or solid-state drive (SSD). These store the operating system, applications, user data, etc.

[0191] Motherboard: A board that connects all hardware components and provides power and data communication. The motherboard contains the CPU socket, RAM slots, expansion slots, I / O ports, etc.

[0192] Graphics Processing Unit (GPU): A specialized processor for graphics and image processing. GPUs are used for 3D graphics rendering, video decoding / encoding, machine learning tasks, etc.

[0193] Power Supply Unit (PSU): A device that provides power to a computer's hardware components. A PSU converts AC power into the DC voltage required by the computer.

[0194] Cooling system: A system that manages a computer's temperature and prevents damage from overheating. Cooling systems include heat sinks, fans, and liquid coolers.

[0195] Input devices: These are devices that allow a user to enter information into a computer, such as a keyboard, mouse, touchpad, or touchscreen.

[0196] Output devices: Monitors, printers, speakers, etc. are devices that communicate information from a computer to a user. A monitor displays images and text, a printer prints documents, and speakers output sound. Any output device can be used to communicate information from a computer to a user.

[0197] Network Interface: An interface that connects a computer to a network, such as an Ethernet port or Wi-Fi adapter, allowing access to the Internet or sharing resources on a local network.

[0198] Expansion Card: A card that plugs into a motherboard's expansion slot to provide additional functionality or performance, such as a graphics card, sound card, or network card.

[0199] Optical drive: A device for reading and writing optical media such as CDs, DVDs, and Blu-ray discs. In computers, the optical drive is sometimes omitted.

[0200] Case: An enclosure that protects and houses the hardware components of a computer. The case provides access to the hardware, supports the cooling system, and influences the design and shape of the computer system.

[0201] These components work together to make the whole computer system function: when you use your computer, these components work together to process data, perform tasks, and display or output information.

[0202] Below we list some of the software that is important for running a computer and explain the functions and operations of each.

[0203] Operating System (OS): The underlying software that manages a computer's hardware and software resources and allows users and applications to access them. Typical operating systems include Windows, macOS, and Linux.

[0204] Device driver: Software that is responsible for communication between the operating system and the hardware devices and peripherals in a computer. Device drivers are necessary for the correct operation of hardware such as keyboards, mice, printers, and graphics cards.

[0205] System software: Software that supports the basic functions of a computer, such as file management, system settings, disk management, and network connections. Examples include File Explorer and Disk Utility.

[0206] Security software: Software that protects your computer from security threats. This includes antivirus software, firewalls, and anti-malware tools.

[0207] Web browser: Software that displays web pages on the Internet and allows users to browse and search for online information. Popular web browsers include Google Chrome, Mozilla Firefox, and Microsoft Edge.

[0208] Utility software: A group of software for performing tasks related to improving productivity such as document creation, spreadsheet creation, and presentation creation. This also includes software that performs the processes for implementing the present invention.

[0209] Communications software: Software used for computer-based communication, such as email clients, instant messengers, and video conferencing tools.

[0210] Multimedia software: Software for playing, editing, and creating multimedia content such as audio, images, and video.

[0211] Backup software: Software that regularly backs up data on your computer, reducing the risk of data loss or system failure.

[0212] Development tools: Tools used to develop software and applications using programming languages ​​and development frameworks. These include integrated development environments (IDEs), text editors, and version control systems.

[0213] These software programs provide the basic functions required for a computer to operate and help users perform various tasks efficiently. Each software program is designed for a specific purpose and works together to improve the overall functionality of the system.

[0214] The process by which the server returns data in response to a client request is as follows. HTTP, HTTPS, etc. are used as communication protocols.

[0215] The client (usually a web browser) sends an HTTP request to the server by specifying the URL and HTTP method (GET, POST, etc.). The server analyzes the received HTTP request and processes it according to the content of the request.

[0216] If the required data or resources are available on the server side, the server retrieves the data from a database, file system, etc. The server-side program (PHP, Python, Ruby, etc.) processes the retrieved data or resources and generates a result. The server creates an HTTP response to return the generated result (HTML, JSON, XML, etc.) to the client. At this time, a status code (200 OK, 404 Not Found, etc.) and header information are also set. The server then sends the created HTTP response to the client.

[0217] The client analyzes the received HTTP response and displays or processes it in the appropriate format. For example, a web browser displays HTML, and JavaScript processes JSON data.

[0218] Through these processes, the server returns data in response to an HTTP request from the client. The protocol used for this exchange is HTTP (HyperText Transfer Protocol), which makes it possible to exchange information over the web. As mentioned above, messenger software or email can also be used for communication between the client and server.

[0219] It should be noted that instead of displaying the data on a web browser, the data may be displayed on the client terminal via email or messenger software.

[0220] In this embodiment, a computer program and an information processing device are provided that automatically create drawings showing the structure of a patent from the scope of claims using a computer. Creating drawings that visually express the structure of an invention is an important task, but doing it manually takes time and effort. Therefore, the computer program and information processing device in this embodiment analyze the scope of claims using language processing technology and automatically generate drawings, thereby significantly streamlining this task.

[0221] The program and information processing device have two main components. The first component is a conversion component that analyzes input claims using large-scale language models (LLMs) and converts them into code for creating drawings. This process extracts the elements of the invention and their relationships from the claims. LLMs can be deep learning models specialized for natural language processing, such as GPT-3, BERT, XLNet, and RoBERTa. These models are pre-trained with large amounts of text data and have excellent capabilities for understanding context and extracting meaningful information. The conversion component uses LLMs to analyze the sentence structure of the claims and identify subjects (main elements), predicates (relationships between elements), and objects (subordinate elements). This information is then used to output program code (e.g., SVG code) that generates vector data for computer graphics.

[0222] The second means is a drawing means that illustrates the claims in the form of a block diagram or flowchart in accordance with the code obtained by the conversion means. The converted code includes information such as the type of diagram (block diagram or flowchart), the layout of each component, and their connections (although at least some of this information may be omitted). Based on this information, the drawing means automatically generates a diagram that appropriately expresses the content of the claims. At this time, arrows are added to the diagram to represent the process flow, time series, and signal flow. A specific implementation of the drawing means could utilize visualization libraries such as Python's Matplotlib or JavaScript's D3.js. These libraries provide a wide range of functions for drawing shapes from vector data.

[0223] As described above, the computer program and information processing device of this embodiment combine language processing and graphic drawing technologies to automatically generate drawings from patent claims. The conversion means uses advanced natural language analysis using LLM to accurately understand the content of claims without human intervention and extract the necessary and sufficient information for drawing. Furthermore, the drawing means applies computer graphics technology to allow the program to autonomously create easy-to-read and understand drawings based on the extracted information. This significantly reduces the workload of drawing creation, which has previously been done manually, and contributes to the efficiency of patent application procedures.

[0224] It is also possible to process claims directly to create image data in various formats, such as vector graphics, bitmaps, JPEG, GIF, and SVG, without converting them into drawing code. In this case, the results of language analysis by the LLM are directly input into a drawing library or data conversion library to create the drawing. For example, by implementing a program that converts the output of the LLM into SVG or PNG image data, drawings can be generated more directly.

[0225] Furthermore, this invention can be used not only to create drawings required for patent applications, but also to illustrate patent publications and patent claims. In other words, by retrieving published patent documents from a database and analyzing the claims therein to create drawings, it is possible to visually express the contents of existing patents in an easy-to-understand manner. This is expected to improve the efficiency of patent searches and prior art searches. (Example of display) FIG. 5 is a diagram showing the results of drawing based on code written in the Dot language output from a claim of a patent by a computer program according to this embodiment.

[0226] Here, the results of drawing based on the description of claim 1 of the claims are displayed. Claim 1 states, "An information processing device that uses a computer to create drawings showing the configuration of a patent claim, an input means for inputting claims; an acquisition means for processing the input claims through a large-scale language model (LLM) to acquire code written in a language for creating drawings illustrating the input claims; a drawing means for drawing a block diagram or a flowchart illustrating the input claims in accordance with the code obtained by the obtaining means, When a component is included in the input claim, the acquiring means draws the component as a block. Assume that the following is defined. The elements of this claim are input means, acquisition means, and drawing means, which are illustrated as blocks. Each block describes the steps of the process executed in that block in the form of a flowchart. The acquisition means includes a statement indicating its configuration, such as "an acquisition means for processing the input claim through a large-scale language model (LLM) and acquiring code written in a language for creating a drawing illustrating the input claim," and a statement indicating the operation and function of the configuration, such as "when the input claim includes a component, the acquisition means draws the component as a block." In this case, only the former may be described within the block for the "acquisition means," and the latter may be described outside the block, or both may be described within a box as shown in Figure 5. The former and latter processes may be described as separate flowcharts, or as a single flowchart as shown in Figure 5.

[0227] Also, it is possible to describe only the flow chart portion of FIG. 5 and not draw the blocks.

[0228] FIG. 6 is a diagram showing the results of drawing based on code written in the Mermaid language output from the description of a claim by the computer program of this embodiment.

[0229] In addition to the description in Figure 5, the preamble of the claim ("An information processing device that uses a computer to create drawings showing the configuration from the claims") is also included in the illustration.

