Article generation support device

Abstract

A text generation support device 1 that supports the generation of text that explains a group of patent documents comprises: a statistical data acquisition module 11 that acquires statistical data indicating the results of statistical processing performed on the group of patent documents; a notable portion acquisition module 12 that acquires, based on the statistical data, information that identifies notable portions in the statistical data and reasons for noticing the notable portions; a text acquisition module 13 that acquires, based on the information that identifies the notable portions, text data for each of a plurality of patent documents included in the group of patent documents and that correspond to the notable portions; and a first prompt generation module 14 that generates a first prompt that causes a first generation AI to output an explanatory text based on the notable portions, the reasons for noticing, and the text data.

Description

[Technical Field]

[0001] The present invention relates to a text generation support device that uses generation AI to support the generation of text that explains a group of patent documents that includes multiple patent documents. [Background technology]

[0002] Various statistical processing methods have been proposed to analyze trends in patent applications by specific companies or in specific technical fields. Meanwhile, in recent years, methods utilizing generative AI have also been explored. For example, generative AI could be given the results of statistical processing on patent applications to write explanatory text, or it could be given patent documents such as patent application documents and patent gazettes to write explanatory text. The text generated by generative AI has become so accurate that it is indistinguishable from text written by humans at first glance.

[0003] However, even if the results of statistical processing are fed into a generative AI, the AI may only be able to output superficial information that can be understood by looking at graphs and numbers. Furthermore, when a generative AI is fed a patent document, it may be able to extract and summarize information in a balanced manner from the document, but it may not be able to demonstrate the deep insight required to analyze trends in patent applications. For this reason, it is not easy to get a generative AI to generate explanatory text that resembles the text written by an expert. Summary of the Invention [Problem to be solved by the invention]

[0004] One aspect of the present invention relates to the generation of specific and persuasive explanatory text by using the results of statistical processing of patent applications to incorporate statistical characteristics that are difficult to know by simply reading the patent documents, and also by delving into the specific content of the patent documents. [Means for solving the problem]

[0005] A text generation assistance device according to one aspect of the present invention comprises: A text generation support device that supports the generation of text that explains a group of patent documents, a statistical data acquisition module for acquiring statistical data indicating the results of statistical processing of a group of patent documents; an attention point acquisition module that acquires information specifying an attention point in the statistical data and an attention reason for focusing on the attention point based on the statistical data; a text acquisition module that acquires text data for each of a plurality of patent documents included in the patent document group and corresponding to the portion of interest based on information identifying the portion of interest; a first prompt generation module that generates a first prompt that causes the first generation AI to output an explanatory sentence based on the attention part, the attention reason, and the text data; Equipped with. [Brief explanation of the drawings]

[0006] [Figure 1] FIG. 1 shows a writing generation assistance device 1 according to an embodiment and an external device connected to the writing generation assistance device 1. [Figure 2] FIG. 2 shows a first example of statistical data acquired by the statistical data acquisition module 11 in the embodiment. [Figure 3] FIG. 3 shows a textual representation of the statistical data of FIG. [Figure 4] FIG. 4 shows a second example of statistical data acquired by the statistical data acquisition module 11 in the embodiment. [Figure 5] FIG. 5 shows a list of higher and lower classifications of the FI classification symbol "C03C 3 / 076" shown on the vertical axis of FIG. [Figure 6] Figure 6 shows statistical data in which the vertical axis of Figure 4 is replaced with specific classification names. [Figure 7] FIG. 7 shows a textual representation of the statistical data of FIG. [Figure 8]FIG. 8 shows an example of a prompt that causes the generation AI3 to perform a general analysis of statistical data in an embodiment. [Figure 9] FIG. 9 shows an example of a response from Production AI3 to the prompt shown in FIG. [Figure 10] FIG. 10 shows an example of a prompt generated by the interest acquisition module 12 in the embodiment. [Figure 11] FIG. 11 shows an example of a response from Production AI3 to the prompt shown in FIG. [Figure 12] FIG. 12 shows an example of text data acquired by the text acquisition module 13 in the embodiment. [Figure 13] FIG. 13 shows an example of a first prompt generated by the first prompt generation module 14 in the embodiment. [Figure 14] FIG. 14 shows an example of a response from Production AI3 to the first prompt shown in FIG. [Figure 15] FIG. 15 shows an example of a second prompt generated by the second prompt generation module 15 in the embodiment. [Figure 16] FIG. 16 shows an example of some of the responses from Production AI3 to the second prompt shown in FIG. DETAILED DESCRIPTION OF THE INVENTION

