A large model-based scientific and technological achievement supply and demand matching method and system

Abstract

The application discloses a kind of based on big model's science and technology achievement supply and demand matching method and system, belong to artificial intelligence and science and technology achievement transformation technical field.The method includes: obtaining science and technology achievement structured information and enterprise technology demand structured information, and respectively generate corresponding initial semantic representation data vector;Big language model is used to respectively carry out semantic extension to science and technology achievement structured information and enterprise technology demand structured information, and fusion generates science and technology achievement fusion representation vector and enterprise technology demand fusion representation vector;Through big language model, key information extraction is carried out to user matching request, and combined with historical enterprise technology demand fusion representation vector, semantic enhancement and demand reconstruction are carried out, and generate standardization search request;Based on standardization search request, vector semantic recall and full-text search recall are carried out, and candidate matching result set is obtained;Text semantic matching degree and label semantic matching degree between candidate matching result set and standardization search request are calculated, and sorting and screening are carried out based on comprehensive matching score, and final matching result is obtained.The system is used to realize the above-mentioned method process.The application can improve the accuracy and matching efficiency of science and technology achievement supply and demand matching.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and data processing technology, specifically a method and system for matching the supply and demand of scientific and technological achievements based on a large model. Background Technology

[0002] With the continuous development of the technology transfer system, the efficient matching of scientific and technological achievements with the technological needs of enterprises has become a crucial link in promoting industrial innovation. Currently, enterprises often need to quickly acquire scientific and technological achievements that match their needs during the process of technological upgrading and product development, while a large number of scientific and technological achievements generated by research institutions, universities, and other units also urgently need to be industrialized. Therefore, building an efficient and accurate supply and demand matching mechanism is of great significance.

[0003] Traditional methods for matching supply and demand for scientific and technological achievements typically rely on keyword retrieval, tag matching, or vector similarity calculation. Existing solutions structure scientific and technological achievement data and enterprise demand data, constructing vector representations for each. Upon receiving a user query, they utilize a combination of vector retrieval and full-text retrieval to recall candidate results, and then reorder them using a ranking model to obtain the final matching result. Furthermore, some technologies employ large language models to identify user intent and reconstruct questions from the input, generating solutions and combining them with tag information for multi-path recall and reordering, thereby improving the accuracy of supply and demand matching.

[0004] However, existing technologies still have the following shortcomings: Insufficient semantic understanding: Existing methods mostly rely on keyword matching or shallow vector similarity calculation, which makes it difficult to accurately understand the complex semantic relationships implied in the technical needs of enterprises. Especially when facing cross-domain and multi-technical requirements, matching deviations are prone to occur.

[0005] There are differences in semantic expression across scenarios: Enterprise technical needs are usually described with application scenarios and engineering indicators as the guide, while scientific and technological achievements focus on expressing technical principles and innovations. There are significant differences between the two at the semantic level, which can easily lead to semantic misalignment during the matching process and affect the accuracy of the matching results.

[0006] Lack of knowledge enhancement and expansion capabilities: Most existing solutions are based directly on matching raw text or simple structured information, lacking in-depth mining of technical knowledge, industry chain relationships and technical correlation information, making it difficult to achieve higher-level semantic association and reasoning.

[0007] Therefore, there is an urgent need to provide a method and system for matching the supply and demand of scientific and technological achievements based on a large model, so as to improve the semantic understanding of enterprises' technological needs, achieve deep semantic alignment between scientific and technological achievements and enterprise needs, and improve the accuracy and reliability of matching. Summary of the Invention