[0230] 7 to 9 are diagrams showing the results of drawing based on code written in Mermaid language output from a claim by a computer program in this embodiment. Figures 7 to 9 are diagrams that have been divided from a single figure for ease of viewing, and were originally drawn on a single sheet.

[0231] This diagram shows the results of analyzing the scope of claims, including claims 1 to 5. Each claim is shown as a block representing the highest concept. The dependent relationships between claims are indicated by arrows.

[0232] Furthermore, the elements included in each claim are described as blocks within the block representing that claim. Here, only the name of the block is written within the block representing the element. An explanation of the function, operation, etc. of each block is written near the block (or near the arrow leading from the block). The relationship between blocks representing elements is expressed by arrows. For example, if a signal or information is sent from block A to block B, an arrow is drawn from block A to block B, and the name or explanation of the element (signal, information, etc.) being sent is written overlapping or near the arrow.

[0233] As shown in the figure, the relationship between blocks representing elements within one claim is represented by an arrow, and the relationship between blocks representing elements within different claims is also represented by an arrow. Lines may be used instead of arrows. Lines may be straight, curved, or dotted.

[0234] Furthermore, a lead line may be added from the block, and an explanation of that block may be written. The specification (such as the embodiment of the invention) may also be analyzed using LLM along with the claims, and the terms of the embodiment corresponding to the claims, the position where they appear (paragraph number, page, line), etc. may be displayed. The terms of the embodiment corresponding to the elements in the claims, the position where they appear (paragraph number, page, line), etc. may also be displayed. These may be written within the block of the claim or element, or may be written nearby. They may also be displayed in a table format separate from the figures.

[0235] The claim blocks may be drawn in order starting from claim 1, or the order may be automatically determined for layout reasons or other reasons to make the relationships easier to understand as shown in the drawing.

[0236] 10 and 11 are diagrams showing the results of drawing based on code written in the Dot language output from a claim by a computer program according to this embodiment. Figures 10 and 11 are diagrams that have been divided into parts for ease of viewing, and were originally drawn on a single sheet.

[0237] Here, the shape of the blocks is changed depending on the type of element. Also, arrows indicating the dependency of claims do not need to be drawn (for example, when the dependency can be understood by describing the relationship between elements across claims). (Variation 1) In the above embodiment, we have described a method for generating drawings from claims by analyzing the claim text using LLM and generating code for creating drawings. In Variation 1, rule-based natural language processing technology is used instead of LLM. Specifically, a syntactic analysis of the claim text is performed, and rules are defined for extracting key components and their relationships from the resulting syntax tree. These rules are created manually in advance based on knowledge about how claims are written. While rule-based methods are less versatile than LLMs, they are considered suitable for claim writing because they can perform analysis specialized for specific sentence structures. (Variation 2) In the above embodiment, when generating drawings from claims, they are expressed in the form of block diagrams or flowcharts. In Variation 2, other types of drawings, such as circuit diagrams and sequence diagrams, can also be generated. Circuit diagrams are used to represent the configuration of electrical and electronic circuits, and are suitable when the claims describe the components of a circuit and their interconnections. Sequence diagrams, on the other hand, are used to represent the chronological interactions between multiple components, and are useful when the claims describe the operation of a communication system or software. Expanding the types of drawings allows patents in a wider range of technical fields to be accommodated. (Variation 3) In the above embodiment, drawings are generated from a single claim. In Variation 3, drawings are generated by combining multiple claims. Claims may consist of independent claims and dependent claims. Dependent claims cite the independent claims and add further limitations. Therefore, by analyzing the independent claims and dependent claims together, it is possible to generate drawings that express the overall structure of the invention. In this case, one possible method is to create a basic drawing from the independent claims and then add the content of the dependent claims to further refine the drawing.

[0238] The above describes modified examples of the present invention. Modification 1 proposes a method of using rule-based natural language processing technology instead of LLM. Modification 2 shows a method of responding to patents in various technical fields by expanding the types of drawings to be generated. Modification 3 explains a method of combining multiple claims to create drawings. These modifications further develop the automatic drawing generation technology in the above embodiment, and are expected to contribute to improving the efficiency of patent application work.

[0239] The invention can also be limited by the following points. (Limitation 1) When analyzing patent claims, we use natural language processing and take into account terminology and expression patterns specific to the technical field to which the invention pertains, enabling us to perform a more accurate analysis. This allows us to more accurately extract the characteristics of the invention in that technical field and reflect them in the drawings. (Limitation 2) When creating diagrams, to express the relationships between the components of an invention, rather than just using simple arrows, arrows of different shapes and colors are used depending on the type of relationship. For example, by distinguishing between arrows that indicate data flow and arrows that indicate control flow, the operation of the invention can be expressed in more detail. (Limitation 3) To provide an interface that allows users to modify and adjust drawings automatically. Specifically, it provides a function that allows users to change the layout of components in the generated drawings or add new elements. This enables semi-automatic drawing creation that includes human judgment, rather than completely automatic generation. (Limitation 4) The effects of the invention described in the claims are added as text information in the drawings. Because the effects of the invention may not be directly apparent from the claims, they are added to the drawings while also referencing the description in the specification. This allows the technical significance of the invention to be more clearly shown. (Limitation 5) The system will have a function to check the consistency of drawings generated based on the claims and descriptions in the specification by comparing them. Specifically, it will check whether the components and relationships contained in the drawings deviate from the descriptions in the specification, and will output a warning if any deviations are found. This will prevent inconsistencies between the claims and the description in the specification. (Limitation 6) The numerical ranges stated in the claims should be reflected in the drawings. For example, if the claim states that the temperature is in the range of 50°C to 100°C, this temperature range should be clearly indicated in the corresponding part of the drawings. This will allow the technical features of the invention to be expressed in more detail.

[0240] We have proposed six points for limiting an invention. These points clarify the technical features of an invention from various perspectives, such as how to analyze the scope of the patent claims, how to express the drawings, the user interface, and checking consistency with the specification. By combining these points, it is expected that stronger patent rights can be obtained.

[0241] The following configurations of the invention are also possible. (Example 1: Generating claims using LLM) It is also possible to generate patent claims by converting the invention structure extracted from drawings into text data and then inputting it into an LLM. Large-scale language models such as GPT-3 and T5 can be used as LLMs. These models learn the sentence structure and expression patterns of patent claims by pre-training with large amounts of patent document data. Therefore, when text data representing the invention structure is input into an LLM, it can be automatically verbalized according to the claim format. Fine-tuning the LLM in this process enables more appropriate sentence generation. (Example 2: Generating claims using ontology) Each technical field has its own unique terminology and expressions used in patent claims. Therefore, one method is to build an ontology that systematizes knowledge in each technical field in advance and use it to generate claims. The ontology describes the key concepts and terms in that technical field, as well as their relationships. By mapping the components of the invention extracted from drawings to this ontology, appropriate terminology can be selected and claims written in a manner appropriate to the technical field. Using an ontology also makes it possible to add more detailed invention limitations to claims. (Example 3: Rule-based claim generation) There is a rule-based method for defining sentence structures and expression patterns when generating claims from drawings. For example, rules such as "element A in the drawing should be described as the main element of the claim" or "if element A and element B are connected, use the expression 'has a structure in which A and B are connected'" are defined in advance. These rules are then applied based on the analysis of the drawings to generate the claim text. Rule-based methods are less flexible than LLM, but have the advantage of being able to strictly control how claims are written. (Example 4: Creating Multiple Claims) When the amount of information contained in the drawings is large, one method is to gradually limit the structure of the invention by generating multiple claims. Specifically, first, extract the main components of the invention from the entire drawing and describe them as independent claims. Next, focus on the detailed structure and operation in the drawing and describe them as limitations in dependent claims. At this time, to properly express the relationship between the independent claim and the dependent claim, use expressions such as "the above" to link them. By generating multiple claims, the structure of the invention can be protected from multiple angles. (Example 5: Creating claims by combining drawings and specifications) There is a method for generating more complete patent claims by utilizing the contents of the specification in addition to the drawings. First, the basic structure of the invention is extracted from the drawings and used as the outline of the claims. Next, detailed descriptions and effects of the invention not shown in the drawings are extracted from the specification and added to the claims. LLM and rule-based methods can be applied to analyzing the specification. By combining the information from the drawings and specification, it is possible to incorporate more essential features of the invention into the claims.

[0242] We have proposed five specific methods for automatically generating patent claims from drawings: a method that utilizes LLM, a method that uses ontology, a rule-based method, a method that generates multiple claims, and a method that combines drawings and specifications. By appropriately selecting and combining these methods, we believe it will be possible to automatically generate patent claims with greater accuracy.

[0243] The configurations of other embodiments will be described below.

[0244] In this embodiment, a system is provided that automatically generates embodiments based on the configuration of the invention described in the claims. Embodiments in a patent specification are descriptions that specifically explain the invention described in the claims. However, refining the content of the claims and providing clear and sufficient explanations requires advanced intellectual work. The system in this embodiment analyzes the description of the claims using natural language processing technology and automatically generates embodiment sentences based on the results, thereby significantly streamlining this work.

