A method and system for matching technical needs with patent solutions

Abstract

The application discloses a kind of technical demand and the matching method and system of patent scheme, applied to data processing technical field, method includes obtaining technical demand text and candidate patent text;The technical demand text and candidate patent text are processed, and demand side field set and patent side field set are obtained;Demand side field set is processed, and demand constraint determination result is obtained;Based on candidate patent text, patent side field set is processed, and prototype problem is generated;And the implication between source field and prototype problem is processed, and prediction bridge object is obtained;Demand side field set and prediction bridge object are analyzed, and bridge matching factor is obtained;Demand side field set and patent side field set are analyzed, and direct matching factor is obtained;Based on demand constraint determination result, bridge matching factor and direct matching factor, optimal matching result is obtained.The method provided by the application can improve the matching precision of patent retrieval.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for matching technical requirements with patented solutions. Background Technology

[0002] In technology research and development, commercialization of research results, and patent navigation scenarios, enterprises, universities, and research institutions often need to quickly locate existing technical solutions that match current technological needs from large-scale patent literature databases.

[0003] However, technical requirement texts usually focus on what problem needs to be solved, what effect needs to be achieved, and what constraints need to be imposed, while patent texts focus more on what technical means were used, what technical effects were achieved, and what claim features are defined. The two types of texts naturally occupy different semantic spaces.

[0004] Existing patent search methods mostly use keyword retrieval. These methods typically map the technical requirement text and the patent text directly to a unified representation space and then sort them based on semantic similarity. However, this method has difficulty distinguishing the role difference between problem expression and solution, and is prone to mismatch between the requirement intent and the patent technical means, thereby reducing search accuracy. Summary of the Invention

[0005] This invention provides a method and system for matching technical requirements with patent solutions, which solves the problem of semantic role mismatch caused by the difference in expression space between technical requirement text and patent text, thereby improving the semantic understanding depth and matching accuracy of patent retrieval.

[0006] To address the aforementioned technical problems, this invention provides a method for matching technical requirements with patented solutions, the method comprising: Obtain the user's technical requirements text and candidate patent texts from the literature database; Semantic extraction processing is performed on the technical requirement text and the candidate patent text to obtain a set of requirement-side fields and a set of patent-side fields. The set of demand-side fields is parsed and judged to obtain the demand constraint judgment result; Based on each candidate patent text, the set of patent-side fields is selected, and a prototype question is generated based on at least two source fields. The implication relationship between the source fields and the prototype question is verified to obtain a prediction bridging object that matches the candidate patent text. Analyze the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain the bridging matching factor; analyze the demand-side field set and the patent-side field set to obtain the direct matching factor; Based on the demand constraint determination result, the bridging matching factor, and the direct matching factor, the optimal matching result is obtained.

[0007] As one preferred embodiment, the semantic extraction processing of the technical requirement text and the candidate patent text to obtain a requirement-side field set and a patent-side field set includes: Text preprocessing and domain concept recognition processing are performed on the technical requirement text and the candidate patent text to obtain the requirement expression paradigm of the technical requirement text and the patent expression paradigm of the candidate patent text. Semantic role labeling and field recognition are performed on the aforementioned demand expression paradigm to obtain a demand-side field set, which includes at least a demand intent field, a target effect field, a constraint condition field, an application scenario field, and a technology preference field. Semantic role labeling and field identification are performed on the patent expression paradigm to obtain a set of patent-side fields. The set of patent-side fields includes at least the fields of technical problem to be solved, technical effect, technical means, claim limiting features, applicable conditions, and technical field.

[0008] As one preferred embodiment, the step of parsing and judging the demand-side field set to obtain the demand constraint judgment result includes: The fields in the demand-side field set are identified and extracted to obtain an initial constraint set; The initial constraint set is processed using natural language processing and semantic parsing techniques to obtain constraint types and logical elements; The constraint type and the logical element are logically generated and strength assigned to obtain the requirement constraint determination result.

[0009] As one preferred embodiment, the step of verifying the implication relationship between the source field and the prototype question to obtain a prediction bridging object that matches the candidate patent text includes: Using a natural language reasoning model, the implication relationship between the source field and the prototype question is validated to obtain an implication relationship score. Based on the implication relation score, the prototype problem is subjected to confidence filtering to obtain a predicted bridging object that matches the candidate patent text.