[0008] This invention provides a method and system for matching the supply and demand of scientific and technological achievements based on a large model, in order to solve the problem of poor accuracy in the matching results of supply and demand of scientific and technological achievements in the existing technology.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for matching the supply and demand of scientific and technological achievements based on a large model, the process of which is as follows: Obtain structured information on scientific and technological achievements and structured information on enterprise technology needs; encode the structured information on scientific and technological achievements into initial semantic representation data vectors of scientific and technological achievements; and encode the structured information on enterprise technology needs into initial semantic representation data vectors of enterprise technology needs. The structured information of scientific and technological achievements and the structured information of enterprise technology needs are semantically extended using a large language model, resulting in extended semantic information of scientific and technological achievements and extended semantic information of enterprise technology needs. The extended semantic information of scientific and technological achievements and extended semantic information of enterprise technology needs are encoded into extended semantic vectors, and then fused with the corresponding initial semantic representation data vectors to obtain fused representation vectors of scientific and technological achievements and fused representation vectors of enterprise technology needs. By extracting key information from the user's scientific and technological achievement matching request using a large language model, and combining the historical enterprise technology demand fusion representation vector in the enterprise demand database, the user matching request is semantically enhanced and the demand is reconstructed to generate a standardized retrieval request. Based on standardized retrieval requests, multi-path recall is performed on the fusion representation vectors of scientific and technological achievements in the scientific and technological achievements database, and the recall results are merged and deduplicated to obtain a set of candidate fusion representation vectors of scientific and technological achievements. Calculate the comprehensive matching score between each candidate scientific and technological achievement fusion representation vector and the standardized retrieval request, and sort the candidate scientific and technological achievement fusion representation vectors in descending order according to the comprehensive matching score; The candidate scientific and technological achievement fusion representation vectors, sorted in descending order, are filtered to obtain a number of candidate scientific and technological achievement fusion representation vectors as the final matching results.

[0010] Furthermore, the original text data of scientific and technological achievements is collected, and information such as title, abstract, technical field, key technology, application direction, source, production time, and maturity is extracted from the original text data to obtain structured information of scientific and technological achievements. Collect raw text data of enterprise technical requirements, and extract enterprise name information, requirement title information, requirement description information, application scenario information, technical target information, performance index information, constraint information, and requirement submission time information from the raw text data of enterprise technical requirements, thereby obtaining structured information of enterprise requirements.

[0011] Furthermore, the structured information of scientific and technological achievements is input into the large language model, and corresponding extended semantic information of scientific and technological achievements is generated based on the prompt templates corresponding to the scientific and technological achievements. The extended semantic information of scientific and technological achievements includes: technical implementation path information, applicable industry information, potential application scenario information, key technical difficulties information, and upstream and downstream related technical information. The structured information of enterprise technology requirements is input into the large language model, and corresponding extended semantic information of enterprise technology requirements is generated based on the prompt template corresponding to the enterprise technology requirements. The extended semantic information of enterprise technology requirements includes: technical problem breakdown information, feasible technical route information, potential alternative technical solution information, related application scenario information, and constraint analysis information.

[0012] Furthermore, the initial semantic representation data vectors of scientific and technological achievements and the initial semantic representation data vectors of enterprise technology needs are fused with their respective extended semantic vectors according to their respective preset weights to obtain the fused representation vectors of scientific and technological achievements and the fused representation vectors of enterprise technology needs.

[0013] Furthermore, key information is extracted from the user's input request for matching scientific and technological achievements through a large language model. The key information includes the technical topic, application scenario, core objective, and constraints.

[0014] Furthermore, the multi-path recall includes vector semantic recall and full-text retrieval recall.

[0015] Furthermore, the comprehensive matching score includes text semantic matching degree and tag semantic matching degree.

[0016] Furthermore, a score threshold is set to filter the candidate scientific and technological achievement fusion representation vectors sorted in descending order, and the final matching result is obtained by selecting multiple candidate scientific and technological achievement fusion representation vectors whose final scores are greater than or equal to the score threshold.

[0017] A system for implementing the above-mentioned large-scale model-based method for matching the supply and demand of scientific and technological achievements includes: The data processing module is used to collect, clean, parse, and structure the raw data of scientific and technological achievements and the raw data of enterprise technology needs, and generate corresponding structured information and initial semantic representation data. The semantic extension module is used to call the large language model to semantically extend the structured information of scientific and technological achievements and the structured information of enterprise needs, and form a fusion representation of scientific and technological achievements and a fusion representation of enterprise technology needs. The request parsing module is used to extract key information from the user's input technology achievement matching request, and combine it with the historical enterprise technology demand fusion representation vector in the enterprise demand database to perform semantic enhancement and demand reconstruction, and generate standardized search requests. The multi-path recall module is used to perform multi-path recall on the fusion representation vectors of scientific and technological achievements in the scientific and technological achievements database based on standardized retrieval requests, and generate a set of candidate fusion representation vectors of scientific and technological achievements. The matching calculation module is used to calculate the comprehensive matching score between the candidate scientific and technological achievement fusion representation vector and the standardized retrieval request, and to rank the candidate scientific and technological achievement fusion representation vector according to the comprehensive matching score. The results output module is used to perform threshold filtering on the ranked candidate scientific and technological achievement fusion representation vectors and output the final matching results to the user terminal.