[0245] This system is broadly composed of three functional blocks. The first is an analysis unit that analyzes the scope of claims and extracts the elements of the invention and their relationships. The second is a generation unit that generates a sentence for an embodiment based on the analysis results. The third is a proofreading unit that evaluates the quality of the generated sentence and makes corrections as necessary. Each functional block is explained in detail below. (Configuration of the analysis unit) The analysis unit contains multiple modules that receive the claim text as input and analyze it using language processing techniques. First, the morphological analysis module divides the claim text into words and identifies the part of speech and inflected forms of each word. Next, the syntactic analysis module analyzes the dependency relationships between words to clarify the syntactic structure of the text. Finally, the semantic analysis module extracts the components of the invention and their relationships based on the syntactic structure. Using LLM in this process enables advanced semantic understanding. (Configuration of the generation unit) The generation unit has the function of converting the information about the configuration of the invention obtained by the analysis unit into a description of the embodiment. Specifically, by combining a template-based generation module and a neural network-based generation module, it generates more natural and readable text. The template-based generation module generates a basic description of the embodiment by applying the components of the invention to a pre-prepared text template. On the other hand, the neural network-based generation module uses LLM to generate text that explains the embodiment in more detailed and flexible terms. By appropriately combining the outputs of both modules, a description of the embodiment that conforms to the format of a patent specification is automatically created. (Configuration of the calibration section) The proofreading unit has the function of evaluating the generated text of an embodiment and improving its quality. Specifically, it consists of a proofreading module that points out formal errors in the text based on the description rules of patent specifications, and a proofreading module that points out semantic errors in the text based on the technical content of the invention. The former determines whether the text complies with the description requirements under the Patent Act and suggests corrections as necessary. The latter determines whether the relationships between the components of the invention are properly explained and provides additional explanations if there are any unclear points. The proofreading unit's functions can improve the quality of the text of an automatically generated embodiment.

[0246] As described above, the system in this embodiment automates a series of processes, including analyzing claims, generating embodiments, and proofreading text. The analysis unit applies various natural language processing techniques to extract detailed information about the configuration of the invention. The generation unit combines template-based and neural network-based techniques to efficiently create descriptions that conform to the format of a patent specification. The proofreading unit checks the quality of the text from both the requirements of the Patent Act and the technical content of the invention, thereby improving the completeness of the automatically generated text. By linking these functional blocks, a system for automatically generating embodiments of an invention from claims is realized.

[0247] By using this system, it is possible to significantly reduce the human burden involved in preparing patent specifications and improve the efficiency of the patent application process. In addition, by automatically generating an embodiment that accurately describes the technical content of an invention, it is expected to contribute to improving the quality of patent rights.

[0248] Next, a computer program for automatically generating claims from embodiments of the invention, and the configuration of an information processing device will be described in detail.

[0249] In this embodiment, a system is provided that automatically generates patent claims from embodiments of an invention described in a patent specification. The claims are important details for determining the scope of patent protection, and must limit the structure of the invention as necessary and sufficient. However, extracting the essential features of an invention from the description of the embodiments and creating appropriate claims requires advanced intellectual work. The system in this embodiment analyzes the description of the embodiments using natural language processing technology and automatically generates the text of the claims based on the results, thereby significantly streamlining this process.

[0250] This system is broadly composed of three functional blocks. The first is an analysis unit that analyzes the description of the embodiment and extracts the components of the invention and their relationships. The second is a generation unit that generates the text of the claims based on the analysis results. The third is a proofreading unit that evaluates the quality of the generated text and makes corrections as necessary. Each functional block is explained in detail below. (Configuration of the analysis section) The analysis unit includes multiple modules that receive the sentence of the embodiment as input and analyze it using language processing techniques. First, the morphological analysis module divides the sentence into words and identifies the part of speech and inflected forms of each word. Next, the syntactic analysis module analyzes the dependency relationships between words to clarify the syntactic structure of the sentence. Furthermore, the semantic analysis module extracts the components of the invention and their relationships based on the syntactic structure. In this process, the use of LLM enables advanced semantic understanding. Furthermore, since an embodiment may include multiple examples, the essential features of the invention are extracted by comparative analysis of these examples. (Configuration of the generation unit) The generation unit has the function of converting the information about the invention structure obtained by the analysis unit into the text of patent claims. Specifically, it consists of a module that generates independent claims and a module that generates dependent claims. In generating independent claims, the essential elements of the invention extracted from the embodiments are written in a predetermined format. In this process, LLM is used to learn the sentence structure and expression patterns of claims, allowing for the generation of more natural and appropriate text. Meanwhile, in generating dependent claims, in addition to the structure of the independent claims, desirable structures and effects of the invention described in the embodiments are added as limitations. This automatically creates a group of claims that gradually limit the structure of the invention. (Configuration of the calibration section) The proofreading unit has the function of evaluating the generated claim text and improving its quality. Specifically, it consists of a proofreading module that points out formal errors in the claims based on the description requirements of the Patent Act, and a proofreading module that points out substantive errors in the claims based on the technical content of the invention. The former determines whether the claim description format complies with the Patent Act Enforcement Regulations and suggests corrections as necessary. The latter determines whether the content of the claim deviates from the description of the embodiment and suggests corrections if there is an inconsistency. The proofreading unit's functions can improve the quality of the automatically generated claim text.

[0251] As described above, the system in this embodiment automates a series of processes, including analyzing embodiments of the invention, generating claims, and proofreading text. The analysis unit applies various natural language processing techniques to extract detailed information about the configuration of the invention. The generation unit efficiently creates a group of claims that progressively limit the essential and preferred configurations of the invention by appropriately combining independent claims and dependent claims. The proofreading unit checks the quality of the text from both the requirements of the Patent Act and the technical content of the invention, thereby improving the completeness of the automatically generated text. By linking these functional blocks, a system for automatically generating claims from embodiments of the invention is realized.

[0252] By using this system, it is possible to significantly reduce the human burden involved in preparing patent specifications and improve the efficiency of the patent application process. Furthermore, by automatically generating patent claims that adequately limit the essential and preferred configurations of an invention, it is expected to contribute to improving the quality of patent rights.

[0253] Furthermore, a modified example of processing with LLM will be described with reference to past patent publications. (Variation 1: Learning a sentence generation model using past patent publications) This modification aims to generate more natural and appropriate sentences by using sentences from past patent publications as training data when generating sentences for patent claims and embodiments of inventions. Specifically, correspondences between patent claims and embodiments of inventions are extracted from a large number of patent publications and input into the LLM as training data. The LLM then learns the sentence structure and expression patterns of patent claims and embodiments of inventions from this training data. This enables the generation of sentences that conform to the writing styles frequently found in patent publications. Furthermore, by selectively training patent publications related to a specific technical field, it becomes possible to generate sentences that appropriately use terms and expressions specific to that technical field. (Variation 2: Extraction of elements of inventions using past patent publications) In this modified version, when generating claims from embodiments of an invention, knowledge of invention elements extracted from past patent publications is utilized to generate more appropriate claims. Specifically, the system extracts the elements of inventions described in claims and their relationships between superordinate and subordinate concepts from a large number of patent publications, and builds a database of invention elements. Then, by referencing this database when analyzing embodiments, the system extracts the elements of inventions more accurately. Furthermore, when generating claims, the system combines the extracted elements based on the relationships between superordinate and subordinate concepts to create claims that appropriately limit the essential and preferred elements of the invention. This allows the system to utilize past knowledge to generate appropriate claims even from embodiments in which the elements of the invention are insufficiently described. (Variation 3: Determining claim format using past patent publications) In this variant, when generating embodiments of inventions from patent claims, knowledge of claim writing styles learned from past patent publications is utilized to generate more appropriate embodiments. Specifically, an LLM that has learned the relationship between claim writing styles and their appropriateness from a large number of patent publications is used to judge the quality of the generated claims. Then, based on the judgment results, explanatory text that conforms to the claim writing style is automatically added when generating embodiments. For example, if a claim uses the expression "comprises...", a sentence that specifically explains the configuration is added to the embodiment. This makes it possible to efficiently generate embodiments that are consistent with the content of the claims.

[0254] We have proposed three variations of LLM processing that utilize past patent publications. Variation 1 uses patent publications as training data, enabling more natural and appropriate sentence generation. Variation 2 extracts knowledge of the components of an invention from patent publications and uses this knowledge to generate claims. Variation 3 learns knowledge of the claim writing format from patent publications and uses this knowledge to generate embodiments. All of these variations aim to improve the accuracy of automatic generation of patent specifications by effectively utilizing past knowledge.

[0255] Furthermore, we explain the configuration of a computer that uses LLM to check patent specifications and claims before filing.

[0256] In this embodiment, a system is provided that automatically checks the content of patent specifications and claims before filing a patent application. This system uses LLM to analyze the text of the specification and claims and evaluate their consistency and appropriateness. The purpose of this is to improve the quality of the specification and claims before filing and increase the likelihood of patent rights being granted.