[0007] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The embodiments described below are merely examples of the present invention and are not intended to limit the scope of the present invention. Furthermore, not all of the configurations and operations described in the embodiments are necessarily essential as the configurations and operations of the present invention. Note that the same components are designated by the same reference numerals, and redundant explanations will be omitted.

[0008] <1. Configuration and Function> 1 shows a writing generation support device 1 according to an embodiment and an external device connected to the writing generation support device 1. The writing generation support device 1 is a computer system including a CPU, a memory, etc. (not shown). The writing generation support device 1 may be configured with a single computer or may be configured with multiple computers connected via a network. The writing generation support device 1 is connected to external devices such as a patent database 2 and a generation AI 3.

[0009] The patent database 2 stores patent documents. Patent documents include patent application documents or patent gazettes (unexamined patent gazettes, patent publication gazettes, etc.). Patent documents are not limited to patents defined by the Patent Act and may include utility model documents, and may include foreign patent documents as well as Japanese patent documents. The patent database 2 may further store bibliographic information such as examination progress information that is not published in the gazettes. The patent database 2 is not limited to data stored in a single storage device, but may also be data stored distributed across multiple storage devices. The writing generation support device 1 acquires data from the patent database 2 and performs various processes.

[0010] The generation AI 3 includes a large-scale language model (LLM). The large-scale language model is a language model constructed using large amounts of text data and deep learning technology, and processes tasks such as sentence generation, sentence summarization, and question answering in response to prompts sent from the writing generation assistance device 1. The generation AI 3 may include multiple large-scale language models. The multiple large-scale language models may include both models provided by external providers and used via a network, and models implemented in a local environment. The multiple large-scale language models may include both high-performance models and models that are lower in performance but lightweight and inexpensive. The writing generation assistance device 1 sends prompts to the generation AI 3 and obtains output from the generation AI 3.

[0011] The writing production support device 1 includes a statistical data acquisition module 11, a noteworthy part acquisition module 12, a text acquisition module 13, a first prompt generation module 14, and a second prompt generation module 15. These modules are implemented by loading programs into a memory included in the writing production support device 1 and executing them by a CPU.

[0012] The statistical data acquisition module 11 acquires a patent document group including a plurality of patent documents from the patent database 2, and acquires statistical data by performing statistical processing on this patent document group. The patent document group may be, for example, a collection of documents of patent applications filed by a specific company (e.g., Company A), a collection of documents of patent applications filed by multiple companies belonging to a specific industry, or a collection of documents of patent applications in a specific technical field. Examples of statistical data will be described later with reference to FIGS. 2 to 7. Alternatively, the statistical data acquisition module 11 is not limited to performing statistical processing, and may acquire the statistical data by receiving the results of statistical processing performed by another device.

[0013] The interest point acquisition module 12 acquires information identifying an interest point in the statistical data and an interest reason for focusing on the interest point based on the statistical data acquired by the statistical data acquisition module 11. Specifically, the interest point acquisition module 12 generates a prompt that causes the generation AI 3 to respond with the interest point and the interest reason. The interest point acquisition module 12 acquires the interest point and the interest reason by sending a prompt to the generation AI 3 and receiving a response from the generation AI 3. An example of the prompt will be described later with reference to FIG. 10, and examples of the interest point and the interest reason will be described later with reference to FIG. 11.

[0014] The method of acquiring the points of interest and the reasons for attention is not limited to having the generation AI 3 respond. The points of interest acquisition module 12 may acquire the points of interest and the reasons for attention for each reason for attention for which a point of interest is found by determining whether or not there is a corresponding point of interest for each candidate reason for attention (such as the year when the number of cases peaked or the year when the growth rate peaked), respectively. The points of interest are not limited to being acquired on an annual basis, but may be acquired on an arbitrary basis such as on a quarterly basis. Furthermore, if statistical data is aggregated not on an annual basis but on a technical classification, status, or other basis, the points of interest may be acquired on such a basis.