[0018] Compared with the prior art, the advantages of the present invention are: This invention presents a method and system for matching the supply and demand of scientific and technological achievements based on a large-scale model. By semantically expanding and fusing data on scientific and technological achievements and enterprise technology demand data, and combining historical enterprise demand enhancement, multi-path recall, and a comprehensive matching scoring mechanism, it achieves a deep semantic understanding of enterprise technology demands and accurate matching of scientific and technological achievements. Compared to traditional methods, this invention supports demand expression in natural language, improving the accuracy and efficiency of matching the supply and demand of scientific and technological achievements, thereby enhancing the level of intelligence in the process of technology transfer. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method according to an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the structured processing and fusion representation construction process of scientific and technological achievements and enterprise technology demand data in the method of the present invention.

[0021] Figure 3 This is a schematic diagram of the user matching request parsing and standardized retrieval request generation process in the method of this embodiment of the invention.

[0022] Figure 4 This is a schematic diagram of the multi-channel recall and comprehensive matching scoring process in the method of this invention embodiment.

[0023] Figure 5 This is a structural block diagram of the system according to an embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] like Figure 1 As shown in the figure, this embodiment discloses a method for matching the supply and demand of scientific and technological achievements based on a large model. The process is as follows: Step S1, as follows Figure 2As shown, raw text data of scientific and technological achievements and raw text data of enterprise technology requirements are collected. The collected raw text data are preprocessed before information extraction, resulting in structured information of scientific and technological achievements and structured information of enterprise technology requirements. Furthermore, the structured information of scientific and technological achievements is encoded into an initial semantic representation data vector of scientific and technological achievements, and the structured information of enterprise technology requirements is encoded into an initial semantic representation data vector of enterprise technology requirements.

[0026] In this embodiment, the original text data of the collected scientific and technological achievements comes from at least one of the following data sources: (1) Patent specification, patent abstract and claims text; (2) Title, abstract, and keywords of the paper; (3) Registration information of scientific and technological achievements of universities, research institutes, key laboratories or industrial technology research institutes; (4) Semi-structured documents such as technology contracts, project acceptance and completion results, and expert recommendation results.

[0027] Because the original text data of scientific and technological achievements often contains problems such as missing fields, redundant descriptions, inconsistent formats, and non-standard terminology, this embodiment preprocesses the collected original text data of scientific and technological achievements. The preprocessing includes deduplication, noise removal, field normalization, and missing field completion.

[0028] Then, the title information, abstract information, technical field information, key technology information, application direction information, source information, production time information, and maturity information of the scientific and technological achievements are extracted from the preprocessed original text data, thereby obtaining the structured information of the scientific and technological achievements. This structured information of the scientific and technological achievements includes the title information, abstract information, technical field information, key technology information, application direction information, source information, production time information, and maturity information from the original text data of the scientific and technological achievements.

[0029] In this embodiment, the raw text data of enterprise technical requirements is collected from online enterprise submissions, technical manager input, meeting minutes, survey records, or document uploads. Raw text data of enterprise technical requirements typically features extensive natural language descriptions, inconsistent formats, and numerous implicit constraints. Therefore, this embodiment performs the same preprocessing on the collected raw text data of enterprise technical requirements, including deduplication, noise removal, field normalization, and missing field completion.

[0030] Then, from the preprocessed original text data of enterprise technical requirements, information such as enterprise name, requirement title, requirement description, application scenario, technical objective, constraint, and requirement submission time are extracted to obtain structured information of the enterprise technical requirements. This structured information of enterprise technical requirements includes the enterprise name, requirement title, requirement description, application scenario, technical objective, constraint, and requirement submission time information from the original text data of the enterprise technical requirements.

[0031] In this embodiment, a vector coding model is used to encode the structured information of scientific and technological achievements and the structured information of enterprise technology requirements, respectively, to obtain the initial semantic representation data vector of scientific and technological achievements and the initial semantic representation data vector of enterprise technology requirements.

