RAG-based scientific and technological achievement dynamic matching method and computer program product

Abstract

The invention discloses a scientific and technological achievement dynamic matching method based on RAG and a computer program product, and relates to the technical field of scientific and technological achievement supply and demand matching. The scientific and technological achievement dynamic matching method based on the RAG comprises the steps that a knowledge base is pre-constructed; obtaining user matching data, and analyzing the user matching data as demand matching data or result matching data; when the user matching data is demand matching data, searching scientific and technological achievement data related to the demand matching data in an index structure, and sorting the searched scientific and technological achievement data to form a first related queue; and in combination with the first correlation queue or the second correlation queue, generating a matching suggestion of the user matching data according to a big language model induction. By adopting the technology provided by the invention, rapid conversion and application of scientific and technological achievements can be promoted, deep integration of industry, university and research is further improved, and the marketization and commercialization efficiency of the scientific and technological achievements is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of scientific and technological achievement supply and demand matching, and in particular to a RAG-based scientific and technological achievement dynamic matching method and a computer program product. Background Art

[0002] The matching of supply and demand of scientific and technological achievements refers to the effective connection between the supply side and the demand side of scientific and technological achievements, so as to promote the commercialization and productization of scientific and technological achievements and accelerate the transformation of scientific and technological achievements from laboratory to market. Due to the unstructured nature of technology demand information and scientific and technological achievement information, the over-specialization of scientific and technological achievement information and the over-colloquialism of technology demand information, the matching accuracy between the supply side and the demand side is low.

[0003] Based on this, the current traditional matching of supply and demand of scientific and technological achievements focuses on processing data, using keywords and tags to identify and match the similarity between scientific and technological achievements and demand. For example, the Chinese invention patent document (CN115455304A) discloses a method for matching supply and demand of scientific and technological achievements based on big data, which mainly consults literature materials and labels them, that is, matches them with general tags and characteristic tags to generate unique tags, and then calculates weights to identify the matching degree. However, this type of data processing-based method is limited to the surface matching of keywords, lacks a deep understanding of semantics, and is difficult to handle complex matching logic and multi-level matching requirements. Summary of the invention

[0004] The present invention provides a RAG-based dynamic matching method for scientific and technological achievements and a computer program product to solve the problem that the scientific and technological achievement matching method in the prior art is limited to the surface matching of keywords and lacks a deep understanding of semantics.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is to provide a dynamic matching method of scientific and technological achievements based on RAG, and the dynamic matching method of scientific and technological achievements based on RAG includes: A knowledge base is pre-constructed, wherein the knowledge base stores multi-dimensional information of scientific and technological achievements and current candidate requirements, wherein the knowledge base also includes an index structure created by the multi-dimensional information of scientific and technological achievements and the current candidate requirements.

[0006] User matching data is obtained, and the user matching data is analyzed to be demand matching data or outcome matching data.

[0007] When the user matching data is demand matching data, scientific and technological achievement data related to the demand matching data is retrieved in the index structure, and the retrieved scientific and technological achievement data is sorted to form a first related queue.

[0008] When the user matching data is achievement matching data, candidate demand data related to the achievement matching data is retrieved in the index structure, and the retrieved candidate demand data is sorted to form a second related queue.

[0009] In combination with the first related queue or the second related queue, matching suggestions for the user matching data are generated based on the large language model.

[0010] The technical solution provided by the present invention has the following beneficial effects compared with the prior art: By storing the multi-dimensional information of scientific and technological achievements and current candidate demands through the pre-built knowledge base, and creating an index structure for the multi-dimensional information of scientific and technological achievements in each dimension, it is possible to facilitate users to quickly retrieve and improve matching accuracy. Among them, when users need to query scientific and technological achievements, they retrieve the scientific and technological achievement data related to the demand data in the index structure. When users need to match their own scientific and technological achievements with candidate demands, they retrieve the candidate demand data most relevant to the scientific and technological achievements in the index structure, which can satisfy the shared data docking between the supply side and the demand side. Promote the rapid transformation and application of scientific and technological achievements, thereby increasing the deep integration of industry, academia and research, and improving the efficiency of the marketization and commercialization of scientific and technological achievements.

[0011] In addition, the retrieved scientific and technological achievements are sorted into the first relevant queue, and the retrieved candidate needs are sorted into the second relevant queue. That is, the scientific and technological achievement data or candidate needs that best match the user are arranged and displayed according to user needs, and a large language model is used to summarize and generate summary language for customers to view and understand.