[0015] The text acquisition module 13 acquires text data from the patent database 2 for each of the multiple patent documents corresponding to the noteworthy portion of the patent document group, based on the information identifying the noteworthy portion acquired by the noteworthy portion acquisition module 12. However, it is not necessary to acquire text data for all patent documents corresponding to the noteworthy portion; it is also possible to identify a predetermined number of patent documents in descending order of the importance score assigned to the patent document or in descending order of the filing date, and acquire text data for each of the identified patent documents. If there are multiple combinations of noteworthy portion and reason for attention, multiple patent documents are identified for each reason for attention.

[0016] The first prompt generation module 14 generates a first prompt that causes the generation AI 3 to output an explanatory sentence based on the attention points and attention reasons acquired by the attention point acquisition module 12 and the text data acquired by the text acquisition module 13. The first prompt generation module 14 acquires the explanatory sentence by sending the first prompt to the generation AI 3 and receiving a response from the generation AI 3. An example of the first prompt will be described later with reference to FIG. 13, and an example of the explanatory sentence will be described later with reference to FIG. 14. When there are multiple combinations of attention points and attention reasons, an explanatory sentence is acquired for each attention reason.

[0017] The second prompt generation module 15 generates a second prompt that causes the generation AI 3 to generate a report that integrates the explanatory text obtained for each reason for attention by the first prompt generation module 14. The second prompt generation module 15 obtains a report of the group of patent documents by sending a second prompt to the generation AI 3 and receiving a response from the generation AI 3.

[0018] In the present application, the generation AI3 that generates an explanatory text in response to a first prompt may be referred to as the first generation AI, the generation AI3 that generates a report in response to a second prompt may be referred to as the second generation AI, and the generation AI3 that answers the attention point and the reason for attention in response to the prompt generated by the attention point acquisition module 12 may be referred to as the third generation AI. The first to third generation AIs may be the same generation AI or may be different generation AIs.

[0019] <2. Specific Examples> <2-1.Statistical Data> 2 shows a first example of statistical data acquired by the statistical data acquisition module 11 in the embodiment. The first example shows the number of patent applications filed by Company A by each filing year. The statistical data shown in FIG. 2 is used to analyze the trend in the number of applications filed by Company A.

[0020] Figure 3 shows a text representation of the statistical data in Figure 2. Figure 3 includes a line showing the label (application year), a line showing the data (number of cases), and a line showing the boundary between the label and the data, with the vertical bar "|" used as a delimiter in each line to represent the list. If the generation AI 3 is good at processing text data, it is desirable to use the format shown in Figure 3.

[0021] FIG. 4 shows a second example of statistical data acquired by the statistical data acquisition module 11 in the embodiment. The second example shows the number of patent applications (excluding expired applications) filed by Company A by technology classification (FI) and the ratio of the number of applications by status (filed only, under examination, and continuing patent) of patent applications belonging to each technology classification. Note that "filed only" refers to applications for which no request for examination has been filed. The statistical data shown in FIG. 4 is used to analyze which technical fields Company A is focusing on in filing patent applications and which technical fields it is focusing on in terms of examination and granting patents.

[0022] FIG. 5 shows a list of higher and lower classes of the FI classification symbol "C03C 3 / 076" shown on the vertical axis of FIG. 4. For a person viewing the analysis results of FIG. 4, or for the generator AI3, it may be difficult to understand the specific technical field from the FI classification symbol alone. While replacing the FI classification symbol with a specific classification name may facilitate understanding, the definition of "C03C 3 / 076"—"having 40% to 90% silica by weight"—may still be difficult to understand. In an embodiment, the list of higher and lower classes of "C03C 3 / 076" shown in FIG. 5 may be provided to the generator AI3, which may then generate a concise and easy-to-understand classification name.

[0023] Figure 6 shows statistical data in which the vertical axis of Figure 4 has been replaced with specific classification names. By providing a list of higher and lower classifications for each FI classification symbol shown in Figure 4 to generation AI3, the classification names shown in Figure 6 can be obtained. "C03C 3 / 076" has been replaced with "glass containing 40% to 90% silica."