[0032] The vector encoding model used in this embodiment is a general text embedding model or an industry-specific fine-tuned embedding model. The Qwen3-Embedding-4B general embedding model is preferred, and the output is set to a 1024-dimensional vector.

[0033] The structured information of scientific and technological achievements, including title information, abstract information, technical field information, key technology information, application direction information, source information, output time information, and maturity information, is sequentially concatenated to form a concatenated text. The concatenated text is then input into a vector encoding model, which generates an initial semantic representation data vector of the scientific and technological achievements.

[0034] The enterprise name, requirement title, requirement description, application scenario, technical objective, constraint, and requirement submission time information in the structured information of enterprise technical requirements are sequentially concatenated to form a concatenated text. The concatenated text is then input into a vector encoding model, which generates an initial semantic representation data vector of the enterprise technical requirements.

[0035] Step S2, as follows Figure 2 As shown, semantic expansion is performed on the structured information of scientific and technological achievements and the structured information of enterprise needs, respectively, resulting in expanded semantic information of scientific and technological achievements and expanded semantic information of enterprise technical needs. Next, the expanded semantic information of scientific and technological achievements and the expanded semantic information of enterprise technical needs are encoded into vectors. Then, the initial semantic representation data vector of scientific and technological achievements and the expanded semantic information vector of scientific and technological achievements are fused to obtain a fused representation vector of scientific and technological achievements, and the initial semantic representation data vector of enterprise technical needs and the expanded semantic information vector of enterprise technical needs are fused to obtain a fused representation vector of enterprise technical needs. Finally, the fused representation vector of scientific and technological achievements is written into the scientific and technological achievements database, and the fused representation vector of enterprise technical needs is written into the enterprise needs database, for use in subsequent user requests for matching scientific and technological achievements with historical enterprise needs semantic enhancement.

[0036] Specifically, in this embodiment, a large language model is used to semantically expand the structured information of scientific and technological achievements and the structured information of enterprise technical requirements according to the corresponding prompt templates. The large language model uses Qwen3.5-27B, or DeepSeek-V3.2, or Llama4 Scout Chinese Enhanced Version. Preferably, Qwen3.5-27B is used to complete the semantic expansion and requirement decomposition.

[0037] The structured information of scientific and technological achievements is input into a large language model, which then generates corresponding extended semantic information based on the given prompt templates. These prompt templates include items related to technical implementation paths, applicable industries, potential application scenarios, key technical challenges, and upstream and downstream related technologies. Therefore, the extended semantic information generated by the large language model based on these prompt templates includes: technical implementation path information, applicable industry information, potential application scenario information, key technical challenges information, and upstream and downstream related technology information.

[0038] The structured information of enterprise technology requirements is input into a large language model. The large language model then generates corresponding extended semantic information for the enterprise technology requirements based on the corresponding prompt template. The prompt template includes items such as technical problem breakdown, feasible technical routes, potential alternative technical solutions, related application scenarios, and constraint analysis. Therefore, the extended semantic information for enterprise technology requirements generated by the large language model based on the corresponding prompt template includes: technical problem breakdown information, feasible technical route information, potential alternative technical solution information, related application scenario information, and constraint analysis information.

[0039] Next, in this embodiment, the technical implementation path information, applicable industry information, potential application scenario information, key technical difficulties information, and upstream and downstream related technical information in the extended semantic information of scientific and technological achievements are sequentially concatenated into concatenated information and then input into the vector encoding model, which generates the extended semantic information vector of scientific and technological achievements.

[0040] Furthermore, the technical problem breakdown information, feasible technical route information, potential alternative technical solution information, related application scenario information, and constraint analysis information in the extended semantic information of enterprise technical requirements are sequentially concatenated to form a concatenated text, which is then input into the vector encoding model, and the vector encoding model generates the extended semantic vector of enterprise technical requirements.

[0041] In this embodiment, the initial semantic representation data vector of the scientific and technological achievement and the extended semantic vector of the scientific and technological achievement are fused according to a preset weight to obtain the fused representation vector Fr of the scientific and technological achievement, as shown in the following formula: Fr = 0.65 × Emb (structured information of scientific and technological achievements) + 0.35 × Emb (extended semantic information of scientific and technological achievements) The initial semantic representation data vector of enterprise technology requirements and the extended semantic vector of enterprise technology requirements are fused according to preset weights to obtain the fused representation vector Fd of enterprise technology requirements, as shown in the following formula: Fd = 0.60 × Emb (structured information of enterprise technology requirements) + 0.40 × Emb (extended semantic information of enterprise technology requirements) Where Emb(·) represents the encoding function of the vector encoding model; Fr represents the fusion representation vector of scientific and technological achievements; and Fd represents the fusion representation vector of enterprise technology needs.

