Technical problem solution exploration support system, technical problem solution exploration support method, and technical problem solution exploration support program

Abstract

This system provides a technical problem solution search support system that searches for literature based on the directness of the technical problem described in natural language, thereby suppressing premature convergence on peripheral literature. [Solution] The technical problem solution search support system determines the degree of specificity of a technical problem in natural language based on a specificity score, presents a specificity question if the degree of specificity does not meet predetermined conditions, and generates or regenerates a problem characteristic phrase based on additional input. Furthermore, it identifies the problem description area in the patent document, uses the correspondence between the matching feature quantity containing the problem characteristic phrase and the description in the candidate document, evaluates the correspondence in the problem description area highly, calculates the degree of direct impact, sets a search starting document based on the degree of problem matching or the degree of target matching, and tracks and classifies documents starting from that search starting document.

Description

Technical Field

[0001] The present invention relates to a technical problem solution search support system, a technical problem solution search support method, and a technical problem solution search support program that receive a development-related technical problem as an input, search published patent documents that contribute to solving the technical problem, organize and present them as technical reference information.

Background Art

[0002] Conventionally, since patent documents systematically describe technical problems, solving means, functions and effects, it has been common practice to use patent documents as technical information sources and search for documents that contribute to solving development-related technical problems.

[0003] In addition, as search and analysis methods targeting patent documents, there are known techniques including: a technology for searching patent documents using natural language processing or machine learning (Patent Document 1), a technology for extracting descriptive sentences representing concepts and related words from patent specifications to search for similar patent specifications (Patent Document 2), and a technology for creating a search formula using a reference set of patent documents and classification information to extract matching documents (Patent Document 3).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Patent Document 3

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, the technical problems entered by users often consist of terms used in the field or abstract natural language, and searches using the entered terms as they are tend to miss literature directly addressing the problem. In addition, patent documents sometimes describe object names, problems, structures, or effects using higher-level conceptual expressions, and there was a problem in that the object names and terms representing problems entered by users did not always match the expressions in the literature.

[0006] Furthermore, similarity searches targeting the full text or claims of documents had the problem that superficial co-occurrence of words or general semantic similarity tended to dominate, rather than determining whether the documents dealt with the same type of problem, making it difficult to evaluate the direct relevance of the problem. In addition, once a single related document was obtained, the search would quickly converge on peripheral documents, making it difficult to systematically grasp documents that directly address the core of the problem and the surrounding documents. Moreover, when using generative methods, there was a risk of presenting identifiers for non-existent documents or incorrectly attributing the content of documents.

[0007] Therefore, the present invention aims to provide a technical problem solution search support system, a technical problem solution search support method, and a technical problem solution search support program that can identify literature based on the directness of the problem, even from technical problems described in abstract or field-specific natural language, and can suppress premature convergence to peripheral literature. [Means for solving the problem]

[0008] A technical problem solution search support system according to one aspect of the present invention includes: an input receiving unit that receives a technical problem described in natural language from a user; a specificity determination unit that determines the degree of specificity of the technical problem based on the number of fulfilled attribute categories extracted and a specificity score including the specificity of problem phenomenon words, and if the degree of specificity does not meet predetermined conditions, generates a specificity question that asks for at least one of the object, problem phenomenon, current configuration, or expected effect regarding the missing or low specificity attribute and presents it to the user; a phrase generation unit that generates or regenerates a problem characteristic phrase based on the technical problem and additional input from the user to the specificity question; and a problem description area that describes the problems of the prior art, targets for improvement, unresolved problems, or objectives of the invention from patent documents, and uses the correspondence between matching feature quantities including the problem characteristic phrase and the descriptions in candidate documents. The system includes: a direct impact calculation unit that evaluates the response in the problem description area more highly than the response in descriptions outside the problem description area, and calculates a direct impact for each candidate document using at least the problem match and the target match; a starting point setting unit that sets a candidate document as a starting point document based on the direct impact and on the condition that the problem match or the target match satisfies predetermined conditions; a document tracking unit that tracks at least one of the cited documents, cited documents, prior art documents, and similar documents of the starting point document; a document classification unit that classifies the documents tracked by the starting point document and the document tracking unit into one of a plurality of document categories, including direct impact documents, prior art documents, cited documents, similar documents, and related documents; and an output control unit that outputs the technical key points of each classified document along with the identifier of the document.

[0009] This configuration allows for the generation or regeneration of a problem feature phrase based on the input technical problem if the input lacks sufficient specificity. This is achieved by obtaining additional input regarding the missing or low-specificity attributes through specificity questions. The regenerated problem feature phrase is then compared with the problem description area in the patent document, and based on the comparison results, candidate documents that match the problem or target object can be set as search starting documents. This makes it easier to use documents that directly correspond to the input technical problem as search starting points, compared to simple search term expansion, full-text similarity search, or claim similarity search.

[0010] The specificity determination unit calculates the specificity score based on the number of attribute categories for which significant words were obtained, the specificity of the problem phenomenon words, and whether or not the problem location, occurrence conditions or causal structure, and expected effect have been identified. Problem phenomenon words that represent general defects and do not specify the problem location, occurrence conditions or causal structure may be treated as problem phenomenon words with low specificity.

[0011] The impact calculation unit may identify the area containing the problem based on at least one of the following: a heading string or section label, a sentence containing a word indicating the problem, determination of the problem description text by a trained model, or extraction from the paragraph structure of a document page or machine translation.

[0012] The direct impact calculation unit may calculate the degree of match to the problem by multiplying the matching of words appearing in the problem description area by a coefficient α, and multiplying the matching of words appearing in the claims, examples, or detailed descriptions of solutions by a coefficient β smaller than α (including 0). This allows for a relatively higher evaluation of the match in the problem description area, unlike when the entire text of the patent document or the claims are uniformly compared.