[0024] Figure 7 shows a text representation of the statistical data in Figure 6. Like Figure 3, Figure 7 also contains a row showing labels, a row showing data, and a row showing the boundary between labels and data, with each row separated by a vertical bar "|" to represent the table. Figure 7 also shows the number of applications by status.

[0025] <2-2. General analysis of statistical data> Fig. 8 shows an example of a prompt that causes the generation AI 3 to perform a general analysis of statistical data in an embodiment. In Fig. 8 and subsequent figures, a case is shown in which the first example shown in Fig. 2 or 3 is used as the statistical data. The prompt shown in Fig. 8 includes information (#instructions) indicating the viewpoint of analysis and the positioning of the patent document group, the same statistical data (#aggregated results) as in Fig. 3, and information specifying the format of the answer (#output result format), and is generated by the writing generation assistance device 1.

[0026] Figure 9 shows an example of a response from the generation AI 3 to the prompt shown in Figure 8. Using the results of the overall analysis as shown in Figure 9, the attention point acquisition module 12 can acquire an accurate attention point and reason for attention, and the second prompt generation module 15 can generate an accurate second prompt.

[0027] <2-3. Obtaining points of interest and reasons for interest> Fig. 10 shows an example of a prompt generated by the noteworthy portion acquisition module 12 in the embodiment. The prompt shown in Fig. 10 includes an instruction (#instruction) to find a noteworthy portion in the statistical data based on the "problem awareness," the general analysis result (##data investigation result) obtained in Fig. 9, the same statistical data (##data) as in Fig. 3, and information specifying the format of the answer (##result output format). The "problem awareness" is information indicating a general way of looking at the statistical data depending on the purpose of the statistical processing. For example, when analyzing the trends in the number of applications as in Figs. 2 and 3, it can be information such as "Pay particular attention to the portions with large changes in the number of applications or the portions with large numbers of applications (multiple portions are possible)."

[0028] Figure 11 shows an example of a response from Production AI3 to the prompt shown in Figure 10. Figure 11 lists three notable reasons: "The year when the number of patent applications reached its peak" "A year in which the number of patent applications decreased significantly" "A year in which the number of patent applications remained stable"

[0029] Furthermore, the application year indicating the point of interest for each of the reasons for attention is shown in Figure 11. The point of interest for each of the reasons for attention may be a period spanning multiple years.

[0030] <2-4. Acquiring text data> 12 shows an example of text data acquired by the text acquisition module 13 in the embodiment. The text acquisition module 13 acquires text data of a plurality of patent documents corresponding to the noted passages for each of the noted reasons.

[0031] The text data does not have to be the entire text of each patent document. It can be only the "abstract," or only the "abstract" and "claims," or it can also include "embodiments," or it can be a portion up to a predetermined number of characters from the beginning. Alternatively, the text data can be obtained by having the generation AI 3 summarize the entire text of each patent document. It can also include various attribute data of the patent document, such as the applicant's name and application date, converted into text.

[0032] <2-5. Generating explanatory text> 13 shows an example of a first prompt generated by the first prompt generation module 14 in the embodiment. The first prompt includes, for each reason for attention, a portion of attention and a reason for attention (see FIG. 11), text data (see FIG. 12), and an instruction to generate a sentence explaining what kind of technology it is. The first prompt may also include information on "problem awareness" (see FIG. 10).

[0033] Figure 14 shows an example of a response from the generation AI3 to the first prompt shown in Figure 13. In Figure 14, for the years in which the number of patent applications filed by Company A peaked (2010 and 2011), an explanatory text is generated that incorporates the content of the text data, rather than simply analyzing the number of applications. It is desirable that the first prompt be generated for other reasons for attention, and that explanatory text be generated by the generation AI3.

[0034] <2-6. Generate Report> 15 shows an example of a second prompt generated by the second prompt generation module 15 in the embodiment. The second prompt includes instructions to create a report on a group of patent documents (##Tasks you should perform), the overall analysis results obtained in FIG. 9 (###Data investigation results), and explanatory text obtained in FIG. 14 for each reason for attention (####Year when the number of patent applications peaked). The second prompt includes multiple combinations of reasons for attention and explanatory text.