[0042] Finally, in this embodiment, the fusion representation vector of scientific and technological achievements is written into the scientific and technological achievements database, and the fusion representation vector of enterprise technology needs is written into the enterprise needs database.

[0043] Step S3, as follows Figure 3 As shown, the system extracts key information from the user's input of the scientific and technological achievement matching request through a large language model, and combines the historical enterprise technology demand fusion representation vector in the enterprise demand database to perform semantic enhancement and demand reconstruction on the user's matching request, thereby generating a standardized retrieval request.

[0044] In this embodiment, the user's text requesting a matching of scientific and technological achievements is input into a large language model. The large language model understands and analyzes the semantics of the technical requirements in the user's matching request and automatically identifies the user's current intention to match scientific and technological achievements.

[0045] When the user's intent is identified as a demand for matching scientific and technological achievements, the large language model extracts key information structured fields from the user's matching request text. These key information structured fields include the technical topic, application scenario, core objective, and constraints, thereby obtaining the corresponding key information from the user's matching request text.

[0046] Next, in this embodiment, the extracted key information is encoded into a request semantic vector, and the request semantic vector is used to perform similarity matching with the historical enterprise technology demand fusion representation vector in the enterprise demand database. The historical enterprise technology demand fusion representation vector that is similar to the current user's request semantics is then recalled from the enterprise demand database.

[0047] Furthermore, the large language model combines the historical enterprise technology demand fusion representation vector obtained from the recall to perform semantic enhancement and demand reconstruction on the technology themes, application scenarios, core objectives, and constraints in the user matching request.

[0048] Specifically, in this embodiment, the large language model combines the technical problem breakdown information, feasible technical route information, potential alternative technical solution information, related application scenario information, and constraint analysis information in the historical enterprise technical needs fusion representation vector to complete and standardize the unstructured, colloquial, or vaguely expressed text content in the user matching request.

[0049] Finally, based on the user matching requests after semantic enhancement and demand reconstruction, the large language model generates standardized retrieval requests suitable for subsequent recall and matching calculations.

[0050] Step S4, as follows Figure 4 As shown, based on the standardized retrieval request obtained in step S3, vector semantic recall and full-text retrieval are performed on the fusion representation vectors of scientific and technological achievements in the scientific and technological achievements database. The recall results are then merged and deduplicated to obtain a set of candidate matching results composed of multiple candidate fusion representation vectors of scientific and technological achievements.

[0051] In this embodiment, the vector semantic recall process is as follows: The standardized retrieval request is input into the vector encoding model, which generates a query vector Q. Then, the cosine similarity score between the query vector Q and the fusion representation vectors Fr of each scientific and technological achievement in the scientific and technological achievement database is calculated. The cosine similarity score Sim_vec(Q,Fr) is calculated as follows: Sim_vec(Q,Fr) = (Q·Fr) / (||Q||×||Fr||) Where Sim_vec(Q,Fr) represents the cosine similarity score between the query vector Q and the fusion representation vector of scientific and technological achievements Fr; Q represents the request semantic vector corresponding to the standardized retrieval request; Fr represents the fusion representation vector of scientific and technological achievements; Q·Fr represents the dot product of the request semantic vector Q and the fusion representation vector of scientific and technological achievements Fr; ||Q|| represents the vector norm of the request semantic vector Q; and ||Fr|| represents the vector norm of the fusion representation vector of scientific and technological achievements Fr.

[0052] Based on the descending order of cosine similarity scores, a predetermined number of scientific and technological achievement fusion representation vectors with the highest ranking in the descending order are selected as candidate scientific and technological achievement fusion representation vectors, forming the first recall result set R1 corresponding to vector semantic recall.

[0053] The full-text search recall process in this embodiment is as follows: Based on the technical topics, application scenarios, core objectives, and constraints in the standardized search request, a full-text search is performed on the title information, abstract information, key technology information, and application direction information of the scientific and technological achievements database. The fusion representation vector of scientific and technological achievements containing at least two key pieces of information is selected as the candidate fusion representation vector of scientific and technological achievements, which constitutes the second recall result set R2 corresponding to the full-text search recall.