[0013] The direct impact calculation unit may calculate the degree of match to the issue, the degree of match to the target object, the degree of match to the structure, the degree of match to the effect, and the degree of match to the feature phrase for each candidate document, and calculate the direct impact based on the weighted sum of these. The starting point setting unit may set a candidate document as the starting point for the search if the direct impact is equal to or greater than the first threshold, and at least one of the degree of match to the issue or the degree of match to the target object is equal to or greater than the second threshold.

[0014] The document tracking unit or the phrase generation unit may, if none of the retrieved candidate documents meet the conditions set as the starting document for the search, perform an additional search by changing the search term, or track citations, cited documents, or similar documents. This prevents the search from ending prematurely at the stage of the first retrieved related documents.

[0015] Prior to outputting a document identifier, the output control unit may issue a request to retrieve the document page corresponding to the identifier, set a retrieval success flag based on the retrieval response or retrieval success information, output the identifier and the technical summary associated with the retrieval result for identifiers for which the retrieval success flag has been set, and suppress the output of identifiers for which the retrieval success flag has not been set. This can prevent the output of non-existent document identifiers or document identifiers whose retrieval has not been confirmed.

[0016] A method for supporting the search for technical problem solutions according to another aspect of the present invention includes the steps of: receiving a technical problem described in natural language from a user; determining the degree of specificity of the technical problem based on the number of fulfilled attribute categories extracted and a specificity score including the specificity of problem phenomenon words, and if the degree of specificity does not meet predetermined conditions, generating and presenting concretization questions regarding the missing or less specific attributes; generating or regenerating problem characteristic phrases based on additional input to the technical problem and the concretization questions; identifying problem description areas from patent documents that describe problems of the prior art, targets for improvement, unresolved issues, or objectives of the invention, and using the correspondence between matching feature quantities including problem characteristic phrases and descriptions in candidate documents, and the correspondence in the problem description areas This method involves a computer performing the following steps: calculating a degree of direct impact for each candidate document using at least the degree of match to the issue and the degree of match to the target object, evaluating the response in the document more highly than the response in the document outside the area described in the issue; setting a candidate document as a starting document for the search based on the degree of direct impact, and on the condition that the degree of match to the issue or the degree of match to the target object meets predetermined conditions; tracking at least one of the cited documents, cited documents, prior art documents, or similar documents starting from the starting document for the search; classifying the starting document for the search and the tracked documents into one of several document categories, including direct impact documents, prior art documents, cited documents, similar documents, and related documents; and outputting the technical key points of each classified document along with the identifier of the document.

[0017] A technical problem solution search support program according to yet another aspect of the present invention includes a process for a computer to receive a technical problem described in natural language from a user; a process for determining the degree of specificity of the technical problem based on a specificity score including the number of attribute categories satisfied and the specificity of problem phenomenon words, and if the degree of specificity does not meet predetermined conditions, a process for generating and presenting a concretization question; a process for generating or regenerating a problem characteristic phrase based on additional input to the technical problem and the concretization question; and a process for identifying a problem description area from patent documents that describes the problems of the prior art, targets for improvement, unresolved issues, or objectives of the invention, and using the correspondence between matching feature quantities including the problem characteristic phrase and the descriptions in candidate documents, This program performs the following steps: evaluates the response in the problem description area more highly than the response in descriptions outside the problem description area, and calculates the degree of direct impact for each candidate document using at least the degree of match to the problem and the degree of match to the target object; sets the candidate document as the starting document for the search based on the degree of direct impact, and on the condition that the degree of match to the problem or the degree of match to the target object meets predetermined conditions; tracks at least one of the cited documents, cited documents, prior art documents, or similar documents of the starting document for the search; classifies the starting document for the search and the tracked documents into one of several document categories; and outputs the technical key points of each classified document along with the identifier of the document. [Effects of the Invention]

[0018] According to the present invention, when the level of specificity of a technical problem described in natural language does not meet predetermined conditions, it is possible to generate and present concretization questions regarding the missing or less specific attributes, and to generate or regenerate problem characteristic phrases based on additional input to these concretization questions. This makes it possible to concretize the matching features used in subsequent literature searches, even when an abstract or field-specific technical problem is input.

[0019] Further, according to the present invention, a problem description area that describes the problems of prior art in patent documents, objects to be improved, unsolved problems or the purpose of the invention is identified, the direct hit degree can be calculated by using the correspondence between the collated feature quantity including the problem feature phrase and the description of the candidate document, and evaluating the correspondence in the problem description area higher than the correspondence in descriptions other than the problem description area. This makes it easier to search for documents that directly correspond to the input technical problem, compared to similarity search that uniformly targets the full text of documents or the scope of claims.

[0020] Further, according to the present invention, a candidate document can be set as a search starting document based on the direct hit degree and on the condition that at least problem matching or object matching is recognized. This can suppress that peripheral documents lacking direct hit to the problem or the object are set as search starting documents.

[0021] Further, according to the present invention, starting from the search starting document, at least one of cited documents, citing documents, prior art documents and similar documents can be traced, and the traced documents can be classified into any of a plurality of document categories. This makes it easy to systematically grasp the documents directly corresponding to the problem and the surrounding document group according to document categories such as direct hit documents, prior art documents, citing documents, similar documents or peripheral documents.

[0022] Further, according to one aspect of the present invention, before outputting a document identifier, an acquisition request for a document page corresponding to the identifier is issued, an acquisition success flag is set based on an acquisition response or acquisition success information, and the identifier for which the acquisition success flag is set can be output together with technical key points associated with the acquisition result. This can suppress output of non-existent document identifiers or document identifiers for which acquisition has not been confirmed, and can improve the reliability of the output document identifiers and document contents.