[0035] Figure 16 shows an example of a portion of the response from generation AI3 to the second prompt shown in Figure 15. In Figure 16, a report with multifaceted and in-depth content is generated by incorporating the content of the text data for each reason for attention obtained based on statistical data showing the number of patent applications filed by Company A by each filing year.

[0036] The generation AI 3 may attempt to generate an answer using all the information provided evenly, but in the case of time-series data such as those shown in Figures 2 and 3, users may have little interest in information that is too old. Therefore, a statement specifying the emphasis of the report depending on the type of statistical processing, such as "Please pay particular attention to recent conditions," may be added to the second prompt shown in Figure 15.

[0037] <2-7. Multiple statistical processing> 8 to 16 have been described with reference to cases where a reason for attention and a location of interest are obtained using the results of one statistical process shown in FIGS. 2 and 3, but embodiments are not limited to this. Other reasons for attention and locations of interest may be obtained using the results of the statistical processes shown in FIGS. 4 to 7, or a report including more multifaceted content may be generated by obtaining reasons for attention and locations of interest from the results of each of multiple statistical processes. Even when a system using the results of multiple statistical processes is implemented, buttons may be displayed in association with each of the graphs shown in FIGS. 2 and 6, for example, so that clicking one of the buttons creates a report using only the results of the corresponding graph.

[0038] <2-8.Other> The prompt for a general analysis of statistical data shown in Fig. 8, the prompt for answering points of interest and reasons for attention shown in Fig. 10, and the first and second prompts shown in Fig. 13 and 15 may each be divided into multiple prompts. The generation AI3 may perform processing multiple times by answering multiple prompts in sequence.

[0039] <3. Effects> (1) According to an embodiment, a text generation support device 1 that supports the generation of text that explains a group of patent documents includes a statistical data acquisition module 11, a noteworthy part acquisition module 12, a text acquisition module 13, and a first prompt generation module 14. The statistical data acquisition module 11 acquires statistical data indicating the results of statistical processing performed on a group of patent documents. The noteworthy part acquisition module 12 acquires information for identifying a noteworthy part in the statistical data and a noteworthy reason for noticing the noteworthy part based on the statistical data. The text acquisition module 13 acquires text data for each of a plurality of patent documents included in the patent document group and corresponding to the portion of interest, based on information identifying the portion of interest. The first prompt generation module 14 generates a first prompt that causes the generation AI 3 to output an explanatory sentence based on the attention portion, the attention reason, and the text data.

[0040] Even if the generation AI 3 is given a patent document included in the group of patent documents to be analyzed and asked to generate an explanatory text, the result may be a mere summary. Even if the generation AI 3 is given statistical data and asked to generate an explanatory text, the result may be a superficial explanatory text that can be understood by looking at a graph. According to an embodiment, by acquiring notable points and reasons for attention based on statistical data, acquiring text data corresponding to the notable points, and providing these notable points, reasons for attention, and text data to the generation AI 3, it is possible to generate an explanatory text that covers the main points and goes into specific content.

[0041] (2) According to the embodiment, the writing production assistance device 1 further includes a second prompt generation module 15. The noteworthy part acquisition module 12 acquires, based on the statistical data, information specifying a plurality of noteworthy parts in the statistical data and a noteworthy reason for each noteworthy part. The first prompt generation module 14 generates a first prompt that causes the generation AI 3 to output an explanation for each attention reason based on each attention part, attention reason, and text data. The second prompt generation module 15 generates a second prompt that causes the generation AI 3 to output a report that integrates the explanatory texts output for each of the multiple reasons for attention.

[0042] According to this, by providing the attention points, attention reasons, and text data for a plurality of attention reasons to the generation AI 3, it is possible to generate commentary that incorporates specific content from a variety of perspectives.

[0043] (3) According to the embodiment, The statistical data acquisition module 11 acquires statistical data indicating the results of a plurality of statistical processes performed on a group of patent documents for each of the plurality of statistical processes. The noteworthy point acquisition module 12 acquires, based on the statistical data, information specifying a plurality of noteworthy points including a noteworthy point in the statistical data corresponding to each of the plurality of statistical processes, and a noteworthy point reason for focusing on each noteworthy point. The first prompt generation module 14 generates a first prompt that causes the generation AI 3 to output an explanation for each attention reason based on each attention part, attention reason, and text data. The second prompt generation module 15 generates a second prompt that causes the generation AI 3 to output a report that integrates the explanatory texts output for each of the multiple reasons for attention.