[0010] As one preferred embodiment, the analysis of the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain bridging matching factors includes: The demand-side field set and the prediction bridging object are subjected to association channel filtering to obtain an association channel set; Based on the set of associated channels, semantic similarity calculation is performed on the technical issue field and / or expected effect field in the demand-side field set and the bridging issue in the predicted bridging object to obtain the demand-bridging similarity. The implication relationship score of the predicted bridging object and the demand-bridging similarity are weighted to obtain the initial bridging matching value; The initial bridging matching value is normalized to obtain the bridging matching factor.

[0011] This invention also provides a system for matching technical requirements with patented solutions, comprising: The acquisition module is used to acquire the user's technical requirement text and candidate patent texts from the literature database; The extraction module is used to perform semantic extraction processing on the technical requirement text and the candidate patent text to obtain a set of requirement-side fields and a set of patent-side fields. The determination module is used to parse and determine the set of demand-side fields to obtain the demand constraint determination result. The verification module is used to select and process the set of patent-side fields based on each candidate patent text, generate a prototype question based on at least two source fields, and verify the implication relationship between the source fields and the prototype question to obtain a prediction bridging object that matches the candidate patent text. The matching module is used to analyze the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain the bridging matching factor; and to analyze the demand-side field set and the patent-side field set to obtain the direct matching factor. The generation module is used to obtain the optimal matching result based on the requirement constraint determination result, the bridging matching factor, and the direct matching factor.

[0012] As one preferred embodiment, the semantic extraction processing of the technical requirement text and the candidate patent text to obtain a requirement-side field set and a patent-side field set includes: Text preprocessing and domain concept recognition processing are performed on the technical requirement text and the candidate patent text to obtain the requirement expression paradigm of the technical requirement text and the patent expression paradigm of the candidate patent text. Semantic role labeling and field recognition are performed on the aforementioned demand expression paradigm to obtain a demand-side field set, which includes at least a demand intent field, a target effect field, a constraint condition field, an application scenario field, and a technology preference field. Semantic role labeling and field identification are performed on the patent expression paradigm to obtain a set of patent-side fields. The set of patent-side fields includes at least the fields of technical problem to be solved, technical effect, technical means, claim limiting features, applicable conditions, and technical field.

[0013] As one preferred embodiment, the step of parsing and judging the demand-side field set to obtain the demand constraint judgment result includes: The fields in the demand-side field set are identified and extracted to obtain an initial constraint set; The initial constraint set is processed using natural language processing and semantic parsing techniques to obtain constraint types and logical elements; The constraint type and the logical element are logically generated and strength assigned to obtain the requirement constraint determination result.

[0014] As one preferred embodiment, the step of verifying the implication relationship between the source field and the prototype question to obtain a prediction bridging object that matches the candidate patent text includes: Using a natural language reasoning model, the implication relationship between the source field and the prototype question is validated to obtain an implication relationship score. Based on the implication relation score, the prototype problem is subjected to confidence filtering to obtain a predicted bridging object that matches the candidate patent text.

[0015] As one preferred embodiment, the analysis of the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain bridging matching factors includes: The demand-side field set and the prediction bridging object are subjected to association channel filtering to obtain an association channel set; Based on the set of associated channels, semantic similarity calculation is performed on the technical issue field and / or expected effect field in the demand-side field set and the bridging issue in the predicted bridging object to obtain the demand-bridging similarity. The implication relationship score of the predicted bridging object and the demand-bridging similarity are weighted to obtain the initial bridging matching value; The initial bridging matching value is normalized to obtain the bridging matching factor.

[0016] Compared with the prior art, the beneficial effects of the present invention are at least one of the following: This invention obtains a user's technical requirement text and candidate patent texts from a literature database; performs semantic extraction processing on the technical requirement text and the candidate patent texts to obtain a demand-side field set and a patent-side field set; performs parsing and judgment processing on the demand-side field set to obtain a demand constraint judgment result; selects from the patent-side field set based on each candidate patent text, generating a prototype problem based on at least two obtained source fields; verifies the implication relationship between the source fields and the prototype problem to obtain a predicted bridging object matching the candidate patent text; analyzes the predicted bridging objects matching the demand-side field set and the candidate patent text to obtain a bridging matching factor; analyzes the demand-side field set and the patent-side field set to obtain a direct matching factor; and obtains the optimal matching result based on the demand constraint judgment result, the bridging matching factor, and the direct matching factor.