[0023] However, Patent Documents 1 to 3 mentioned above do not sufficiently disclose a specific configuration that enables specifying documents based on the direct impact of a problem, suppressing early convergence to peripheral documents, and ensuring the reliability of identifiers and contents of the presented documents, even when the technical problem is described in an abstract or on-site natural sentence. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] [Figure 1] FIG. 1 is a block diagram showing an overall configuration of a technical problem solution search support system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a flowchart showing an overall process executed by the technical problem solution search support system. [Figure 3] FIG. 3 is a flowchart showing processes related to specificity determination, generation of a specific question, reception of an additional input, and regeneration of a problem feature phrase. [Figure 4] FIG. 4 is a diagram showing a data structure used for specifying a problem description area, performing weighted matching on the problem description area, and calculating a direct impact degree. [Figure 5] FIG. 5 is a diagram showing tracing of a citation network starting from a search starting document and classification into document categories. [Figure 6] FIG. 6 is a flowchart showing processes related to an output verification gate including confirmation of acquisition of a document identifier and output suppression. [Figure 7] FIG. 7 is a diagram showing an example of output including document categories, document identifiers, technical main points, and distinction between facts and estimations or technical suggestions. DESCRIPTION OF EMBODIMENTS

[0025] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The embodiment described below is an example embodying the present invention, and does not limit the technical scope of the present invention. In the following description, the same or corresponding components are denoted by the same reference numerals, and overlapping description may be omitted.

[0026] <1. System Configuration> As shown in Figure 1, the technical problem solution search support system 1 according to this embodiment comprises a search support server 10 and a user terminal 30. The search support server 10 and the user terminal 30 are connected to each other via a network so that they can communicate with each other. The search support server 10 is configured to acquire patent documents or document pages corresponding to patent documents from a patent document source 40. In addition, the search support server 10 may, if necessary, use a trained language model 50 to perform natural language analysis, generate feature phrases, determine problem description areas, or classify document categories.

[0027] The search support server 10 comprises an input reception unit 11, a specificity determination unit 12, a phrase generation unit 13, a direct impact calculation unit 14, a starting point setting unit 15, a literature tracking unit 16, a literature classification unit 17, an output control unit 18, a control unit 19, and a storage unit 20. The control unit 19 includes, for example, a processor and executes a program stored in the storage unit 20 to realize the functions of the input reception unit 11, the specificity determination unit 12, the phrase generation unit 13, the direct impact calculation unit 14, the starting point setting unit 15, the literature tracking unit 16, the literature classification unit 17, and the output control unit 18.

[0028] The memory unit 20 stores a vocabulary conversion dictionary 21, a problem decomposition pattern 22, and a concretization question template 23. The vocabulary conversion dictionary 21 is a dictionary for converting problem words into function words, action words, effect words, synonyms, or superordinate conceptual words. The problem decomposition pattern 22 is a pattern for decomposing a technical problem described in natural language into attribute categories such as object, problem phenomenon, current configuration, constituent elements, action, effect, and constraints. The concretization question template 23 is a template for generating questions that request additional input from the user regarding attributes that lack sufficient or have low specificity.

[0029] The user terminal 30 receives input from the user in the form of a technical problem described in natural language and displays the literature category, literature identifier, technical key points, and technical implications output from the search support server 10. The user terminal 30 may be, for example, a personal computer, tablet, smartphone, or other information processing terminal.

[0030] The patent document source 40 is a database or web page source that provides published patent documents, registered patent documents, or corresponding document pages. The patent document source 40 is not limited to a single database, but may include multiple patent document databases, web document pages, machine translations, citation information, or similar document information provided by external services.

[0031] <2. Overall process> Refer to Figure 2 to explain the overall process performed by the technical problem solution search support system 1.

[0032] First, the input receiving unit 11 receives a technical problem described in natural language from the user via the user terminal 30 (ST1). The technical problem may be in natural language that includes the name of the object and an abstract problem phenomenon, such as, "It is difficult to walk with a walking aid. Are there any ways to improve it?"

[0033] The specificity determination unit 12 extracts attribute categories from the input technical problem, calculates a specificity score that includes the number of satisfied attribute categories and the specificity of the problem phenomenon words, and determines the specificity of the technical problem based on the specificity score (ST2). The specificity determination unit 12 determines whether the specificity satisfies predetermined conditions (ST3).

[0034] If the level of specificity does not meet the predetermined conditions, the specificity determination unit 12 generates and presents to the user a specificity question asking for at least one of the following attributes that are missing or have low specificity: the object, the problem phenomenon, the current configuration, or the expected effect (ST4). The input reception unit 11 accepts additional input from the user in response to the specificity question (ST5). The phrase generation unit 13 generates or regenerates a problem feature phrase based on the input technical problem and additional input, and updates the matching feature set Q60 (ST6). After that, the process proceeds again to specificity determination or candidate literature acquisition.

[0035] If the level of specificity meets the predetermined conditions, the direct impact calculation unit 14 obtains candidate documents from the patent document source 40 and identifies the problem description area for each candidate document (ST7). The direct impact calculation unit 14 uses the correspondence between the matching feature set Q60 and the description in the candidate document, and evaluates the correspondence in the problem description area more highly than the correspondence in the description outside that area, to calculate the direct impact for each candidate document (ST8).

[0036] The starting point setting unit 15 determines whether a candidate document satisfies the starting point conditions based on the degree of direct impact and on the condition that at least a match for the issue or a match for the target object is recognized (ST9). If no candidate document satisfies the starting point conditions, the phrase generation unit 13 or the document tracking unit 16 performs an additional search to reacquire candidate documents (ST10).