[0257] This system is broadly composed of four functional blocks. The first is the input unit that receives the text of the patent specification and claims as input. The second is the analysis unit that analyzes the input text and extracts the components of the invention and their relationships. The third is the evaluation unit that evaluates the consistency and appropriateness of the specification and claims based on the analysis results. The fourth is the output unit that presents the evaluation results to the user and suggests modifications as necessary. Each functional block is explained in detail below. (Input section configuration) The input section provides an interface for receiving the text data of the patent specification and claims created by the user. Specifically, it has a screen that allows users to upload files and copy and paste text. It also allows users to input the specification and claims separately, making it easy to distinguish between the two texts. (Configuration of the analysis section) The analysis unit has the functionality to analyze the input text of the specification and claims using LLM. Specifically, it is equipped with modules for morphological analysis, syntactic analysis, and semantic analysis, and performs detailed analysis of the structure and semantic content of the text. In particular, it places emphasis on extracting the components of the invention and their relationships, clarifying the correspondence between the specification and claims. It also checks whether the claim format complies with the Patent Law Enforcement Regulations. (Configuration of evaluation section) The evaluation unit has the function of evaluating the consistency and appropriateness of the specification and claims based on the results of the analysis unit. Specifically, the evaluation is carried out from the following perspectives: -Are the essential elements of the invention described in both the specification and the claims? Whether the elements recited in the claims are explained in the specification to the extent that they can be implemented - Whether the claims are supported by the descriptions in the specification - Whether the claim format complies with the Patent Law Enforcement Regulations These evaluation items are scored using LLM to quantitatively assess the quality of the specification and claims. (Output section configuration) The output section presents the results of the evaluation section to the user and provides an interface to assist in revising the specification and claims. Specifically, it visualizes areas with low evaluation scores and displays comments explaining the reasons. It also uses LLM to automatically generate revision suggestions for areas that require revision and presents them to the user. Users can refer to this information to improve the content of the specification and claims.

[0258] As described above, the system in this embodiment uses LLM to automatically check the contents of patent specifications and claims. The input section receives the text of the specification and claims, and the analysis section analyzes them in detail. The evaluation section then evaluates the consistency and appropriateness of the specification and claims, and the output section presents the results to the user. By linking these functional blocks, a system is realized that efficiently improves the quality of specifications and claims before patent applications are filed.

[0259] By using this system, it is possible to significantly reduce the human burden during the preparation stage of a patent application and support the creation of higher quality patent application documents. In addition, by ensuring the consistency and appropriateness of the specification and claims before filing, it is expected that reasons for rejection will be prevented during patent examination and the likelihood of patent rights being established will be increased.

[0260] Below, we explain the configuration of a computer that uses LLM to search past documents for the purpose of examining novelty and inventive step, based on input claims.

[0261] In this embodiment, a system is provided that uses LLM to efficiently search related past literature to evaluate the novelty and inventive step of a patent application. This system analyzes the content of the input patent claims and extracts keywords that describe the characteristics of the invention. It then uses these keywords to search past patent documents and other technical literature to identify literature related to the novelty and inventive step of the invention. This aims to reduce the burden of prior art searches on examiners and improve the quality of examinations.

[0262] This system is broadly composed of four functional blocks. The first is an input unit that receives the text of the patent claims as input. The second is an analysis unit that analyzes the input text and extracts keywords that represent the characteristics of the invention. The third is a search unit that uses the extracted keywords to search past literature. The fourth is an output unit that presents search results to the user and provides information for evaluating the novelty and inventive step of the invention. Each functional block is explained in detail below. (Input section configuration) The input unit provides an interface for receiving the claim text of the patent application being examined as text data. Specifically, by entering a patent application number, the corresponding claim text is automatically retrieved from the patent publication database. It is also possible to input the claim text directly. (Configuration of the analysis section) The analysis unit has the function of analyzing the input claim text using LLM. Specifically, it is equipped with morphological analysis, syntactic analysis, and semantic analysis modules, and performs detailed analysis of the structure and semantic content of the text. In particular, it focuses on the components of the invention and their relationships, and extracts keywords that represent the characteristics of the invention. When extracting keywords, it takes into account not only the text of the claims but also the content of the specification. In addition, it uses LLM to generate synonyms and related words for the extracted keywords, improving the comprehensiveness of the search. (Search section configuration) The search unit has the function of searching past patent documents and other technical literature using the keywords extracted by the analysis unit. Specifically, it searches patent publication databases and academic literature databases for documents containing the keywords. In this process, LLM is used to collect a wide range of related literature by taking into account the variety of keyword combinations and expressions. LLM is also used to generate summaries of the contents of the documents found in the search results and evaluate their relevance to the invention. This allows examiners to efficiently understand prior art. (Output section configuration) The output section presents the user with information about the documents retrieved by the search section, providing an interface to assist in the evaluation of the novelty and inventive step of the invention. Specifically, it displays a list of documents from the search results, and shows each document's summary and a score for its relevance to the invention. It also highlights parts of the documents that correspond to the elements of the invention, allowing comparison with the claims. Furthermore, it uses LLM to automatically generate findings regarding the novelty and inventive step of the invention, supporting the examiner's judgment. Examiners can use this information to compile the results of their prior art searches.

[0263] As described above, the system in this embodiment uses LLM to analyze the content of patent claims and efficiently search for related past literature. The input unit receives the claim text, and the analysis unit extracts keywords that describe the characteristics of the invention. The search unit then uses the keywords to search for related literature, and the output unit presents the results to the user. By linking these functional blocks, a system is realized that supports prior art searches during patent examination.

[0264] The system will significantly reduce the burden of prior art searches on examiners, improving the efficiency and quality of examinations. It is also expected that by providing objective information on the novelty and inventive step of inventions, it will contribute to ensuring fairness and consistency in examinations.

[0265] Below, we will explain the configuration of a computer using LLM that examines whether the invention described in the claims has novelty and inventive step over prior documents.

[0266] This embodiment provides a system that automatically evaluates the novelty and inventive step of patent application inventions based on comparison with prior art documents. This system uses LLM to analyze the structure of the invention described in the claims and the structure of the invention described in prior art documents, and identifies the differences between them to determine whether the invention is novel or inventive. This aims to support examiners' judgments and improve the efficiency and quality of examinations.

[0267] This system is broadly composed of five functional blocks. The first is an input unit that receives the patent claims and text from prior documents as input. The second is an analysis unit that analyzes the input text and extracts the elements of the invention and their relationships. The third is a comparison unit that compares the patent claims with the structures of the inventions in prior documents and identifies the differences. The fourth is an evaluation unit that evaluates the presence or absence of novelty and inventive step based on the identified differences. The fifth is an output unit that presents the evaluation results to the user and supports the examiner's decision. Each functional block is explained in detail below. (Input section configuration) The input section provides an interface for receiving the text data of the claims of the patent application under examination and the text data of the related prior documents. Specifically, by inputting the patent application number and the document number or URL of the prior document, the respective text data is automatically retrieved from the patent publication database and the literature database. (Configuration of the analysis section) The analysis unit has the functionality to analyze input claims and prior literature sentences using LLM. Specifically, it is equipped with morphological analysis, syntactic analysis, and semantic analysis modules to perform detailed analysis of sentence structure and semantic content. In particular, it focuses on the components of an invention and their relationships, expressing them in a unified format. This allows the structure of inventions in claims and prior literature to be compared using a common standard. (Configuration of comparison section) The comparison unit compares the claims extracted by the analysis unit with the structures of inventions in prior documents to identify differences. Specifically, it uses LLM to determine whether the elements described in the claims are also described in the prior documents. It also compares the relationships between elements and the specific content of the elements to evaluate whether the claimed invention is identical to the invention in the prior document or whether it could be easily arrived at from the invention in the prior document. (Configuration of evaluation section) The Evaluation Department has the function of evaluating the novelty and inventive step of the claimed invention based on the differences identified by the Comparison Department. Specifically, if the claimed invention contains new elements compared to the invention in the prior document, or if it combines known elements in a way that would not be easily conceivable, it is determined to have novelty and inventive step. On the other hand, if the claimed invention is identical to the invention in the prior document, or if it could be easily conceived from the invention in the prior document, it is determined to lack novelty and inventive step. This determination is made using the LLM, taking into account the common general knowledge of a person skilled in the art. (Output section configuration) The output section presents the results of the evaluation section to the user and provides an interface to support the examiner's decision. Specifically, it displays a comparison table between the claimed invention and inventions in prior documents, clearly indicating the differences between their respective components. It also displays the evaluation results for novelty and inventive step, along with information on the prior documents that serve as the basis. Furthermore, it uses LLM to automatically generate explanatory text about the evaluation results, helping the examiner understand them. The examiner can refer to this information when making their final decision.

[0268] As described above, the system in this embodiment uses LLM to analyze the claims and the contents of prior documents, automatically evaluating the novelty and inventive step of an invention. The input section receives the claims and the text of the prior document, and the analysis section extracts the configuration of the invention. The comparison section then identifies differences between the two, and the evaluation section determines whether the invention has novelty or inventive step based on those differences. Finally, the output section presents the evaluation results to the user. By linking these functional blocks, a system is realized that supports the determination of the patentability of inventions during patent examination.

[0269] By using this system, it is possible to significantly reduce the workload of examiners and improve the efficiency and quality of examinations. Furthermore, by clearly indicating the basis for determining the novelty and inventive step of an invention, it is expected to contribute to fulfilling accountability to applicants. However, it is important to note that this system is merely a tool to support examiners' judgments, and the final decision on patentability should be made based on the examiner's specialized knowledge and experience.