[0012] In some embodiments, the knowledge base also includes an index structure created by the multidimensional information of scientific and technological achievements and the current candidate needs, including: vectorizing the multidimensional information of scientific and technological achievements to generate a multidimensional vector of scientific and technological achievements; setting a primary index according to the multidimensional information of scientific and technological achievements, and filling the multidimensional vector of scientific and technological achievements into the primary index according to the dimension; setting a secondary index according to the classification of the primary index, and filling the multidimensional vector of scientific and technological achievements into the secondary index according to the classification; setting a tertiary index according to the sub-classification of the secondary index, and filling the multidimensional vector of scientific and technological achievements into the tertiary index according to the sub-classification.

[0013] In some implementation schemes, vectorizing the multidimensional information of scientific and technological achievements to generate a multidimensional vector of scientific and technological achievements includes: extracting multidimensional information of scientific and technological achievements from a knowledge base, wherein the multidimensional information of scientific and technological achievements includes at least the name of the scientific and technological achievements, a brief introduction to the scientific and technological achievements, and application scenarios of the scientific and technological achievements; extracting candidate demand data from the knowledge base, wherein the candidate demand data includes a demand title, a demand description, and a demand technical field.

[0014] Preprocess the multi-dimensional information of scientific and technological achievements and the candidate demand data; vectorize the processed multi-dimensional information of scientific and technological achievements and the candidate demand data to generate a multi-dimensional information vector of scientific and technological achievements and a candidate demand vector.

[0015] In some embodiments, when the user matching data is demand matching data, searching the index structure for scientific and technological achievement data related to the demand matching data, and sorting the retrieved scientific and technological achievement data to form a first related queue, includes: Vectorize the demand matching data to form a demand matching vector; calculate the similarity between the demand matching vector and the multi-dimensional information vector of scientific and technological achievements in the index structure; preliminarily sort the retrieved scientific and technological achievement data according to the similarity; use Rerank technology to optimize the preliminary sorting and generate the first related queue.

[0016] In some implementation schemes, obtaining user matching data and analyzing the user matching data into demand matching data or outcome matching data includes: The user matching data is analyzed using natural language processing technology; and the index level of the index structure corresponding to the user matching data is determined according to a routing model, wherein the routing model is pre-trained by user query history records and user feedback.

[0017] In some embodiments, combining the first related queue or the second related queue and generating a response to the user matching data according to a large language model inductively includes: Collect satisfaction feedback, the satisfaction feedback including response scores and comments on the user matching data; adjust the index structure and relevant matching parameters of the user matching data according to the satisfaction feedback; adjust the knowledge base according to the satisfaction feedback, including updating the description of scientific and technological achievements in the knowledge base.

[0018] In some implementation schemes, the pre-built knowledge base includes: structuring the storage of the multi-dimensional information of scientific and technological achievements according to dimensions; automatically capturing the latest scientific and technological achievement information from the data source according to a periodic threshold; cleaning the latest scientific and technological achievement information, and structuring the storage of the cleaned latest scientific and technological achievement information according to dimensions in the knowledge base; and updating the index structure according to the latest scientific and technological achievement information.

[0019] In some embodiments, when the user matching data is outcome matching data, retrieving candidate demand data related to the outcome matching data in the index structure, and sorting the retrieved candidate demand data to form a second related queue, includes: The result matching data is vectorized to form a result matching vector; the retrieved current demands are preliminarily sorted according to the similarity; the preliminarily sorted is optimized using the Rerank technology to generate the second related queue.

[0020] In some embodiments, combining the first related queue or the second related queue and generating matching suggestions for the user matching data according to a large language model inductively includes: An external Internet knowledge source is connected, wherein the large language model generates a matching suggestion for the user matching data according to the first relevant queue, the second relevant queue and the external Internet knowledge source.

[0021] In some embodiments, the present application also provides a computer program product, including a computer program / instruction, which implements the above-mentioned RAG-based scientific and technological achievement dynamic matching method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work, among which: Figure 1 This is a schematic diagram of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present invention. Figure 1 ; Figure 2 It is a technical structure diagram of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present invention; Figure 3 This is a schematic diagram of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present invention. Figure 2 ; Figure 4 This is a schematic diagram of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present invention. Figure 3 ; Figure 5 This is a schematic diagram of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present invention. Figure 4 . DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0024] See also Figure 1 to Figure 2 As shown, Figure 1 The schematic diagram shows a process of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present application. Figure 1 ; Figure 2 A technical structure block diagram of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present application is shown.

[0025] In some embodiments, the RAG-based dynamic matching method for scientific and technological achievements includes: Step S100: pre-build a knowledge base.