[0037] In ST10, if no candidate documents are obtained, the phrase generation unit 13 may perform additional searches based on changes to search terms, paraphrasing, broadening of concepts, or addition of foreign language expressions. Furthermore, if candidate documents are obtained but none satisfy the starting conditions, the document tracking unit 16 may track citations, cited documents, or similar documents using the obtained candidate documents. This prevents the search from ending prematurely at the stage of the initially obtained relevant documents, thus avoiding convergence on peripheral documents unrelated to the core of the issue.

[0038] If there are candidate documents that meet the starting point conditions, the starting point setting unit 15 sets the candidate document as the search starting document 70 (ST11). The document tracking unit 16 tracks at least one of the following starting points: cited documents, cited documents, prior art documents, and similar documents (ST12). The document classification unit 17 classifies the search starting document 70 and the tracked documents into one of several document categories (ST13). The output control unit 18 confirms the acquisition of the document identifier using the output verification gate (ST14) and outputs the technical key points for each classified document along with the document identifier (ST15).

[0039] <3. Assessment of Specificity and Specific Questions> Referring to Figure 3, the processes related to determining the level of specificity, generating specificity questions, and regenerating the problem characteristic phrases will be explained.

[0040] The phrase generation unit 13 analyzes the input technical problem sentence and, referring to the problem decomposition pattern 22, extracts the target narrative, problem phenomenon words, words representing the current structure, constituent words, action words, effect words, and words representing constraints. For example, if the input sentence contains the word "walker," that word may be extracted as the target narrative. Also, if the word "difficult to walk" is included, that word may be extracted as a problem phenomenon word.

[0041] The specificity determination unit 12 determines the degree of specificity based not only on the number of satisfied attribute categories extracted, but also on a specificity score that includes the specificity of the problem phenomenon words. The specificity score is calculated, for example, based on the number of satisfied attribute categories from which significant words were obtained, the specificity of the problem phenomenon words, whether or not the problem area is identified, whether or not the occurrence conditions or cause structure is identified, and whether or not the expected effect is identified.

[0042] Here, words expressing general problems such as "difficult to walk," "difficult to use," or "doesn't work properly" are extracted as problem phenomenon words, but if the problem location, conditions for occurrence, or causal structure are not specified, they are treated as problem phenomenon words with low specificity. Therefore, even if a target narrative is obtained, inputs where the problem phenomenon is abstract and the problem location, causal structure, or expected effect are unclear may be judged as having low specificity.

[0043] If the level of specificity does not meet the predetermined conditions, the level of specificity determination unit 12 identifies the attribute categories that are lacking or have low specificity, and generates specificity questions by referring to the specificity question template 23. Specificity questions are, for example, questions asking about the specific content of the problem phenomenon, the location where the problem occurs, the current configuration, the configuration considered to be the cause, or the expected effect. The number of specificity questions may be, for example, no more than a predetermined number, preferably a maximum of three questions.

[0044] When the user answers a specificity question, the phrase generation unit 13 accepts the answer as additional input and extracts the object, problem phenomenon, current configuration, components, action, or effect included in the additional input. Then, the phrase generation unit 13 generates or regenerates a problem characteristic phrase based on the additional input and updates the matching feature set Q60. In this way, the specificity determination, specificity question, additional input, and feature update are linked as a series of processes, and the matching features used in the subsequent directness calculation are made specific.

[0045] The phrase generation unit 13 may refer to the vocabulary conversion dictionary 21 and generate, in addition to the object name included in the technical problem, at least one of the object's higher-level conceptual name, usage name, function name, product name or method name after problem resolution, or a foreign language expression thereof. The generated word or phrase is included in the problem characteristic phrase. This helps to suppress search omissions due to discrepancies between the object name entered by the user and the invention name or technical field name described in the patent document.

[0046] <4. Identifying the problem area and calculating the degree of impact> Referring to Figure 4, we will explain how to identify the problem area and calculate the degree of impact.

[0047] The direct impact calculation unit 14 identifies the problem description area for each candidate document. The problem description area refers to the area in the patent document that describes the problems, areas for improvement, unresolved issues, or objectives of the invention of the prior art. The problem description area is not limited to areas with explicit headings such as "Problem to be Solved by the Invention," "Background Art," "Summary," or "Background." The problem description area may also be extracted from the text of the document as sentences or paragraphs that describe the shortcomings, problems, issues, objectives, or areas for improvement of the prior art.

[0048] Identifying the area describing the problem is performed based on at least one of the following methods: extraction based on heading strings or section labels, extraction of sentences containing words indicating the problem, determination of problem-described sentences using a trained model, or extraction from the paragraph structure of the literature page or machine translation. Words indicating the problem may include, for example, "problem," "objective," "conventional," "insufficient," "difficult," and "improvement."

[0049] The direct impact calculation unit 14 compares the matching feature set Q60 with the corresponding sections of the candidate literature. The matching feature set Q60 includes, for example, Q_object, Q_issue, Q_configuration, Q_effect, and Q_phrase. Q_object is a feature related to the object, Q_issue is a feature related to the issue phenomenon or issue content, Q_configuration is a feature related to the current configuration or causal configuration, Q_effect is a feature related to the expected effect, and Q_phrase is a feature related to the issue feature phrase.

[0050] The direct impact calculation unit 14 calculates the issue match S1, object match S2, construct match S3, effect match S4, and feature phrase match S5 for each candidate document. Each match may be calculated as a weighted overlap between corresponding word sets, a cosine similarity between vectorized features, or an evaluation value by a trained model.