[0270] In the above-described embodiment, a computer program and an information processing device are provided that automatically generate patent claims from drawings showing the configuration of an invention using a computer. In a patent application, claims are important disclosures for identifying the technical scope of an invention. However, verbalizing the content of drawings to create appropriate claims requires advanced intellectual work. The computer program and information processing device in this embodiment analyze the information contained in the drawings using natural language processing technology and image recognition technology, and automatically generate patent claims based on the results, thereby significantly streamlining this process.

[0271] The program and information processing device have two main components. The first component is an analysis component that uses a computer vision algorithm to analyze the input drawing and extract the components contained in the drawing and their relationships. The analysis component uses image recognition technologies such as object detection, segmentation, and OCR to identify each element in the drawing and obtain their labels. Furthermore, the analysis component recognizes the lines and arrows that indicate the connections between elements to understand the relationships between elements. This allows the configuration of the invention expressed in the drawing to be extracted in a format that is understandable by a computer.

[0272] The second means is a conversion means that converts the information about the configuration of the invention obtained by the analysis means into text in accordance with the format of the patent claims. The conversion means describes the components extracted from the drawings as the main elements of the claims. It also verbalizes the relationships between the components using expressions such as "comprises..." or "connected to...". In this process, it generates appropriate sentences by taking into account grammatical rules regarding how claims are written and the use of terminology in each technical field. The output of the conversion means is text data that conforms to the format of the patent claims.

[0273] As described above, the computer program and information processing device of this embodiment combine image analysis of drawings with natural language generation technology to automatically create patent claims from drawings. The analysis means uses a computer vision algorithm to accurately recognize the structure of the invention depicted in the drawings and express it in a format that can be processed by a computer. Furthermore, the conversion means implements rules for verbalizing the extracted structure of the invention, allowing for the automatic generation of text that conforms to the format of the claims. This significantly reduces the workload of verbalizing the contents of drawings and contributes to the efficiency of patent application procedures.

[0274] In the above explanation, the types of drawings are not limited to block diagrams and flowcharts, but can also be applied to various other drawings that express the configuration of an invention, such as circuit diagrams and structural diagrams. Furthermore, it is possible to analyze drawings published in patent documents, as well as drawings at the time of patent application, and generate corresponding patent claims. This makes it useful as an auxiliary tool for analyzing the scope of rights of existing patents.

[0275] Next, a multimodal embodiment will be described in which the configurations described in the claims and the corresponding elements of the drawings are displayed in association with each other.

[0276] In this embodiment, a system is provided that visually correlates the structure of an invention described in the claims with the corresponding elements in the drawings to aid in understanding patent specifications. This system uses LLM to analyze the text of the claims and the images in the drawings, identifying the correspondence between them and visualizing the structure of the invention in an easy-to-understand manner. This allows users to intuitively grasp the contents of patent specifications, thereby improving the convenience of examiners and general users.

[0277] This system is broadly composed of four functional blocks. The first is an input unit that receives the text of the claims and images of the drawings as input. The second is an analysis unit that analyzes the input text and drawings and extracts the elements of the invention and their relationships. The third is a matching unit that matches the elements of the claims with the elements of the drawings. The fourth is an output unit that visually presents the results of the matching. Each functional block is explained in detail below. (Input section configuration) The input section extracts the claims and images of drawings from patent specifications and provides an interface for receiving them as text data and image data, respectively. Specifically, by inputting a patent publication document file or PDF file, the claims and drawings are automatically extracted. It is also possible to upload the claims and drawings separately. Images can be either vector data or raster data. (Configuration of the analysis section) The analysis unit has the functionality to analyze the input claim text and drawing images using LLM. Specifically, it applies morphological analysis, syntactic analysis, and semantic analysis modules to the claim text to extract the components of the invention and their relationships. On the other hand, it applies image recognition technologies such as object detection, segmentation, and OCR to the drawing images to identify each element in the drawing and obtain their labels. This allows it to extract information about the structure of the invention from both the claims and drawings. (Configuration of the mapping part) The matching unit has the function of matching the elements of claims extracted by the analysis unit with elements in drawings. Specifically, it uses LLM to evaluate the semantic similarity between words and phrases representing elements of claims and the labels of elements in drawings. It then identifies highly similar combinations as corresponding elements. It also improves the accuracy of the matching by comparing the relationships between elements in claims with the positional relationships of elements in drawings. This automatically identifies the correspondence between the text of claims and drawings. (Output section configuration) The output unit provides an interface for visually presenting the results of the matching unit. Specifically, the output unit displays the claim text and the drawings side by side, and clearly shows the correspondence between corresponding elements and drawing elements by drawing lines. Furthermore, the output unit highlights the elements in the claim text and the corresponding elements in the drawings in the same color, allowing the user to intuitively understand the visual correspondence. Through these displays, the user can easily grasp the relationship between the configuration of the invention described in the claims and the specific embodiments shown in the drawings.

[0278] As described above, the system in this embodiment uses LLM to analyze the claims and drawings and visually present the correspondence between them. The input unit receives the claims and drawings, and the analysis unit extracts their respective components. The matching unit then matches the components of the claims with the elements of the drawings, and the output unit visually presents the results. By linking these functional blocks, a multimodal system is realized that allows users to intuitively understand the contents of patent specifications.

[0279] Using this system will make it easier to understand the contents of patent specifications, which is expected to improve the efficiency of examiners' examinations and promote the use of patent information by general users. Furthermore, by clearly showing the correspondence between claims and drawings, it is believed to contribute to an accurate understanding of the technical scope of inventions. However, it should be noted that the correspondence determined by this system is merely the result of automatic processing, and the final judgment must be made by a human being, taking into account the entire description of the patent specification.

[0280] The following describes a computer configuration that associates each component of the claims with the description of the embodiment of the invention in which it is described.

[0281] In this embodiment, a system is provided that automatically associates and presents the elements of the invention described in the claims with the corresponding description of the embodiments of the invention to aid in understanding patent specifications. This system uses LLM to analyze the text of the claims and the text of the embodiments of the invention, and by identifying the correspondence between the two, it correlates and presents the description of the configuration of the invention. This allows the content of the claims to be understood in conjunction with the specific description of the embodiments of the invention, enabling a deeper understanding of the content of the patent specification.

[0282] This system is broadly composed of four functional blocks. The first is an input unit that receives the claims and the text of the embodiments of the invention as input. The second is an analysis unit that analyzes the input text and extracts the elements of the invention and their relationships. The third is a matching unit that matches the elements of the claims with the description of the embodiments. The fourth is an output unit that presents the results of the matching to the user. Each functional block is explained in detail below. (Input section configuration) The input unit extracts the claims and the description of the preferred embodiment from the patent specification and provides an interface for receiving them as text data. Specifically, by inputting a patent publication document file or PDF file, the claims and the description of the preferred embodiment are automatically extracted. It is also possible to upload the claims and the description of the preferred embodiment separately. (Configuration of the analysis section) The analysis unit has the function of analyzing the input claims and embodiments using LLM. Specifically, it applies morphological analysis, syntactic analysis, and semantic analysis modules to perform a detailed analysis of the structure and meaning of the sentences. In particular, for claims, it focuses on the elements of the invention and their relationships, and expresses them in a unified format. On the other hand, for embodiments, it focuses on how the elements of the invention are specifically realized, and identifies where they are described. This allows it to extract information about the structure of the invention from both the claims and embodiments. (Configuration of the mapping part) The matching unit has the function of matching the claim elements extracted by the analysis unit with the description of the embodiment. Specifically, it uses LLM to evaluate the semantic similarity between the words and phrases representing the claim elements and the description of the embodiment. Then, highly similar combinations are identified as corresponding elements and description locations. In addition, the accuracy of the matching is improved by comparing the relationship between the elements in the claim and the context of the description in the embodiment. This automatically matches each element of the claim with the description of the embodiment where it is specifically explained. (Output section configuration) The output unit provides an interface for presenting the results of the matching unit to the user. Specifically, it displays the claims and the sentences of the embodiments of the invention side by side, and sets links between each element of the claim and the corresponding description of the embodiment. When the user clicks on an element of the claim, they are able to jump to the corresponding description of the embodiment, and vice versa. In addition, the elements of the claim and the corresponding description of the embodiment are highlighted in the same color, allowing the user to intuitively understand the visual correspondence. Through these functions, the user can easily grasp the relationship between the configuration of the invention described in the claims and its specific embodiment.

[0283] As described above, the system in this embodiment uses LLM to analyze the claims and the description of the embodiment of the invention and present the correspondence between them. The input unit receives the claims and the description of the embodiment, and the analysis unit extracts their respective components and description content. The matching unit then matches the components of the claims with the description of the embodiment, and the output unit presents the results to the user. By linking these functional blocks, a system is realized that provides a deeper understanding of the contents of patent specifications.

[0284] By using this system, it is possible to understand the structure of the invention described in the claims in relation to the specific description of the embodiment of the invention, enabling accurate understanding of the contents of the patent specification. This is important for accurately interpreting the technical scope of a patent right, and is expected to contribute to the utilization of patent information and the efficiency of patent examination. However, it should be noted that the correspondence created by this system is merely the result of automatic processing, and the final judgment must be made by a human being, taking into account the entire description of the patent specification.