[0017] Compared with existing technologies, this invention addresses the mismatch problem caused by differences in semantic roles and expression spaces between technical requirement texts and patent texts. It first extracts the corresponding field sets from the semantics of both types of texts, then generates prototype questions based on the patent-side fields and verifies the implication relationships to obtain predicted bridging objects, thus completing the intermediate carrier for semantic transformation. Subsequently, it calculates bridging matching factors and direct matching factors based on the original fields, while incorporating requirement constraint judgment results for multi-dimensional fusion analysis. This accurately identifies and adapts to the semantic role differences between the two types of texts, achieving bidirectional semantic alignment and effectively avoiding mismatch. This not only deepens the semantic understanding of patent retrieval but also significantly improves matching accuracy. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for matching technical requirements with patented solutions in one embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a system for matching technical requirements with patent solutions in one embodiment of the present invention; Figure label: The module consists of: 11. Acquisition module; 12. Extraction module; 13. Judgment module; 14. Verification module; 15. Matching module; and 16. Generation module. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0021] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items. Those skilled in the art will understand the specific meaning of these terms in this invention based on the specific circumstances.

[0022] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0023] One embodiment of the present invention provides a method for matching technical requirements with patented solutions. For details, please refer to [link to relevant documentation]. Figure 1 , Figure 1 The diagram shown illustrates a method for matching technical requirements with patent solutions in one embodiment of the present invention. The method includes: S1: Obtain the user's technical requirements text and candidate patent texts from the literature database; S2: Perform semantic extraction processing on the technical requirement text and the candidate patent text to obtain a set of requirement-side fields and a set of patent-side fields; S3: Parse and determine the set of demand-side fields to obtain the demand constraint determination result; S4: Based on each candidate patent text, select the set of patent-side fields, generate a prototype question based on at least two source fields, and verify the implication relationship between the source fields and the prototype question to obtain a prediction bridging object that matches the candidate patent text; S5: Analyze the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain the bridging matching factor; analyze the demand-side field set and the patent-side field set to obtain the direct matching factor; S6: Based on the requirement constraint determination result, the bridging matching factor, and the direct matching factor, the optimal matching result is obtained.

[0024] Specifically, it is necessary to obtain the technical requirements text submitted by the user, as well as the patent titles, abstracts, specifications, and claims texts from the patent literature database.

[0025] Semantic role structure extraction is performed on the technical requirement text and the candidate patent text respectively.

[0026] Specifically, semantic extraction processing is performed on the technical requirement text and the candidate patent text to obtain a demand-side field set and a patent-side field set, including: performing text preprocessing and domain concept recognition processing on the technical requirement text and the candidate patent text to obtain the demand expression paradigm of the technical requirement text and the patent expression paradigm of the candidate patent text; performing semantic role labeling and field recognition on the demand expression paradigm to obtain a demand-side field set, which includes at least a demand intent field, a target effect field, a constraint condition field, an application scenario field, and a technology preference field; performing semantic role labeling and field recognition on the patent expression paradigm to obtain a patent-side field set, which includes at least a technical problem to be solved field, a technical effect field, a technical means field, a claim limiting feature field, an applicable condition field, and a technical field field.

[0027] First, the two types of original texts are cleaned, redundancy is removed, and sentence structure is standardized preprocessed. Then, combined with patent and technology research and development of a domain-specific thesaurus and professional terminology database, the industry-specific concepts and technical terms in the text are identified and labeled. By removing invalid interference information from the text, the fixed sentence expression logic and content structure of the two types of texts are summarized.

[0028] Next, based on the standardized requirement expression paradigm, the text semantic attributes are divided sentence by sentence according to the expression logic of technical requirements. Different semantic units that represent requirement intention, target effect, constraints, application scenarios and technical preferences are classified and labeled. Then, independent semantic fields are extracted based on the labeling results to form a set of requirement-side fields.

[0029] Then, based on the writing logic of the patent text, and targeting the standardized patent expression paradigm, we distinguish the semantic modules in the marked patent that correspond to the technical problem to be solved, technical effect, technical means, claim limiting features, applicable conditions, and technical field, and filter and extract the corresponding independent fields one by one to form a set of patent-side fields.

[0030] The extraction of the required text can be performed using a prompt template-driven large language model, rule model, sequence labeling model, or a combination thereof, and the extraction method is not limited to a specific model implementation.