[0051] In calculating the degree of matching to the problem S1, the direct impact calculation unit 14 may multiply the matching of words appearing in the problem description area by a coefficient α, and multiply the matching of words appearing in the claims, examples, or detailed description of the solution by a coefficient β smaller than α. The coefficient β may also be 0. This allows for a relatively higher evaluation of matching in the problem description area, unlike when the entire text of the candidate document or the claims are uniformly compared.

[0052] The degree of direct impact D is calculated, for example, as a weighted sum of the degree of match with the problem S1, the degree of match with the target object S2, the degree of match with the structure S3, the degree of match with the effect S4, and the degree of match with the feature phrase S5. As an example, the degree of direct impact D is calculated as D = w1·S1 + w2·S2 + w3·S3 + w4·S4 + w5·S5. Here, w1 to w5 are weights, and for example, they may satisfy Σwi = 1. If the directness of the problem is emphasized, w1 may be set to be larger than the other weights.

[0053] In this specification, "problem matching" means that the problem matching degree S1 is at or above a predetermined level, or that the correspondence with respect to the problem is recognized by a predetermined evaluation label such as "applicable" or "medium" or higher. Furthermore, "object matching" means that the object matching degree S2 is at or above a predetermined level, or that the correspondence with respect to the object is recognized by a predetermined evaluation label such as "applicable" or "medium" or higher.Therefore, problem matching and object matching may be determined based on a numerical degree of matching or based on a qualitative evaluation label.

[0054] The starting point setting unit 15 sets candidate documents as search starting documents 70 based on the degree of direct impact D, and on the condition that at least a match to the problem or a match to the target object is recognized. As a numerical example, the starting point setting unit 15 sets candidate documents as search starting documents 70 if the degree of direct impact D is equal to or greater than the first threshold Th1, and at least one of the degree of match to the problem S1 or the degree of match to the target object S2 is equal to or greater than the second threshold Th2. This prevents candidate documents that have a high degree of direct impact D but lack a match to the problem or target object from being set as search starting documents 70.

[0055] Furthermore, the degree of match and degree of direct impact are not limited to numerical values ​​and may be evaluated using labels such as "high," "medium," "low," or "applicable" or "not applicable." In this case, the starting point setting unit 15 may, for example, set a candidate document 70 as the starting point document for the search if the degree of direct impact is "high" and the degree of match with the issue or the degree of match with the target object is "medium" or higher.

[0056] <5. Citation Network Tracking and Literature Classification> Refer to Figure 5 to explain citation network tracking and literature classification.

[0057] The document tracking unit 16 tracks documents related to the search starting document 70, starting from the search starting document 70. The documents to be tracked may include cited documents 71 cited in the search starting document 70, cited documents 72 that cite the search starting document 70, similar documents 73 that have similar content to the search starting document 70, or prior art documents described as background technology for the search starting document 70. Here, prior art documents refer to documents among the cited documents 71 that show the conventional structure or prerequisite technology of the search starting document 70, or documents described as prior art in the background technology section of the search starting document 70.

[0058] The document classification unit 17 classifies the search-starting documents 70 and the documents tracked by the document tracking unit 16 into one of several document categories. These document categories include, for example, direct references, prior art documents, cited documents, similar documents, and related documents 74.

[0059] As an example, the document classification unit 17 classifies documents with high problem matching degree S1 and target matching degree S2 as direct hit documents. The document classification unit 17 also classifies documents cited in the search starting document 70 and showing the conventional structure or prerequisite technology of the search starting document 70 as prior art documents. Furthermore, the document classification unit 17 may classify later application documents that cite the search starting document 70 as cited documents, documents with similar objects or structures to the search starting document 70 as similar documents, and documents that do not directly correspond to the core of the problem as peripheral documents 74.

[0060] This allows users to distinguish between documents directly related to the entered technical problem, prerequisite prior art documents, cited documents from later applications, similar documents, and related documents, rather than simply viewing a list of documents sorted by similarity.

[0061] <6. Output Verification Gate> The output verification gate will be explained with reference to Figure 6.

[0062] Prior to outputting the document identifier, the output control unit 18 performs information processing that involves verifying the identifier. Specifically, the output control unit 18 maintains a candidate document identifier for each candidate document. The candidate document identifier may include the publication number, registration number, application number, URL of the document page, or corresponding identification information.

[0063] The output control unit 18 issues a request to retrieve the document page corresponding to the candidate document identifier. The retrieval request may be, for example, a request to access a URL generated based on the candidate document identifier, a search request to the patent document source 40, or a query request to an external database.

[0064] The output control unit 18 sets an acquisition success flag based on the acquisition response or acquisition success information. The acquisition success information includes, for example, information indicating that the literature page was successfully acquired, that the acquired literature text or metadata contains information corresponding to a candidate literature identifier, or that the acquired page corresponds to a predetermined literature.

[0065] The output control unit 18 stores the acquired document text or metadata as the acquisition result and extracts the key technical points from the descriptions of problems, solutions, actions, or effects included in the acquisition result. The output control unit 18 sets the output permission flag to allow for identifiers for which the acquisition success flag has been set, and sets the output permission flag to disallow for identifiers for which the acquisition success flag has not been set.

[0066] The output control unit 18 outputs the identifier and the technical key points associated with the acquisition result for identifiers for which the output permission flag is set to permission. On the other hand, the output control unit 18 suppresses the output of identifiers for which the output permission flag is set to permission. This prevents the output of non-existent document identifiers or document identifiers whose acquisition has not been confirmed.