[0054] Finally, the candidate scientific and technological achievement fusion representation vectors in the first recall result set R1 and the second recall result set R2 are merged and deduplicated to obtain the candidate matching result set.

[0055] Step S5, as follows Figure 4 As shown, the text semantic matching degree and tag semantic matching degree between each candidate scientific and technological achievement fusion representation vector in the candidate matching result set and the standardized retrieval request are calculated. Based on the text semantic matching degree and tag semantic matching degree, the comprehensive matching score of the corresponding candidate scientific and technological achievement fusion representation vector is calculated. The candidate scientific and technological achievement fusion representation vectors in the candidate matching result set are sorted in descending order according to the comprehensive matching score.

[0056] In this embodiment, the textual semantic similarity between the standardized retrieval request and the fusion representation vector of each candidate scientific and technological achievement is calculated as the textual semantic matching degree S_sem, which represents the semantic closeness between the standardized retrieval request and the fusion representation vector of the candidate scientific and technological achievement at the technical description level.

[0057] Specifically, the standardized retrieval request is input into the vector encoding model to obtain the request semantic vector Q; the candidate scientific and technological achievement fusion representation vector is denoted as Fr_i, where i represents the i-th candidate scientific and technological achievement in the candidate matching result set. Then, the cosine similarity between the request semantic vector Q and the candidate scientific and technological achievement fusion representation vector Fr_i is calculated, and this cosine similarity is used as the text semantic matching degree S_sem_i, calculated as follows: S_sem_i=(Q·Fr_i) / (||Q||×||Fr_i||) Where S_sem_i represents the text semantic matching degree corresponding to the ith candidate scientific and technological achievement fusion representation vector; Q represents the request semantic vector corresponding to the standardized retrieval request; Fr_i represents the ith candidate scientific and technological achievement fusion representation vector; Q·Fr_i represents the vector inner product of the request semantic vector Q and the candidate scientific and technological achievement fusion representation vector Fr_i; ||Q|| represents the vector norm of the request semantic vector Q; and ||Fr_i|| represents the vector norm of the candidate scientific and technological achievement fusion representation vector Fr_i.

[0058] In this embodiment, the degree of tag overlap between the key information in the standardized retrieval request and the structured information of the scientific and technological achievement corresponding to the fusion representation vector of the candidate scientific and technological achievement is compared and used as the tag semantic matching degree S_tag to represent the degree of consistency between the two parties in terms of technical theme, application scenario and core objective dimension.

[0059] Specifically, a request tag set T_q is formed based on the technical topic, application scenario, and core objective in the standardized search request; an achievement tag set T_r_i is formed based on the technical field information, key technology information, and application direction information in the structured information of candidate scientific and technological achievements, where i represents the i-th candidate scientific and technological achievement in the candidate matching result set. Then, the Jaccard coefficient is used to calculate the degree of overlap between the request tag set T_q and the achievement tag set T_r_i, and this degree of overlap is used as the tag semantic matching degree S_tag_i, calculated using the following formula: S_tag_i=|T_q∩T_r_i| / |T_q∪T_r_i| Where S_tag_i represents the semantic matching degree of the tag corresponding to the i-th candidate scientific and technological achievement; T_q represents the set of request tags corresponding to the standardized retrieval request; T_r_i represents the set of achievement tags corresponding to the i-th candidate scientific and technological achievement; T_q∩T_r_i represents the intersection of the request tag set and the achievement tag set; T_q∪T_r_i represents the union of the request tag set and the achievement tag set; |T_q∩T_r_i| represents the number of tags in the intersection; and |T_q ∪ T_r_i| represents the number of tags in the union.

[0060] Finally, in this embodiment, the text semantic matching degree S_sem_i and the tag semantic matching degree S_tag_i are weighted and fused according to preset weights to generate the comprehensive matching score Score_i of the fusion representation vector of the i-th candidate scientific and technological achievement. The calculation formula is as follows: Score_i=0.70×S_sem_i+0.30×S_tag_i Wherein, Score_i represents the comprehensive matching score of the fusion representation vector of the i-th candidate scientific and technological achievement; S_sem_i represents the text semantic matching degree corresponding to the fusion representation vector of the i-th candidate scientific and technological achievement; and S_tag_i represents the tag semantic matching degree corresponding to the i-th candidate scientific and technological achievement.