[0031] Before parsing and judging the demand-side field set, it is necessary to analyze the mapping relationship between the demand-side field set and the patent-side field set to obtain the mapping channel set.

[0032] Specifically, a mapping relationship analysis is performed on the demand-side field set and the patent-side field set using a preset judgment mapping matrix to obtain a mapping channel set.

[0033] The process of constructing the judgment mapping matrix includes: constructing an initial mapping framework with the demand-side field set as rows and the patent-side field set as columns; injecting rules and performing initial confidence processing on the initial mapping framework based on domain expert knowledge to obtain an expert rule matrix; inputting the expert rule matrix into a machine learning model for training and weight tuning to obtain a learning weight matrix; and performing weighted fusion processing on the basic confidence in the expert rule matrix and the learning weights in the learning weight matrix to obtain the judgment mapping matrix.

[0034] Specifically, using various fields from the demand side as the horizontal row dimension of the matrix and various fields from the patent side as the vertical column dimension of the matrix, a two-dimensional initial matrix structure with intersecting rows and columns is constructed to form a blank initial mapping framework.

[0035] Then, combining industry experts' knowledge of patent retrieval and technical semantic matching, and based on the semantic role differences and inherent expression logic of the two types of text, the semantic association logic, adaptation relationship and mapping rationality between fields are defined for each group of rows and columns in the initial framework. Next, different association types such as strong association, weak association, no association, and taboo mismatch are distinguished, and basic association rules are configured for each group of field pairings, and corresponding initial confidence scores are assigned to quantify the reliability of field mapping. The rule filling and score labeling of all positions are completed, and finally an expert rule matrix is ​​formed.

[0036] Next, we collect a large amount of real technical requirements and patent matching sample data. We use historical valid matching and invalid mismatch field pairs as training samples and introduce the expert rule matrix as a priori input into the machine learning model. Based on the sample data, we conduct supervised training, and the model autonomously discovers the hidden deep semantic associations, implicit mapping rules and scenario-based matching features between the requirement fields and the patent fields.

[0037] The model parameters are continuously iterated, the association weights of different field combinations are constantly corrected, the association strength coefficients of each field pairing are adaptively optimized, and after training and parameter convergence are completed, the learning weight matrix that quantifies the data patterns is output.

[0038] By extracting the basic confidence parameters manually set in the expert rule matrix and the dynamic learning weight parameters obtained by model training from the learning weight matrix, the basic confidence and learning weights of the same field pairing position are weighted, calculated, and fused for correction based on the entropy weight method, taking into account both the rigid constraints of manual rules and the flexible adaptation of data learning. The numerical range is standardized and abnormal deviation values ​​are removed to complete the global matrix numerical balance calibration, and finally the judgment mapping matrix is ​​fused to generate the judgment mapping matrix.

[0039] Using a preset judgment mapping matrix, a mapping relationship analysis is performed on the demand-side field set and the patent-side field set to obtain a mapping channel set. This includes: performing field pairing processing on the demand-side field set and the patent-side field set to obtain a candidate field pair set; and using the judgment mapping matrix to sequentially determine the mapping validity and sort the channel priority of the candidate field pair set to obtain a mapping channel set.

[0040] Based on the previously separated fields on the demand side and the patent side, a bidirectional traversal matching process is conducted according to the principle of full coverage of field categories. Each field on the demand side, including the demand intent, target effect, constraints, application scenarios, and technical preferences, is paired with all fields on the patent side, including the technical problem to be solved, technical effect, technical means, claim limiting features, applicable conditions, and technical field. This process generates all cross-combined field pairings, and all combinations are summarized to form a fully covered set of candidate field pairs.

[0041] Next, the validity of the mapping is determined by retrieving the pre-defined judgment mapping matrix that integrates expert rules and machine learning weights. Based on the association confidence and judgment threshold of the corresponding row and column positions in the matrix, each candidate field pair is checked one by one.

[0042] By comparing the pre-defined semantic association rules and quantified scores within the matrix, it is determined whether the current field pair has a reasonable cross-domain mapping relationship. Invalid field pairs with semantic role conflicts, logical mismatches, or no actual relationship are eliminated, and only valid field pairs that conform to technical semantic logic and have mapping value are retained. After selecting valid field pairs, channel priority sorting is performed. The fusion weight values ​​corresponding to each valid field pair in the judgment mapping matrix are read, and sorted according to the degree of association tightness and semantic adaptability scores. All valid field pairs are hierarchically sorted in descending order of association strength to divide the priority level of mapping channels. At the same time, core strong association channels and secondary auxiliary association channels are divided. Finally, the ordered valid field pairs are integrated to form a standardized set of mapping channels.