[0067] The output control unit 18 may output the information described in the acquisition results and the estimations or technical suggestions based on said information in a manner that allows for distinction. For example, items explicitly described in the acquisition results may not be given any special label, items estimated from the acquisition results may be labeled "estimated," and directions for consideration derived from the acquisition results may be labeled "technical suggestion."

[0068] <7. Output Example> Figure 7 shows an example of output from the output control unit 18. The output control unit 18 displays, for example, the document identifier, applicant, technical summary, and technical implications for each document category, such as direct reference documents, prior art documents, and related documents. The document identifier may also be displayed as a link corresponding to the document page for which acquisition has been confirmed.

[0069] As shown in Figure 7, the output control unit 18 may display information based on literature descriptions in a way that distinguishes between estimations or technical suggestions. The output control unit 18 may also display a note indicating that the output is non-exclusive technical reference information.

[0070] Furthermore, if the input from the user or the generated output corresponds to a determination of whether the technology is usable, whether it infringes a patent right, or whether it is freely implementable, the output related to such determination may be suppressed and output limited to technical reference information.

[0071] <8. Implementation examples using generative language models> At least one of the specificity determination unit 12, the phrase generation unit 13, the directness calculation unit 14, and the literature classification unit 17 may be implemented by numerical or qualitative processing that refers to the vocabulary conversion dictionary 21, the problem decomposition pattern 22, and the specificity question template 23. At least one of these may also be implemented by providing command information to the trained language model 50 and causing the trained language model 50 to perform processing according to said command information. Command information is information that specifies the procedure, judgment criteria, and output format for processing the input technical problem or literature to the trained language model 50, and can be written in natural language, structured text, or a combination thereof.

[0072] The instruction information specifies, for example, at least one of the following: (a) criteria for extracting attribute categories from a technical problem and determining the degree of specificity based on the number of fulfilled items and the specificity of the problem phenomenon words; (b) a procedure for generating a predetermined number or fewer clarification questions that ask about missing or less specific attributes if the degree of specificity does not meet predetermined conditions; (c) criteria for matching problem characteristic phrases with the problem description area with a focus on that area and giving higher value to the match in the problem description area than to the claims, examples, or detailed descriptions of solutions; (d) conditions for setting candidate documents in which a match for the problem or subject is found as the starting document for the search; (e) criteria for tracking documents starting from the starting document and classifying them into document categories; (f) a rule for confirming the acquisition of the document page corresponding to the identifier before outputting the identifier of the document, outputting only identifiers whose acquisition has been confirmed along with the technical key points associated with the acquisition results, and suppressing the output of identifiers whose acquisition has not been confirmed; and (g) a rule for distinguishing between matters described in the acquisition results and estimations or technical implications based on said descriptions when outputting.

[0073] Furthermore, the search support server 10 may provide knowledge information corresponding to the vocabulary conversion dictionary 21, the problem decomposition pattern 22, and the concretization question template 23 as a knowledge file that the trained language model 50 can refer to. In other words, the rules or vocabulary used for vocabulary conversion, problem decomposition, and generation of concretization questions may be provided separately from the command information as referable knowledge information.

[0074] Even when using a pre-trained language model 50 in this way, the technical details of the process performed are as follows: <3> from <6> This is the same as the processes described in [previous section]. Therefore, a person skilled in the art can implement this embodiment without excessive trial and error by providing the above-mentioned command information and knowledge information to the trained language model 50. The specific descriptions of the command information and knowledge information, as well as the specific values ​​of the specificity score, coefficients α and β, and weights w1 to w5, are design matters that are appropriately determined according to the technical field or required accuracy.

[0075] Furthermore, each process in this embodiment is not limited to using the trained language model 50. Specificity determination, generation of problem feature phrases, identification of problem description areas, calculation of direct impact, starting point setting, literature tracking, literature classification, and output verification may be implemented by rule-based processing, dictionary lookup processing, statistical processing, machine learning processing, or a combination thereof.

[0076] <9. Examples> Next, we will describe specific examples.

[0077] If the input is "The walker is difficult to walk with. Are there any ways to improve it?", then "walker" is extracted as the target story. On the other hand, "difficult to walk with" extracted as the problem phenomenon word is a general term for a problem that does not specify the problem location, the conditions under which it occurs, or the cause structure. Furthermore, no significant words are obtained in the categories of current structure, constituent elements, or expected effects. For this reason, the specificity determination unit 12 determines that the specificity score for the input is low and that the specificity does not meet the predetermined conditions.

[0078] In this case, the specificity determination unit 12 generates and presents specific questions to the user, for example, asking about the specific details of the difficulty in walking, the current configuration, or the areas of improvement that the user would like to prioritize. If the user inputs, for example, "It is an electric assist type, and the battery and control device are located at the feet, making it difficult to move the legs forward," the phrase generation unit 13 generates additional feature phrases based on this additional input, such as electric assist walker, battery unit, control unit, and securing space for swinging the legs, and updates the matching feature set Q60.

[0079] Subsequently, the direct hit calculation unit 14 uses the updated matching feature set Q60 to focus on matching the problem description area of ​​the candidate document. For example, if the problem description area of ​​the candidate document describes a problem where mounting the battery unit and control unit inside the vehicle frame leaves insufficient space to swing out the legs, then the problem match S1 and object match S2 can be calculated to be high for that candidate document. In this case, the starting point setting unit 15 sets that candidate document as the search starting document 70.

[0080] The document tracking unit 16 tracks cited documents, cited documents, or similar documents, starting from the search starting document 70. The document classification unit 17 may classify the search starting document 70 as a direct reference document, classify documents cited as background technology for the search starting document 70 as prior art documents, and classify other documents relating to the same type of electric walking assist device as related documents 74. The output control unit 18 confirms the acquisition of document pages for each document and then outputs the document identifier, technical summary, and technical implications.