[0285] Below, we will explain a system for preparing an argument and amendment in response to a notice of rejection issued against a patent application.

[0286] In this embodiment, when a notice of rejection is issued for a patent application, a system is provided that analyzes the reasons for rejection and semi-automatically prepares a written argument and amendment in response. This system uses LLM to understand the content of the notice of rejection, construct a logical argument against it, and generate any necessary amendments. This aims to support the response work of applicants and attorneys and enable efficient and effective acquisition of patent rights.

[0287] This system is broadly composed of five functional blocks. The first is an input unit that receives the contents of the notice of reasons for refusal as input. The second is an analysis unit that analyzes the contents of the input notice of reasons for refusal and identifies the prior art and legal provisions that form the basis for the refusal. The third is an opinion generation unit that constructs the logic of a rebuttal to the reasons for refusal and generates the contents of an opinion. The fourth is an amendment generation unit that generates proposed amendments to resolve the reasons for refusal. The fifth is an output unit that presents the contents of the generated opinion and amendment to the user and makes any necessary corrections. Each functional block is explained in detail below. (Input section configuration) The input section provides an interface for receiving the contents of the Office Action Notice issued by the Patent Office as text data. Specifically, by uploading the Office Action Notice document file or PDF file, the contents are automatically extracted. It is also possible to input information on prior art documents cited in the Office Action Notice. (Configuration of the analysis section) The analysis unit has the functionality to analyze the content of the input rejection notice using LLM. Specifically, it applies morphological analysis, syntactic analysis, and semantic analysis modules to analyze the sentence structure of the rejection notice. It then extracts specific parts of prior art documents and provisions of laws and regulations such as the Patent Act that form the basis of the rejection. It also compares the elements of the invention pointed out in the rejection notice with the elements described in the prior art, clarifying the similarities and differences between them. This allows the content of the rejection notice to be organized in a format that can be processed by a computer. (Configuration of opinion generation unit) The opinion generation unit has the function of constructing a logical argument for the rejection based on the content of the reasons for rejection organized by the analysis unit, and generating the content of the opinion. Specifically, using LLM, it focuses on the differences between the elements of the invention and those of the prior art and constructs a logical argument for asserting the novelty and inventive step of the invention. It also analyzes the interpretation of the legal provisions cited in the notice of rejection and argues why they should not be applied to the present invention. By documenting these arguments in accordance with the format of an opinion, it presents a logical argument for the rejection. (Configuration of the amendment generation unit) The amendment generator has the function of generating proposed amendments to resolve the reasons for refusal based on the content of the reasons for refusal compiled by the analysis unit. Specifically, it uses LLM to identify deficiencies in the elements of the invention pointed out in the notice of reasons for refusal and unclear areas of difference from the prior art, and proposes amendments to clarify these. It also generates amendments that limit the elements of the invention to highlight the differences from the prior art. These amendments are reflected in the claims and description to create amendments to resolve the reasons for refusal. (Output section configuration) The output section provides an interface for presenting the contents of the opinion and amendments generated by the opinion generation section and amendment generation section to the user. Specifically, it displays the text of the generated opinion and amendment on the screen, allowing the user to check their contents. It also provides an editing function so that the user can make corrections as needed. It also has a function for checking whether the contents of the opinion and amendments comply with the requirements of laws and regulations such as the Patent Act, and notifies the user of any deficiencies. Through these functions, the user can carefully review the contents of the opinion and amendments generated by the system and complete the final submission.

[0288] As described above, the system in this embodiment uses the LLM to analyze the contents of an Office Action and semi-automatically prepare an argument and amendment in response to the Office Action. The input unit receives the contents of the Office Action, and the analysis unit organizes the reasons for refusal. The opinion generation unit and amendment generation unit then generate the logic for refuting the reasons for refusal and proposed amendments, and the output unit presents the results to the user. By linking these functional blocks, a system for efficiently dealing with reasons for refusal in patent applications is realized.

[0289] Using this system can significantly reduce the time and effort required to respond to office action notices, making it possible to obtain patent rights more quickly and reliably. Furthermore, by utilizing LLM, the process of responding to office action notices, which previously relied on human expertise, can be automated while maintaining a certain level of quality. However, it should be noted that the generation of opinions and amendments by this system is merely supplementary, and the final decision must still be made by a human being, taking into account the interpretation of laws and regulations such as the Patent Act and the technical significance of the invention. (Other configurations) The patent claim processing system using a large-scale language model (LLM) analyzes the input claim text using an LLM. The LLM uses deep learning models specialized for natural language processing, such as GPT-3, BERT, XLNet, and RoBERTa. The LLM analyzes the claim text structure and identifies subjects (main components), predicates (relationships between elements), objects (subordinate components), etc.

[0290] To generate code for the diagram, the LLM analysis results in program code (e.g., SVG code) that generates vector data for computer graphics from the identified components and relationship information. The generated code includes information such as the type of diagram (block diagram or flowchart), the layout of each component, and their connections.

[0291] Drawings based on code are automatically generated based on generated code (SVG, etc.) to appropriately express the content of the claims. Visualization libraries such as Python's Matplotlib and JavaScript's D3.js are used to draw the drawings. When drawing, arrows are automatically added to the drawing to represent processing flow, time series, and signal flow. Components are drawn as blocks, and nested structures (blocks within blocks) can also be expressed. 1. Patent Claim Analysis Using the LLM - Morphological analysis, syntactic analysis, and semantic analysis are applied sequentially to the input patent claim text. - Morphological analysis divides a sentence into words and identifies the part of speech and inflections of each word. - Syntactic analysis analyzes the dependency relationships between words and generates a syntax tree for the sentence. - Semantic analysis identifies semantic roles such as subject, predicate, and object based on the syntax tree. - LLM uses models pre-trained on a large amount of patent literature, specializing in understanding the meaning of sentences. - The input to LLM is a tokenized version of the patent claim text. - The output of LLM is a vector representation that shows the semantic role of each word and the relationships between words. - From these vector representations, the components of the invention and their relationships are extracted. 2. Code Generation Algorithm for Drawing - Extract the information necessary for drawing from the analysis results of LLM. - Information extracted includes the names of components, their relationships, and importance. - Generate SVG code based on this information. - Code generation is performed by applying extracted information to pre-prepared templates. - Templates are provided for each type of diagram (block diagram, flowchart, etc.). - Convert components into SVG rectangle elements and relationships into line and arrow elements. - Element placement is automatically determined based on importance. - Use different types of arrows (solid lines, dotted lines, etc.) and colors to express relationships. 3. Drawing Algorithm - The generated SVG code is passed to a drawing library to generate the drawing. - Python's Matplotlib and JavaScript's D3.js are used as drawing libraries. - These libraries have the ability to interpret SVG code and generate graphics. - When drawing, they automatically optimize the placement of components and the calculation of line and arrow paths. - Drawing blocks within blocks is achieved using a recursive algorithm. - When dealing with multiple claims, each claim is generated as an independent drawing, and then an overall diagram showing the relationships between them is generated.

[0292] By combining natural language processing using LLM and vector graphics generation using SVG, we have achieved a process for automatically generating drawings from patent claims.

[0293] As described above, drawings may be created by processing the claims, or drawings may be created by processing the embodiments and examples of the invention. Claims, embodiments and examples of the invention may also be created from the drawings.

[0294] The processes and flows in this embodiment may be executed by a plurality of software programs.

[0295] The above-described embodiments and the elements contained therein (part of the configurations, part of the processes) can be combined or replaced to create new and different embodiments. [Example of invention configuration] The following configurations of the invention are also possible.

[0296] An information processing device that accepts input from an invention notification form or an invention input form and creates documents for a patent application, a first generating means for extracting prior art from the invention notification form or the input content using a large-scale language model, and generating prior art from training data, a database, or the Internet if prior art cannot be extracted; a second generation means for extracting a problem from the invention notification form or the input content using a large-scale language model, and generating the problem from training data, a database, or the Internet if the problem cannot be extracted; a third generation means for extracting a means for solving the problem from the invention notification or the content of the input using a large-scale language model, and if the means cannot be extracted, generating the means from training data, a database, or the Internet; a fourth generation means for extracting the effect of the invention from the invention notification form or the content of the input using a large-scale language model, and generating the effect from training data, a database, or the Internet if the effect cannot be extracted; a fifth generating means for extracting a title of the invention from the invention notification form or the content of the input using a large-scale language model, and generating the title from training data, a database, or the Internet if the title cannot be extracted; a sixth generating means for extracting drawings from the invention notification form or the input contents, and if the drawings cannot be extracted, generating a drawing or a code for generating a drawing from learning data, a database, or the Internet; a seventh generating means for extracting a description of drawings from the invention notification form or the input content using a large-scale language model, and generating the same from training data, a database, or the Internet if the description cannot be extracted; an eighth generation means for extracting an abstract from the invention notification form or the input content using a large-scale language model, and if the abstract cannot be extracted, for generating the abstract from training data, a database, or the Internet; and output means for outputting documents for a patent application by combining the generated outputs of said first to eighth generation means.

[0297] The following configurations of the invention are also possible.