[0043] The process of parsing and judging the demand-side field set to obtain the demand constraint judgment result includes: identifying and extracting fields from the demand-side field set to obtain an initial constraint set; processing the initial constraint set using natural language processing and semantic parsing techniques to obtain constraint types and logical elements; and performing logical generation and strength assignment processing on the constraint types and logical elements to obtain the demand constraint judgment result.

[0044] Based on each of the candidate patent texts, the set of patent-side fields is selected, and a prototype problem is generated based on at least two source fields obtained.

[0045] Specifically, the source fields are selected from at least two of the technical means field, technical effect field, and claim feature field in the patent-side field set. Using a reverse logic template or sequence generation model, logical reasoning and text reconstruction are performed on the at least two source fields to generate at least one candidate prototype problem. The reverse logic template defines the reasoning path from the technical solution and technical effect to the technical problem.

[0046] Then, the at least one candidate prototype problem is subjected to quality assessment and screening to obtain a final prototype problem. The quality assessment includes at least a syntactic correctness assessment and a logical consistency assessment with the source field.

[0047] The process of verifying the implication relationship between the source field and the prototype question to obtain a predicted bridging object that matches the candidate patent text includes: using a natural language inference model to verify the implication relationship between the source field and the prototype question to obtain an implication relationship score; and performing confidence filtering on the prototype question based on the implication relationship score to obtain a predicted bridging object that matches the candidate patent text.

[0048] Specifically, the patent-side source text on which the prototype question is based is first standardized with the constructed prototype question text to unify the sentence format and semantic expression. Then, the two sets of texts are input into a pre-trained natural language reasoning model, and the model performs a deep comparison and analysis of the semantic logic of the two.

[0049] The model autonomously determines whether the core semantics of the source field can support, deduce, and cover the content of the prototype problem. It identifies the semantic relationship logic between the two one by one, including different relationship types such as semantic inclusion, logical deduction, content overlap, semantic irrelevance, and logical contradiction. Combining multi-dimensional indicators such as semantic overlap, logical deduction rationality, and content matching relevance, it performs quantitative calculations and finally outputs a quantifiable implication relationship score, which intuitively reflects the reliability of the semantic implication between the source field and the prototype problem.

[0050] A reasonable implication score threshold is pre-set as the screening criterion for valid prototype questions. The implication relationship score corresponding to each prototype question is compared with the preset threshold one by one. High-quality prototype questions with scores that reach or exceed the threshold, valid semantic implication logic, and reasonable inference relationships are selected and retained. Invalid prototype questions with low scores, semantic disconnect, logical contradictions, or those that cannot be effectively supported by source fields are eliminated. Erroneous inference results that are semantically distorted or deviate from the actual technical content of the patent are removed. The selected compliant and valid prototype questions are integrated and standardized as an intermediate semantic carrier for converting the patent text from a technical solution description to a problem-type description. Finally, a stable, compliant, and highly compatible predictive bridging object with the candidate patent text is formed.

[0051] In this process, the predictive bridging object is an intermediate semantic carrier specifically used to bridge the semantic differences between the two types of texts, that is, to express the technical solution type of the patent side as a problem type expression with the same logic as the user's technical requirements.

[0052] Preferably, in another embodiment, this embodiment performs independent implication verification for each prototype question. Specifically, the source fields and prototype questions are constructed into natural language inference pairs, and the independent verification model outputs implication, neutral, or contradictory results. A prototype question is accepted as a prediction bridge object only if at least two source fields respectively give an "implication" to the prototype question and there is no "contradiction".

[0053] Preferably, the generation model and the verification model adopt a context isolation strategy, that is, the input of the verification model does not contain the context prompts of the generation model, so as to avoid the verification model making self-proving judgments directly based on the generation basis.

[0054] The prototype questions use natural language that is closer to users' questioning habits, and are used to bridge the question space and the solution space.