[0044] According to this, by providing the generation AI3 with the points of interest, reasons for interest, and text data for multiple reasons for interest obtained based on the results of multiple statistical processes, it is possible to generate explanatory text that incorporates specific content from a variety of perspectives.

[0045] (4) According to the embodiment, the interest part acquisition module 12 generates a third prompt that causes the generation AI 3 to respond with information that identifies the interest part and the reason for attention.

[0046] In order to obtain the points of interest and the reasons for attention through internal processing in the sentence generation support device 1, individual programming or parameter setting is required depending on the type of statistical processing, but by having the generation AI 3 answer the questions, the amount of work required for programming or parameter setting can be reduced.

[0047] (5) According to an embodiment, the third prompt includes information indicating a general view of the statistical data according to the purpose of the statistical processing.

[0048] This allows the generation AI3 to accurately generate the attention portion and the attention reason. [Explanation of symbols]

[0049] 1...Text generation support device, 2...Patent database, 3...Generation AI, 11...Statistical data acquisition module, 12...Notable part acquisition module, 13...Text acquisition module, 14...First prompt generation module, 15...Second prompt generation module

Claims

1. A text generation support device that supports the generation of text that explains a group of patent documents, a statistical data acquisition module for acquiring statistical data indicating the results of statistical processing performed on the group of patent documents; an attention point acquisition module that acquires, based on the statistical data, information that identifies an attention point in the statistical data and an attention reason for focusing on the attention point; a text acquisition module that acquires text data for each of a plurality of patent documents included in the patent document group that correspond to the portion of interest based on the information identifying the portion of interest; a first prompt generation module that generates a first prompt that causes a first generation AI to output an explanatory sentence based on the attention portion, the attention reason, and the text data; A sentence generation assistance device comprising:

2. A text generation support device that supports the generation of text that explains a group of patent documents, a statistical data acquisition module for acquiring statistical data indicating the results of statistical processing performed on the group of patent documents; an attention point acquisition module that acquires, based on the statistical data, information that identifies a plurality of attention points in the statistical data and an attention reason for each attention point; a text acquisition module that acquires text data for each of a plurality of patent documents included in the patent document group, the plurality of patent documents corresponding to each of the plurality of noteworthy portions, based on information identifying each of the noteworthy portions; a first prompt generation module that generates a first prompt, causing a first generation AI to output an explanatory sentence for each of the attention reasons based on the each of the attention points, the attention reasons, and the text data; a second prompt generation module that generates a second prompt that causes a second generation AI to output a report that integrates the explanatory text output for each of the plurality of reasons for attention; A sentence generation assistance device comprising:

3. A text generation support device that supports the generation of text that explains a group of patent documents, a statistical data acquisition module that acquires statistical data indicating results of performing a plurality of statistical processes on the group of patent documents for each of the plurality of statistical processes; an attention point acquisition module that acquires, based on the statistical data, information identifying a plurality of attention points including an attention point in the statistical data corresponding to each of the plurality of statistical processes and an attention reason for each attention point; a text acquisition module that acquires text data for each of a plurality of patent documents included in the patent document group, the plurality of patent documents corresponding to each of the plurality of noteworthy portions, based on information identifying each of the noteworthy portions; a first prompt generation module that generates a first prompt, causing a first generation AI to output an explanatory sentence for each of the attention reasons based on the each of the attention points, the attention reasons, and the text data; a second prompt generation module that generates a second prompt that causes a second generation AI to output a report that integrates the explanatory text output for each of the plurality of reasons for attention; A sentence generation assistance device comprising:

4. In any one of claims 1 to 3, The attention point acquisition module generates a third prompt that causes a third generation AI to respond with information identifying the attention point and the reason for the attention. Sentence generation support device.

5. In claim 4, the third prompt includes information indicating a general view of the statistical data according to the purpose of the statistical processing; Sentence generation support device.

Images