[0061] Step S6, as follows Figure 4 As shown, the fusion representation vectors of each candidate scientific and technological achievement in the candidate matching result set, which have been sorted in descending order, are filtered to obtain the final matching result.

[0062] In this embodiment, a comprehensive matching score threshold is set, and the comprehensive matching scores corresponding to the fusion representation vectors of each candidate scientific and technological achievement in the candidate matching result set are filtered to obtain multiple candidate scientific and technological achievement fusion representation vectors with comprehensive matching scores greater than or equal to the comprehensive matching score threshold.

[0063] Then, from the multiple candidate scientific and technological achievement fusion representation vectors obtained through screening, a predetermined number of candidate scientific and technological achievement fusion representation vectors with the highest ranking are selected as the final matching results in descending order of comprehensive matching scores.

[0064] In this embodiment, the comprehensive matching score threshold is set according to actual business needs. Preferably, the comprehensive matching score threshold is set between 0.60 and 0.75.

[0065] Finally, the final matching results are sent to the user's terminal for display.

[0066] like Figure 5 As shown, this embodiment also discloses a technology supply and demand matching system based on a large model, including: The data processing module is used to collect, clean, parse, and structure the raw data of scientific and technological achievements and the raw data of enterprise technology needs, and generate corresponding structured information and initial semantic representation data. The semantic extension module is used to call the large language model to semantically extend the structured information of scientific and technological achievements and the structured information of enterprise needs, and form a fusion representation of scientific and technological achievements and a fusion representation of enterprise technology needs. The request parsing module is used to extract key information from the user's input technology achievement matching request, and combine it with the historical enterprise technology demand fusion representation vector in the enterprise demand database to perform semantic enhancement and demand reconstruction, and generate standardized search requests. The multi-path recall module is used to perform multi-path recall on the fusion representation vectors of scientific and technological achievements in the scientific and technological achievements database based on standardized retrieval requests, and generate a set of candidate fusion representation vectors of scientific and technological achievements. The matching calculation module is used to calculate the comprehensive matching score between the candidate scientific and technological achievement fusion representation vector and the standardized retrieval request, and to rank the candidate scientific and technological achievement fusion representation vector according to the comprehensive matching score. The results output module is used to perform threshold filtering on the ranked candidate scientific and technological achievement fusion representation vectors and output the final matching results to the user terminal.

[0067] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. These embodiments are merely descriptions of preferred embodiments and are not intended to limit the scope or concept of the invention. The specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. Such combinations, as long as they do not violate the spirit of the present invention, should also be considered as part of this disclosure. To avoid unnecessary repetition, the present invention will not further describe the various possible combinations.

[0068] This invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this invention and without departing from the design idea of ​​this invention, all modifications and improvements made by those skilled in the art to the technical solutions of this invention should fall within the protection scope of this invention. The technical content for which protection is sought in this invention has been fully described in the claims.

Claims

1. A method for matching the supply and demand of scientific and technological achievements based on a large model, characterized in that, The process is as follows: Obtain structured information on scientific and technological achievements and structured information on enterprise technology needs; encode the structured information on scientific and technological achievements into initial semantic representation data vectors of scientific and technological achievements; and encode the structured information on enterprise technology needs into initial semantic representation data vectors of enterprise technology needs. The structured information of scientific and technological achievements and the structured information of enterprise technology needs are semantically extended using a large language model, resulting in extended semantic information of scientific and technological achievements and extended semantic information of enterprise technology needs. The extended semantic information of scientific and technological achievements and extended semantic information of enterprise technology needs are encoded into extended semantic vectors, and then fused with the corresponding initial semantic representation data vectors to obtain fused representation vectors of scientific and technological achievements and fused representation vectors of enterprise technology needs. By extracting key information from the user's scientific and technological achievement matching request using a large language model, and combining the historical enterprise technology demand fusion representation vector in the enterprise demand database, the user matching request is semantically enhanced and the demand is reconstructed to generate a standardized retrieval request. Based on standardized retrieval requests, multi-path recall is performed on the fusion representation vectors of scientific and technological achievements in the scientific and technological achievements database, and the recall results are merged and deduplicated to obtain a set of candidate fusion representation vectors of scientific and technological achievements. Calculate the comprehensive matching score between each candidate scientific and technological achievement fusion representation vector and the standardized retrieval request, and sort the candidate scientific and technological achievement fusion representation vectors in descending order according to the comprehensive matching score; The candidate scientific and technological achievement fusion representation vectors, sorted in descending order, are filtered to obtain a number of candidate scientific and technological achievement fusion representation vectors as the final matching results.