[0055] The process of analyzing predicted bridging objects that match the demand-side field set and the candidate patent text using the mapping channel set to obtain a bridging matching factor includes: using the mapping channel set to perform association channel filtering on the demand-side field set and the predicted bridging objects to obtain an association channel set; based on the association channel set, performing semantic similarity calculation on the technical issue field and / or expected effect field in the demand-side field set and the bridging issue in the predicted bridging object to obtain a demand-bridging similarity; weighting the implication relationship score of the predicted bridging object and the demand-bridging similarity to obtain an initial bridging matching value; and normalizing the initial bridging matching value to obtain the bridging matching factor.

[0056] The system retrieves the mapping channel set previously determined through validity assessment and priority ranking. Based on various structured fields on the demand side and combined with the semantic content of the bridging issues contained in the predicted bridging objects, it iterates through all established valid mapping channels. Each mapping channel is compared with the semantic attributes of the corresponding demand field category and the bridging issue. Valid channels that adapt to the semantics of the demand appeal and enable cross-semantic role connection between demand-side fields and patent bridging issues are selected. Invalid channels with semantic type mismatches, mapping logic conflicts, or inability to establish issue dimension associations are eliminated. The selected and retained valid mapping channels are then uniformly summarized and integrated to form a set of associated channels specifically matched to the current patent bridging object.

[0057] Based on the selected set of related channels, the core key fields on the demand side are identified, and the technical problem fields and expected effect fields that represent user demands are selected as the comparison subjects. At the same time, the core text content of the bridging problem inside the predicted bridging object is extracted. Relying on technologies such as semantic vector encoding and contextual semantic understanding, deep semantic analysis and feature extraction are performed on the text at both ends under the semantic mapping rules limited by the related channels.

[0058] A comprehensive comparison of the problem descriptions, desired effects, core concerns, and technical requirements of both parties is conducted. The degree of content overlap, semantic fit, and requirement matching is calculated, and a standardized requirement-bridging similarity is quantitatively output.

[0059] Next, the implicit relation score obtained in the previous implicit relation verification step, which is used to characterize the authenticity and reliability of the bridging problem, is retrieved. At the same time, the demand-bridging similarity calculated in the previous step is obtained. Combining the technical weight importance of the two types of indicators, a linear weighted fusion calculation is performed according to the set weights. Taking into account the generation credibility of the bridging object itself and the semantic matching degree between the bridging object and the user's demand, the quantitative values ​​of the two dimensions are fused and superimposed to generate an initial bridging matching value that has not been standardized and corrected.

[0060] All initial bridging matching values ​​corresponding to all candidate patent texts are statistically analyzed to determine the maximum and minimum values ​​of the numerical range. A standardized normalization algorithm is used to perform range mapping and scaling correction on the initial bridging matching values ​​of individual patents, transforming the discrete and inconsistent original values ​​into a fixed standard numerical range, which is the bridging matching factor.

[0061] The demand-side field set and the patent-side field set are analyzed to obtain the direct matching factor. Specifically, the dimensional alignment and range filtering of the demand-side and patent-side fields are first completed based on the judgment mapping matrix to determine the candidate field pairs that can be directly compared. Then, shallow semantic and entity feature comparisons are carried out from the levels of professional terms, technical features, and functional limitations. The field matching score is calculated for each group, and the individual group scores are corrected by combining the association weight of the mapping matrix to weaken the interference of invalid fields and obtain the overall original matching score. Then, the overall original matching score is normalized to obtain the direct matching factor.

[0062] Based on the demand constraint determination result, the bridging matching factor, and the direct matching factor, the optimal matching result is obtained.

[0063] Specifically, the results of the demand constraint determination are used as a rigid screening prerequisite. Based on the type, logical rules and constraint strength of each constraint, the candidate patents are checked one by one to see if they meet the user's hard requirements such as scenario restrictions, performance indicators and technical boundaries. Patents that do not meet the constraints or do not conform to the actual application limitations are eliminated to complete the preliminary filtering and screening, and retain the set of compliant and valid candidate patents.

[0064] For the candidate patent set, the bridging matching factor and the direct matching factor are fused and weighted according to the preset weight ratio. The comprehensive matching score of each candidate patent is obtained by comprehensively considering the bridging adaptation degree of the patent after semantic transformation and the direct feature matching degree of the original field. The comprehensive matching scores are sorted from high to low, and the final ranking result is output. This determines the optimal patent matching result that best fits the semantic requirements of the technical needs, meets the constraints, and has the highest technical solution adaptability.