[0298] An information processing device that uses a computer to input an invention notification form and prepare patent application documents, input means for receiving an invention notification form as an input; an invention notification analysis means for decomposing the invention notification into constituent elements; a patent specification generation means for generating each section of a patent specification using a language model based on the data consisting of the components; a consistency assurance means for checking consistency between sections of the generated patent specification and correcting it as necessary; an output means for outputting the patent specification with the consistency ensured; An information processing device comprising:

[0299] In the above device, The invention notification document analysis means extracts the name of the invention, inventor information, an outline of the invention, a detailed description of the invention, drawings, and related prior art documents from the invention notification document, and is a device for creating patent application documents using an invention notification document as input.

[0300] In the above device, An apparatus for creating patent application documents using an invention notification form as input, characterized in that the patent specification generation means generates sections including the title of the invention, an outline of the invention, a detailed description of the invention, claims, and descriptions of drawings.

[0301] In the above device, The consistency assurance means is a device for creating patent application documents using an invention notification form as input, characterized in that it checks the consistency between the outline of the invention and the detailed description, the consistency between the claims and the detailed description of the invention, the consistency between the description of the drawings and the detailed description of the invention, and the consistency of the entire patent specification, and makes corrections as necessary.

[0302] In the above device, 1. An apparatus for creating patent application documents using an invention notification document as input, wherein the language model is pre-trained to match the style and structure of a patent specification.

[0303] In the above device, 1. An apparatus for preparing patent application documents using an invention notification document as an input, further comprising: a feedback learning means for receiving feedback from a user and continuously improving the language model.

[0304] The above-described configuration has the effect of improving efficiency by automatically generating patent specifications from invention notifications, and improving the quality of patent specifications by using language models.

[0305] [Second embodiment] Next, differences between the information processing device of the second embodiment and the above-mentioned embodiments will be described. The information processing device of the second embodiment includes an acquisition means for processing a first character string using a large-scale language model (LLM) and acquiring knowledge data thereof, an update means for processing a second character string using the large-scale language model based on the acquired knowledge data and updating the knowledge data, and a processing means for processing a third character string using the large-scale language model based on the updated knowledge data. The information processing device inputs character strings (character sets) from a first group to an n-th group (n is a natural number equal to or greater than 2) into the LLM in order and processes each of them. Each group of character strings may be, for example, character strings constituting a patent specification, character strings constituting patent claims, character strings constituting an abstract, or the like, or may be a sentence and its translation, or may be any other document.

[0306] When a first string is processed using a large-scale language model (LLM), a prompt to the LLM includes instructions to output the processing results and acquire knowledge data, and based on this, the information processing device acquires the processing results and knowledge data.

[0307] When a second string is processed using a large-scale language model based on the acquired knowledge data, the instruction (prompt) to the LLM includes an instruction to output the processing result based on the knowledge data and to acquire and update new knowledge data, and based on this, the information processing device acquires the processing result and updated knowledge data.

[0308] Similarly, when a third string is processed using a large-scale language model based on the acquired knowledge data, the instruction (prompt) to the LLM includes an instruction to output the processing result based on the knowledge data and to acquire and update new knowledge data, and based on this, the information processing device acquires the processing result and updated knowledge data.

[0309] When the nth character string is processed using a large-scale language model based on updated knowledge data, the instruction (prompt) to the LLM includes an instruction to output the processing result based on the (updated) knowledge data, and the information processing device acquires the processing result based on this. When the nth character string is processed using a large-scale language model, the instruction (prompt) to the LLM may also include an instruction to acquire and update new knowledge data. Finally, the processing results (output) of the character strings (sets of characters) from the first group to the nth group (n is a natural number greater than or equal to 2) may be compiled into a single file.

[0310] Inputting a large amount of data into an LLM at once presents a problem of reduced accuracy in analysis, processing, and generated output (such as character strings). Meanwhile, inputting data divided into multiple parts presents a problem of difficulty in ensuring consistency among the multiple divided parts. The information processing device of this embodiment solves this problem by dividing (e.g., dividing) a character string (or a sentence, paragraph, document, etc.) to be processed into multiple parts, while still making it easier to ensure consistency among the processing results among the multiple parts.

[0311] The information processing device in this embodiment performs the following processes.

[0312] (1) Divide the character string (document) to be processed into multiple processing units. The divisions are made at the end of a sentence, a line break symbol, a paragraph, a page, a chapter, a heading, a division, etc. This division can be done automatically or manually. When processing multiple documents, each document can be used as a processing unit. It is also possible to process multiple documents in order without dividing them. In this case, the present invention can also be implemented.

[0313] (2) The first part of the divided string is processed by the LLM (or an external LLM). The LLM receives the first part and a prompt. The prompt includes at least (A) the processing content to be performed on the divided part (processed data to be output), and (B) instructions to acquire, store, or output the knowledge data obtained by the processing.

[0314] For example, the processing content (processing data to be output) to be performed on the divided part of (A) includes document checking, translation, rewriting, proofreading, etc. The knowledge data obtained by the processing of (B) includes information such as a vocabulary list, bilingual word translations (bilingual word tables), bilingual sentence translations, words and their reference symbols, which spelling variations of words to standardize, whether the desumasu style is used, etc.

[0315] (3) The next (second) part of the divided string is processed by the LLM (this may be processed by an external LLM). The LLM receives the second part and a prompt. The prompt includes (A) the content of the processing to be performed on that divided part (the processed data to be output) (this may be the same as the content of the previous processing), (B) the knowledge data obtained in the previous processing (for example, the processing immediately before), (C) that the processing shall be in accordance with the knowledge data, and (D) instructions to obtain, store, or output the knowledge data obtained by the current processing, and to update the knowledge data obtained so far (append or update data).

[0316] By this process, the process (A) is carried out based on the knowledge data obtained by the previous process (information such as the vocabulary list, bilingual word translations, bilingual sentence translations, words and their reference symbols, which spelling variations are used, whether the desumasu style is used, etc.), and the consistency of the process can be maintained. Also, by processing the second part, the newly obtained knowledge data updates the knowledge data obtained up to that point, and this is used in processing the next divided part.

[0317] (4) The next and subsequent (third and subsequent) parts of the divided document are processed by the LLM. In this case, input to the LLM is performed in the same manner as in (3) above. That is, the prompt includes at least (A) the content of the processing to be performed on the divided part (processing data to be output) (this may be the same as the content of the previous processing), (B) the knowledge data obtained in the previous processing (updated knowledge data), (C) that the processing shall be in accordance with that knowledge data, and (D) instructions to obtain, store, or output the knowledge data obtained by the current processing, and to update (append data) the knowledge data obtained so far.

[0318] By repeating this process (4) for multiple subsequent divisions, processing is performed according to the knowledge data, and the knowledge data obtained up to that point is updated and used to process the next division. This allows each division to be processed according to the same standards, even when the document is divided and processed by LLM, and ensures the same output quality.

[0319] For example, when processing multiple documents, by treating each document as a single division unit, the multiple documents can be processed using the same criteria. Also, by dividing a single document into multiple parts and processing them in order, all parts of the document can be processed using the same criteria.

[0320] The finally obtained knowledge data can be used in other ways. For example, knowledge data can include information such as a vocabulary book, bilingual word translations (bilingual word table), bilingual sentence translations, words and their reference symbols, which spelling variations of words to standardize, and whether the desumasu style is used. This information can be used as a vocabulary book (a list of words used), a bilingual table (a table that associates written parts in a first language with written parts in a second language), a code table (a table that associates nouns with reference symbols), etc. Such knowledge data can be stored and used in processing the next document. The present invention can also be applied to a process of generating a document based on knowledge data. When new knowledge data is obtained for each generation, the knowledge data can be updated.

[0321] According to this embodiment, it is possible to automatically create knowledge data such as a code table (a table that associates nouns with reference symbols), a vocabulary book (a list of words used), and a bilingual table (a table that associates written parts in a first language with written parts in a second language), and to process documents accordingly. Furthermore, it is possible to update the knowledge data by processing new documents and sentences. The technology in this embodiment is particularly useful for document processing, automatic document checking, automatic revision, specification writing, sentence generation, checking of patent specifications, etc., translation, and translation checking.

[0322] Methods for inputting knowledge data into the LLM include including the knowledge data in the prompt, including the knowledge data in the history data (memory data) of past conversations, and including the knowledge data in the data source of search expansion and generation (RAG).

[0323] As described above, when checking, revising, and proofreading a patent specification using an LLM, for example, the first string of characters that make up the patent specification is entered into the LLM along with a prompt. Knowledge data is obtained along with the processing results, and this knowledge data is used to process the next string. The knowledge data is updated each time a string is processed. The resulting knowledge data can be used in the future as a code table (a table that matches nouns with reference symbols) or a vocabulary book (a list of commonly used words). For example, the LLM can be used to check, revise, and proofread the next patent specification based on this code table (a table that matches nouns with reference symbols) or vocabulary book (a list of commonly used words). Alternatively, the code table (a table that matches nouns with reference symbols) and vocabulary book (a list of commonly used words) can be used to draft and generate opinions, amendments, notices of reasons for refusal, requests for appeals, and appeal decisions using the LLM, or even to create new specifications.