[0081] The effects of this embodiment can be confirmed, for example, by the following evaluation method. As an evaluation dataset, a set of technical problems in natural language spanning multiple technical fields is prepared, and for each problem, a set of directly relevant literature is assigned as the correct answer set. Then, keyword search using input words, search applying only lexical conversion, general semantic similarity search on the full text of the literature, and search according to this embodiment are performed on the same dataset.

[0082] The evaluation metrics may include, for example, the percentage of tasks in which the correct answer was directly identified within a predetermined number of results, the percentage of tasks in which the correct answer could not be identified if the search was terminated at the stage of the first related literature obtained, the percentage of tasks in which the distinction between the correct answer and related literature was incorrect, and the percentage of tasks in which the acquisition of the corresponding literature page was confirmed from the outputted identifiers.

[0083] Furthermore, this embodiment allows for the measurement of similar indicators for configurations that exclude specificity determination and feature phrase regeneration, focused matching to the problem description area, or the requirement of problem matching or object matching in the starting conditions, and the changes in the indicators with and without each configuration can be compared. This allows for the confirmation that focused matching to the problem description area contributes to the presentation of literature directly corresponding to the problem, that the starting conditions and additional search processing contribute to suppressing premature convergence to peripheral literature, that specificity determination and feature phrase regeneration contribute to improving search accuracy in low-specificity inputs, and that the output verification gate contributes to the output of document identifiers with supporting evidence.

[0084] <10. Variation> The present invention is not limited to the embodiments described above. For example, the coefficients α, β, weights w1 to w5, the first threshold Th1, and the second threshold Th2 may be set as appropriate depending on the technical field, the number of documents, the required search accuracy, or the user's purpose.

[0085] Furthermore, the patent document source 40 is not limited to databases that provide published patent documents, but may also include information sources that provide registered patent documents, utility model documents, foreign patent documents, non-patent documents, technical documents, or web-based document pages.

[0086] Furthermore, the output literature categories are not limited to direct references, prior art references, cited references, similar references, and related references. Depending on the user's purpose or the target technology field, the literature categories may also include basic references, applied references, alternative technology references, improved technology references, or references.

[0087] Furthermore, specific questions are not limited to those presented to the user in written form; they may also be presented as options, checkboxes, input forms, or interactive interfaces.

[0088] Although embodiments of the present invention have been described above, the present invention is not limited to the above embodiments and modifications, and various changes, substitutions, additions, or omissions can be made without departing from the spirit of the invention as described in the claims. [Explanation of Symbols]

[0089] 1. Technical Challenge Solution Exploration Support System 10. Search support server 11 Input reception section 12 Specificity judgment part 13 Phrase generation unit 14. Direct Hit Calculation Unit 15 Starting point setting section 16. Literature Tracking Department 17 Literature Classification Department 18 Output control unit 19 Control Unit 20 Memory section 21 Vocabulary Conversion Dictionary 22 Problem Decomposition Patterns 23. Specific Question Template 30 User terminals 40 Patent document source 50 pre-trained language models 60 Matching feature set Q 70 Search starting point literature 71 References 72 Cited documents 73 Similar documents 74 Related Literature

Claims

1. A system for supporting the exploration of technical problem solutions, An input reception unit that accepts technical issues written in natural language from users, Regarding the aforementioned technical problem, the specificity determination unit determines the degree of specificity based on the number of fulfilled attribute categories extracted and the specificity score including the specificity of the problem phenomenon words, and if the degree of specificity does not meet predetermined conditions, generates a specificity question asking about at least one of the object, problem phenomenon, current configuration, or expected effect with respect to the missing or low specificity attribute, and presents it to the user. A phrase generation unit generates or regenerates a problem characteristic phrase based on the aforementioned technical problem and additional input from the user regarding the aforementioned specific questions. A direct hit degree calculation unit identifies a problem description area from the patent document that describes the problems of the prior art, areas for improvement, unresolved issues, or objectives of the invention, uses the correspondence between the matching feature quantity containing the problem characteristic phrase and the description in the candidate document, and evaluates the correspondence in the problem description area higher than the correspondence in the description outside the problem description area, and calculates the direct hit degree for each candidate document using at least the problem match degree and the target object match degree. A starting point setting unit sets candidate documents as search starting documents based on the degree of direct impact and on the condition that the degree of matching the issue or the degree of matching the target object satisfies predetermined conditions, A document tracking unit that tracks at least one of the cited documents, cited documents, prior art documents, and similar documents of the aforementioned search starting document, A document classification unit classifies the search-starting documents and the documents tracked by the document tracking unit into one of several document categories, including direct reference documents, prior art documents, cited documents, similar documents, and related documents. For each of the classified documents, an output control unit outputs its technical key points along with the identifier of the document. A system that supports the exploration of technical problem solutions.

2. The specificity determination unit calculates the specificity score based on the number of attribute categories for which significant words are obtained, the specificity of the problem phenomenon words, and whether or not the problem location, occurrence conditions or cause structure, and expected effect are specified, and treats problem phenomenon words that represent general defects that do not specify the problem location, occurrence conditions or cause structure as problem phenomenon words with low specificity, as described in claim 1.

3. The technical problem solution search support system according to claim 1, wherein the direct impact calculation unit identifies the problem description area based on at least one of the following: a heading string or section label, a sentence containing a word indicating the problem, determination of the problem description text by a trained model, or extraction from the paragraph structure of a document page or machine translation text.

4. The direct impact calculation unit calculates the degree of match to the problem by multiplying the matching of words appearing in the problem description area by a coefficient α, and multiplying the matching of words appearing in the claims, examples, or detailed description of solutions by a coefficient β smaller than α (including 0), according to claim 1.