[0065] In a preferred embodiment, the alignment confidence is further calculated based on the term overlap rate, the hit rate of the demand field terms, and the difference between the bridging score and the direct score, and the alignment confidence is used to adaptively adjust the fusion weight of direct matching and bridging matching.

[0066] In another preferred embodiment, a heterogeneous knowledge graph is constructed based on the patent-side structured results. The heterogeneous knowledge graph includes at least patent nodes, technical means nodes, technical effect nodes, technical field nodes, applicable condition nodes, claim-limiting feature nodes, technical object nodes, and prototype problem nodes, and records the semantic relationships between nodes through directed edges.

[0067] Both path template recall and vector nearest neighbor recall can be used simultaneously. Path template recall is used to retrieve candidate patents along defined semantic paths, while vector nearest neighbor recall provides a fallback for non-standard expressions. After merging the two candidate sets, a multi-dimensional ranking is performed based on factors such as field alignment score, bridging score, constraint satisfaction, domain relevance, and graph path strength.

[0068] For those skilled in the art, any equivalent substitutions or modifications made to field categories, bridge object storage formats, recall index types, scoring functions, threshold parameters, graph structures, or model interfaces without departing from the spirit and essence of this invention should be included within the scope of protection of this invention.

[0069] Another embodiment of the present invention provides a matching system for technical requirements and patent solutions. For details, please refer to [link / reference needed]. Figure 2 , Figure 2 The diagram shown illustrates the structure of a system for matching technical requirements with patented solutions in one embodiment of the present invention. The system includes: Module 11 is used to acquire the user's technical requirement text and candidate patent texts from the literature database; Extraction module 12 is used to perform semantic extraction processing on the technical requirement text and the candidate patent text to obtain a set of requirement-side fields and a set of patent-side fields. The determination module 13 is used to parse and determine the set of demand-side fields to obtain the demand constraint determination result; The verification module 14 is used to select the set of patent-side fields based on each candidate patent text, generate a prototype question based on at least two source fields, and verify the implication relationship between the source fields and the prototype question to obtain a prediction bridging object that matches the candidate patent text. Matching module 15 is used to analyze the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain bridging matching factors; and to analyze the demand-side field set and the patent-side field set to obtain direct matching factors. The generation module 16 is used to obtain the optimal matching result based on the requirement constraint determination result, the bridging matching factor and the direct matching factor.

[0070] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for matching technical requirements with patent solutions, characterized in that, include: Obtain the user's technical requirements text and candidate patent texts from the literature database; Semantic extraction processing is performed on the technical requirement text and the candidate patent text to obtain a set of requirement-side fields and a set of patent-side fields. The set of demand-side fields is parsed and judged to obtain the demand constraint judgment result; Based on each candidate patent text, the set of patent-side fields is selected, and a prototype question is generated based on at least two source fields. The implication relationship between the source fields and the prototype question is verified to obtain a prediction bridging object that matches the candidate patent text. Analyze the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain the bridging matching factor; analyze the demand-side field set and the patent-side field set to obtain the direct matching factor; Based on the demand constraint determination result, the bridging matching factor, and the direct matching factor, the optimal matching result is obtained.

2. The method for matching technical requirements with patent solutions as described in claim 1, characterized in that, The semantic extraction process of the technical requirement text and the candidate patent text yields a set of requirement-side fields and a set of patent-side fields, including: Text preprocessing and domain concept recognition processing are performed on the technical requirement text and the candidate patent text to obtain the requirement expression paradigm of the technical requirement text and the patent expression paradigm of the candidate patent text. Semantic role labeling and field recognition are performed on the aforementioned demand expression paradigm to obtain a demand-side field set, which includes at least a demand intent field, a target effect field, a constraint condition field, an application scenario field, and a technology preference field. Semantic role labeling and field identification are performed on the patent expression paradigm to obtain a set of patent-side fields. The set of patent-side fields includes at least the fields of technical problem to be solved, technical effect, technical means, claim limiting features, applicable conditions, and technical field.

3. The method for matching technical requirements with patent solutions as described in claim 1, characterized in that, The process of parsing and judging the set of demand-side fields to obtain the demand constraint judgment result includes: The fields in the demand-side field set are identified and extracted to obtain an initial constraint set; The initial constraint set is processed using natural language processing and semantic parsing techniques to obtain constraint types and logical elements; The constraint type and the logical element are logically generated and strength assigned to obtain the requirement constraint determination result.