[0324] For example, when using an LLM to check, revise, and proofread a translation, the first string (part of the original text and its translation) of a pair of texts (the original text and its translation) is entered into the LLM along with a prompt. Knowledge data is obtained along with the processing results, and this knowledge data is used to process the next string. The knowledge data is updated each time a string is processed. The final knowledge data can be used in the future as a code table (a table that matches nouns with reference symbols) or a vocabulary book (a list of commonly used words). For example, the code table (a table that matches nouns with reference symbols) and vocabulary book (a list of commonly used words) can be used to check, revise, and proofread the next translation, or the code table (a table that matches nouns with reference symbols) and vocabulary book (a list of commonly used words) can be used to generate a new translation from a new source text using the LLM.

[0325] In the second embodiment, the knowledge data is updated after it is created, but it is also possible to create the knowledge data by processing a character string once, and process subsequent character strings based on that knowledge data (without updating). In other words, the knowledge data is (automatically) generated, and subsequent character strings are processed based on (according to) that knowledge data.

[0326] The knowledge data is preferably in the form of a table that associates words and reference codes (e.g., words and their translations, sentences and their translations, sentences and sentences, words and their explanations) one-to-one, one-to-many, or many-to-many, text in CSV format, general text, natural language, etc. Also, as long as the data format can be processed by LLM, it does not necessarily have to be understandable by humans; compressed data, encryption, etc. may also be used.

[0327] In this way, processing using a large-scale language model can be stabilized.

[0328] In the above embodiment, processing of patent specifications and translated documents has been taken as an example, but the information processing device can also be used to process other documents.

[0329] [Third embodiment] Next, the differences between the information processing device of the third embodiment and the above-described embodiments will be described. The device and process of this embodiment may be used alone or in combination with the device and process described above.

[0330] FIG. 12 is a block diagram showing a system configuration according to the third embodiment of the present invention.

[0331] This information processing device comprises an input means, a translation means, and a proofreading means. The input means inputs text written in a first language. The translation means translates the text written in the first language using a large-scale language model. The large-scale language model may be internal or external to the device. If it is external, it communicates with the outside via an API or the like to execute the translation process. The method of the first or second embodiment may be adopted for the translation. The document to be translated may be divided into multiple parts, and translation may be executed for each part. In other words, the translation process is repeated while changing the translation position.

[0332] The proofreading means extracts a portion of the translated text, sends it together with a corresponding portion of the text written in the first language in a prompt, and proofreads the portion through a large-scale language model (which may be the same as or different from the large-scale language model used by the translation means). The proofreading means proofreads the entire text written in the first language by repeating the process of the proofreading means multiple times, changing the portion extracted from the translated text.

[0333] More specifically, information is input from an input means and passed to a translation means. The translation means creates a translation request prompt (including a specification of the language to be translated and output) that includes the input information (text written in a first language), and sends it to a large-scale language model for translation processing. As described above, this may be done by dividing the text into multiple parts, translating them in order, and finally combining them to obtain the entire translation. In other words, the entire translation may be achieved by repeatedly translating parts of the text written in the first language (in this case, the text written in the first language is divided, prompts containing the original text are input to the large-scale language model multiple times, and the translation results are finally combined). Because processing accuracy decreases when the prompt is long, it is desirable to process the text written in the first language by dividing it in this way. The translation result is passed to a proofreading means.

[0334] To proofread and correct the translation result, the proofreading means sends a portion of the translation sentence and a corresponding portion of the text written in the first language, together with construction instructions, to the large-scale language model in a prompt. The large-scale language model proofreads using the translation sentence and the original sentence, generates a final proofread translation, and passes it to the output means. The output means outputs the proofread document received from the large-scale language model. The proofreading means repeats the process by shifting the portion to be proofread from the beginning to the end of the translation sentence, and finally proofreads the entire translation sentence.

[0335] This configuration allows input information to be translated and processed efficiently with high quality. Note that while proofreading is performed once for the entire text here, proofreading may be performed more than once. Also, proofreading may be performed multiple times using multiple types of LLMs. The LLM that performs translation and the LLM that performs proofreading may be the same or different.

[0336] FIG. 13 is a diagram showing the operation concept of the system according to the third embodiment of the present invention.

[0337] The input means receives information, inputting text written in a first language, and the translation means translates the text written in the first language and generates text in a second language using a large-scale language model.

[0338] The proofreading means extracts portions of the translated text and compares them with the corresponding portions of the text written in the first language to evaluate and proofread the translation. If necessary, the translation result is corrected (proofread) using a large-scale language model. The entire translation is proofread by changing the extracted portions. Both the final translation result and the proofread result are passed to the output means and presented to the user.

[0339] The input text in the first language, the generated text in the second language, and the proofreading results may be stored for future translation quality improvements or used to train a large-scale language model.

[0340] This type of processing allows input text to be translated into another language efficiently and with high accuracy. Utilizing large-scale language models in the translation and proofreading processes enables natural translation that takes context into account. Furthermore, accumulating and utilizing the processing results can be expected to continuously improve the language model.

[0341] FIG. 14 is a flowchart showing the processing in the third embodiment of the present invention.

[0342] This flowchart shows the process of taking text written in a first language and translating it into a second language using a large-scale language model. At each step:

[0343] Input text written in a first language. The input text is translated using a large-scale language model. The language model takes context into account to generate natural and appropriate translation results. Translation using LLM can also be done by dividing the source text into multiple parts, translating each part separately, and finally combining the parts into a single translation.

[0344] Next, the quality of the translation is assessed and the translation is proofread by dividing the translated text into parts, extracting each part, and comparing them with the corresponding part of the source text. This is done by sending a prompt to the LLM containing a portion of the translation and its corresponding part in the source text, along with instructions to proofread the translation. If the translation quality does not meet standards, the large-scale language model may be re-used to correct the translation. This step may involve analyzing the differences between the source text and the current translation and replacing them with more appropriate expressions.

[0345] The extracted part is changed and the above process is performed on the entire translation. The process is performed on all parts of the translation, that is, from the beginning to the end of the document, for example. In this way, the entire translation is proofread.

[0346] The proofreading process may be repeated until a translation result that meets quality standards is obtained. This results in the final translation result being output. Proofreading can be performed by correcting the translated text, or the proofread text can be output as new text.

[0347] The original text and its final translation are saved, and may be used as additional training data for large-scale language models to improve translation quality in the future.

[0348] The above process enables high-quality translation of input text in a first language into a second language. By effectively using large-scale language models for translation and proofreading, it is possible to achieve natural-sounding translation that takes context into account, while simultaneously achieving quality assurance and continuous quality improvement. Furthermore, proofreading multiple times can produce more refined translation results. Furthermore, proofreading can be performed by sending only the translation to the LLM without sending the original text to the LLM. This proofreading is performed to check for typos and ensure a natural translation. Furthermore, as in the second embodiment, knowledge data such as bilingual tables and code tables can be used for translation and proofreading.

[0349] The translation into the second language as described above may be reverse-translated into text in the first language using the LLM used for translation, the LLM used for proofreading, or a different LLM, and output. This allows a human checker to check the content of the translation into the second language in the first language. This reverse translation may be performed in bulk using an LLM (all text to be translated may be sent and translated in a single prompt), or it may be performed using machine translation such as a neural network. Furthermore, as in the translation and proofreading described above, a process of translating a portion (a small amount) at a time may be repeatedly executed in a loop, and the output from each loop may be combined into a single translation. The translation reverse-translated into text in the first language may be proofread using the technology described in the above-described embodiment. The proofreading may be performed two or more times.

[0350] Such backward-translated text may be output by comparing the two, such as by paragraph, page, or sentence, so that the correspondence between the original text in the first language and the backward-translated text is easy to understand. For example, the original text and the backward-translated text may be output in a table format, arranged by unit, or the original text and the corresponding backward-translated text may be output at each line break. Two or three of the backward-translated text, the original text in the first language, and the translated text in the second language may also be output by comparing the two or three, such as by paragraph, page, or sentence, so that the correspondence between them is easy to understand. Furthermore, a vocabulary book or code table may be generated using the original text, the translated text, and the backward-translated text, and the correspondence between words and codes may be displayed in an easy-to-understand manner. The device may be configured to check two or three of the backward-translated text, the original text in the first language, and the translated text in the second language using LLM to check for mistranslations or omissions, and output a warning if any are found.

[0351] The processes described in the above embodiments may be executed in appropriate combinations.

[0352] The processes in the above-described embodiments may be performed by software or by using hardware circuits. A program for executing the processes in the above-described embodiments may be provided, or the program may be recorded on a recording medium such as a CD-ROM, a flexible disk, a hard disk, a ROM, RAM, or a memory card and provided to a user. The program is executed by a computer such as a CPU. The program may also be downloaded to a device via a communication line such as the Internet.

[0353] The above-described embodiments should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims, not by the above description, and is intended to include all modifications within the meaning and scope of the claims.

Claims

[Claim 1] input means for inputting text written in a first language; a translation means for translating text written in the first language using a large-scale language model; proofreading means for extracting a portion of the translated text, including it in a prompt together with a corresponding portion of text written in the first language, and proofreading the portion using a large-scale language model; The part extracted by the proofreading means is changed and the process by the proofreading means is repeated a plurality of times to translate and proofread the text written in the first language; an information processing device comprising output means for converting the proofread translated text back into text in the first language and outputting the text;

Images