2. The method for matching supply and demand of scientific and technological achievements based on a large model according to claim 1, characterized in that, Collect original text data of scientific and technological achievements, and extract information such as title, abstract, technical field, key technology, application direction, source, production time, and maturity from the original text data of scientific and technological achievements to obtain structured information of scientific and technological achievements; Collect raw text data of enterprise technical requirements, and extract enterprise name information, requirement title information, requirement description information, application scenario information, technical target information, performance index information, constraint information, and requirement submission time information from the raw text data of enterprise technical requirements, thereby obtaining structured information of enterprise requirements.

3. The method for matching supply and demand of scientific and technological achievements based on a large model according to claim 1, characterized in that, The extended semantic information of scientific and technological achievements is generated by using a large language model according to the prompt templates corresponding to the scientific and technological achievements. The extended semantic information of scientific and technological achievements includes technical implementation path information, applicable industry information, potential application scenario information, key technical difficulties information, and upstream and downstream related technical information. The large language model generates extended semantic information about enterprise technology requirements based on prompt templates corresponding to those requirements. This extended semantic information includes technical problem breakdown information, feasible technical routes information, potential alternative technical solutions information, related application scenario information, application environment constraints information, and search tag information.

4. The method for matching supply and demand of scientific and technological achievements based on a large model according to claim 1, characterized in that, The initial semantic representation data vectors of scientific and technological achievements and the initial semantic representation data vectors of enterprise technology needs are fused with their respective extended semantic vectors according to their respective preset weights to obtain the fused representation vectors of scientific and technological achievements and the fused representation vectors of enterprise technology needs.

5. The method for matching supply and demand of scientific and technological achievements based on a large model according to claim 1, characterized in that, Key information includes the technology topic, application scenarios, core objectives, and constraints.

6. The method for matching supply and demand of scientific and technological achievements based on a large model according to claim 1, characterized in that, The multiple recall methods are vector semantic recall and full-text retrieval recall.

7. The method for matching supply and demand of scientific and technological achievements based on a large model according to claim 1, characterized in that, The overall matching score includes both text semantic matching degree and tag semantic matching degree.

8. The method for matching supply and demand of scientific and technological achievements based on a large model according to claim 1, characterized in that, A score threshold is set to filter the candidate scientific and technological achievement fusion representation vectors that have been re-sorted in descending order, resulting in multiple candidate scientific and technological achievement fusion representation vectors whose final scores are greater than or equal to the score threshold. Then, from the multiple candidate scientific and technological achievement fusion representation vectors obtained by the screening, a preset number of candidate scientific and technological achievement fusion representation vectors with the highest ranking are selected again in descending order as the final matching result.

9. A system for implementing the technology supply and demand matching method based on a large model as described in any one of claims 1-9, characterized in that, include: The data processing module is used to collect, clean, parse, and structure the raw data of scientific and technological achievements and the raw data of enterprise technology needs, and generate corresponding structured information and semantic representation data. The semantic extension module is used to call the pre-trained large language model to semantically extend the structured information of scientific and technological achievements and the structured information of enterprise needs, and form a fusion representation of scientific and technological achievements and a fusion representation of enterprise needs. The request parsing module is used to identify the intent of the matching request input by the user, extract key information and reconstruct the requirements, and generate a standardized search request; The multi-path recall module is used to perform vector semantic recall, full-text search recall, and related knowledge expansion recall based on standardized search requests, and output a set of candidate matching results; The matching calculation module is used to perform multi-dimensional matching calculations on the candidate matching result set and generate a comprehensive matching score and preliminary ranking results. The re-ranking module is used to call a large language model to perform deep semantic analysis on the preliminary ranking results, output re-ranking scores and matching explanation information, and generate the final ranking results. The results output module is used to filter the final sorting results according to a preset threshold and output the final matching results and explanation information to the user terminal.

Images