4. The method for matching technical requirements with patent solutions as described in claim 1, characterized in that, The step of verifying the implication relationship between the source field and the prototype question to obtain a prediction bridging object that matches the candidate patent text includes: Using a natural language reasoning model, the implication relationship between the source field and the prototype question is validated to obtain an implication relationship score. Based on the implication relation score, the prototype problem is subjected to confidence filtering to obtain a predicted bridging object that matches the candidate patent text.

5. The method for matching technical requirements with patent solutions as described in claim 4, characterized in that, The analysis of the predicted bridging objects that match the demand-side field set and the candidate patent texts to obtain bridging matching factors includes: The demand-side field set and the prediction bridging object are subjected to association channel filtering to obtain an association channel set; Based on the set of associated channels, semantic similarity calculation is performed on the technical issue field and / or expected effect field in the demand-side field set and the bridging issue in the predicted bridging object to obtain the demand-bridging similarity. The implication relationship score of the predicted bridging object and the demand-bridging similarity are weighted to obtain the initial bridging matching value; The initial bridging matching value is normalized to obtain the bridging matching factor.

6. A system for matching technical requirements with patented solutions, characterized in that, include: The acquisition module is used to acquire the user's technical requirement text and candidate patent texts from the literature database; The extraction module is used to perform semantic extraction processing on the technical requirement text and the candidate patent text to obtain a set of requirement-side fields and a set of patent-side fields. The determination module is used to parse and determine the set of demand-side fields to obtain the demand constraint determination result. The verification module is used to select and process the set of patent-side fields based on each candidate patent text, generate a prototype question based on at least two source fields, and verify the implication relationship between the source fields and the prototype question to obtain a prediction bridging object that matches the candidate patent text. The matching module is used to analyze the predicted bridging objects that match the demand-side field set and the candidate patent text to obtain the bridging matching factor; and to analyze the demand-side field set and the patent-side field set to obtain the direct matching factor. The generation module is used to obtain the optimal matching result based on the requirement constraint determination result, the bridging matching factor, and the direct matching factor.

7. The matching system for technical requirements and patent solutions as described in claim 6, characterized in that, The semantic extraction process of the technical requirement text and the candidate patent text yields a set of requirement-side fields and a set of patent-side fields, including: Text preprocessing and domain concept recognition processing are performed on the technical requirement text and the candidate patent text to obtain the requirement expression paradigm of the technical requirement text and the patent expression paradigm of the candidate patent text. Semantic role labeling and field recognition are performed on the aforementioned demand expression paradigm to obtain a demand-side field set, which includes at least a demand intent field, a target effect field, a constraint condition field, an application scenario field, and a technology preference field. Semantic role labeling and field identification are performed on the patent expression paradigm to obtain a set of patent-side fields. The set of patent-side fields includes at least the fields of technical problem to be solved, technical effect, technical means, claim limiting features, applicable conditions, and technical field.

8. The matching system for technical requirements and patent solutions as described in claim 6, characterized in that, The process of parsing and judging the set of demand-side fields to obtain the demand constraint judgment result includes: The fields in the demand-side field set are identified and extracted to obtain an initial constraint set; The initial constraint set is processed using natural language processing and semantic parsing techniques to obtain constraint types and logical elements; The constraint type and the logical element are logically generated and strength assigned to obtain the requirement constraint determination result.

9. The matching system for technical requirements and patent solutions as described in claim 6, characterized in that, The step of verifying the implication relationship between the source field and the prototype question to obtain a prediction bridging object that matches the candidate patent text includes: Using a natural language reasoning model, the implication relationship between the source field and the prototype question is validated to obtain an implication relationship score. Based on the implication relation score, the prototype problem is subjected to confidence filtering to obtain a predicted bridging object that matches the candidate patent text.

10. The matching system for technical requirements and patent solutions as described in claim 9, characterized in that, The analysis of the predicted bridging objects that match the demand-side field set and the candidate patent texts to obtain bridging matching factors includes: The demand-side field set and the prediction bridging object are subjected to association channel filtering to obtain an association channel set; Based on the set of associated channels, semantic similarity calculation is performed on the technical issue field and / or expected effect field in the demand-side field set and the bridging issue in the predicted bridging object to obtain the demand-bridging similarity. The implication relationship score of the predicted bridging object and the demand-bridging similarity are weighted to obtain the initial bridging matching value; The initial bridging matching value is normalized to obtain the bridging matching factor.

Images