5. The direct impact calculation unit calculates the degree of match to the problem, the degree of match to the target object, the degree of match to the structure, the degree of match to the effect, and the degree of match to the feature phrase for each candidate document, and calculates the direct impact based on the weighted sum of these, the technical problem solution search support system according to claim 1.

6. The technical problem solution search support system according to claim 5, wherein the starting point setting unit sets candidate documents in which the degree of direct impact is equal to or greater than a first threshold and at least one of the degree of match with the problem or the degree of match with the target object is equal to or greater than a second threshold as the starting point document for the search.

7. The technical problem solution search support system according to claim 1, wherein the document tracking unit or the phrase generation unit performs an additional search by changing the search term, or tracks cited documents, cited documents, or similar documents, if there are no candidate documents among the acquired candidate documents that satisfy the conditions set as the starting document for the search.

8. The technical problem solution search support system according to claim 1, wherein the phrase generation unit generates, in addition to the name of the object included in the technical problem, at least one of the higher-level conceptual name, application name, function name, name of the product or method after the problem is solved, and a foreign language expression thereof of the object, and includes them in the problem characteristic phrase.

9. The technical problem solution search support system according to claim 1, wherein the output control unit, prior to outputting a document identifier, issues a request to obtain a document page corresponding to the identifier, sets an acquisition success flag based on the acquisition response or acquisition success information, outputs the identifier and the technical key points associated with the acquisition result for identifiers for which the acquisition success flag has been set, and suppresses the output of identifiers for which the acquisition success flag has not been set.

10. The technical problem solution search support system according to claim 9, wherein the output control unit outputs the matters described in the acquired results and the estimations or technical suggestions based on said descriptions in a manner that allows for distinction.

11. The technical problem solution search support system according to claim 1, wherein the phrase generation unit generates the problem characteristic phrase by referring to a storage unit that stores a vocabulary conversion dictionary that converts problem words into function words, action words, effect words, synonyms, or superordinate conceptual words.

12. The technical problem solution search support system according to claim 1, wherein the specificity determination unit or the phrase generation unit extracts the attribute category by referring to a storage unit that stores problem decomposition patterns for decomposing a natural language technical problem into at least one of the following: object, problem phenomenon, current configuration, constituent elements, operation, effect, and constraints.

13. The technical problem solution search support system according to claim 1, wherein at least one of the specificity determination unit, the phrase generation unit, the directness calculation unit, and the literature classification unit is realized by processing based on command information for a trained language model.

14. The technical problem solution search support system according to claim 1, wherein the output control unit suppresses the output related to the determination of whether the technology is usable, whether it infringes a patent right, or whether it is freely implementable, when the input from the user or the generated output corresponds to the determination of whether the technology is usable, whether it constitutes patent infringement, or whether it is freely implementable, and outputs limited to technical reference information.

15. A method for supporting the exploration of solutions to technical challenges, The process involves receiving technical challenges from users, written in natural language, With respect to the aforementioned technical problem, the degree of specificity is determined based on the number of fulfilled attribute categories extracted and the specificity score including the specificity of the problem phenomenon words, and if the degree of specificity does not meet predetermined conditions, a step is made to generate and present specificity questions regarding the missing or insufficiently specific attributes, A step of generating or regenerating a problem characteristic phrase based on the aforementioned technical problem and additional input for the aforementioned specific questions, A step of identifying a problem description area from the patent document that describes the problems of the prior art, areas for improvement, unresolved issues, or objectives of the invention; using the correspondence between the matching feature quantity, which includes the problem characteristic phrase, and the description in the candidate document, and evaluating the correspondence in the problem description area more highly than the correspondence in the description outside that problem description area, a step of calculating the direct impact for each candidate document using at least the problem matching degree and the target matching degree; A step of setting candidate literature as search starting literature based on the degree of direct impact and on the condition that the degree of matching the issue or the degree of matching the target object satisfies predetermined conditions, The process involves tracing at least one of the cited documents, cited documents, prior art documents, or similar documents, starting from the aforementioned search starting document. The process involves classifying the aforementioned search-starting documents and traced documents into one of several document categories, including direct references, prior art documents, cited documents, similar documents, and related documents. The process involves outputting the technical key points of each classified document along with the identifier of that document, A computer-based method for assisting in the search for solutions to technical challenges.

16. On the computer, A process for receiving technical issues described in natural language from users, Regarding the aforementioned technical problem, the degree of specificity is determined based on the number of satisfactions in the attribute category and the specificity score including the specificity of the problem phenomenon words, and if the degree of specificity does not meet predetermined conditions, a process is performed to generate and present a specificity question. A process for generating or regenerating a problem characteristic phrase based on the aforementioned technical problem and additional input for the aforementioned specific questions, From the patent documents, identify the problem description area that describes the problems of the prior art, areas for improvement, unresolved issues, or objectives of the invention; use the correspondence between the matching feature quantity, which includes the problem characteristic phrase, and the description in the candidate document, and evaluate the correspondence in the problem description area more highly than the correspondence in the description outside that area, and for each candidate document, calculate the direct impact using at least the problem match rate and the target object match rate. A process to set candidate documents as search starting documents based on the degree of direct impact and on the condition that the degree of matching the issue or the degree of matching the target object satisfies predetermined conditions, A process of tracking at least one of the cited documents, cited documents, prior art documents, or similar documents of the aforementioned search starting document, The process involves classifying the aforementioned starting document and the traced document into one of several document categories. The process involves outputting the technical key points of each classified document along with the identifier of that document, A support program for exploring technical problem solutions to implement the required actions.

Images