[0026] Exemplarily, the knowledge base stores multidimensional information of scientific and technological achievements and current candidate demands, wherein the multidimensional information of scientific and technological achievements includes detailed information on multiple dimensions of scientific and technological achievements, such as name, introduction, application scenario, technical field, technical problem solved, and technical effect that can be achieved. The multidimensional information of scientific and technological achievements is structured and stored in the knowledge base according to the dimensional information. Similarly, candidate demands are also stored in the knowledge base and structured storage is implemented according to technical fields, technical problems, application scenarios, etc. The index structure constructed based on candidate demands and multidimensional information of scientific and technological achievements can facilitate supply-side users and demand-side users to quickly query and match corresponding demands.

[0027] Step S200, obtaining user matching data, and analyzing the user matching data to determine whether the user matching data is demand matching data or achievement matching data.

[0028] The user's query is obtained and the natural language processing (NLP) technology is used to analyze the user matching data. Specifically, when the user enters the demand matching text, the demand matching text is first preprocessed and converted into the same format, and then tokenization, intent detection, named entity recognition (NER) and other technologies are used to understand the actual demand content of the demand matching text.

[0029] Exemplarily, it also includes: real-time monitoring of user matching in the user interface and identifying the user's information needs. For example, Real-time Information Needs Detection (RIND) is used to analyze the user's query pattern, behavior data or real-time events, so that when there is a need in the user's matching data, retrieval is automatically triggered.

[0030] Query Formulation through Self-Attention (QFS) also uses the self-attention mechanism to further analyze user matching data to understand the importance of retrieval intent and expand the original query to optimize the query statement.

[0031] For example, when a user searches for scientific and technological achievements and enters the keyword "smart grid", RIND and QFS can determine that the user intends to retrieve scientific and technological achievements related to "smart grid" and analyze whether Internet search, patent search, and paper search are required, so as to provide matching technology experts and companies to solve the user's smart grid needs when providing matching suggestions. For example, the generation of subsequent matching results also includes the characteristics of smart grid, related technologies, application fields, related experts, current research results, and future development trends.

[0032] Combination Figure 3 As shown, Figure 3 The schematic diagram shows a process of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present application. Figure 2 .

[0033] Step S300: When the user matching data is demand matching data, scientific and technological achievement data related to the demand matching data is retrieved in the index structure, and the retrieved scientific and technological achievement data is sorted to form a first related queue. Specifically, the retrieval model for querying in each dimensional index structure includes: Step S310, vectorizing the demand matching data to form a demand matching vector.

[0034] Converting the demand text after NLP processing into a vector representation can facilitate comparison with the multi-dimensional information vector of scientific and technological achievements to improve the speed of index query.

[0035] Step S320, calculating the similarity between the demand matching vector and the multi-dimensional information vector of scientific and technological achievements in the index structure.

[0036] Exemplarily, the similarity can be calculated by calculating the cosine similarity between the demand matching vector and the multi-dimensional information vector of scientific and technological achievements, or using Euclidean distance to measure the similarity between the demand matching vector and the multi-dimensional information vector of scientific and technological achievements. This application does not limit this.

[0037] Step S330, preliminarily sorting the retrieved scientific and technological achievement data according to similarity.

[0038] For example, the preliminary sorting is to sort by similarity score, and select relevant scientific and technological achievements that exceed or equal to the similarity threshold. The similarity threshold can be manually preset according to the similarity calculated between the demand matching vector and the multi-dimensional information vector of the scientific and technological achievements, or the similarity threshold can be optimized according to user satisfaction feedback.

[0039] Step S340: Use the Rerank technology to optimize the preliminary sorting and generate a first related queue.

[0040] Exemplarily, the Rerank technology is a re-ranking technology, which can further optimize the ranking of scientific and technological achievements based on the above-mentioned preliminary ranking, thereby improving the relevance and matching degree of the retrieved scientific and technological achievements to the user.

[0041] Specifically, specific features are extracted from the above-mentioned preliminary ranked scientific and technological achievements, including content relevance, context information, etc. The most influential features are selected from the extracted features using the Rerank ranking model, and the preliminary ranking results are re-ranked based on the most influential features. Exemplarily, the Rerank ranking model can be a deep learning model.

[0042] Combination Figure 4 As shown, Figure 4 The schematic diagram shows a process of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present application. Figure 3 .

[0043] Step S400: When the user matching data is the result matching data, the candidate demand data related to the result matching data is retrieved in the index structure, and the retrieved candidate demand data is sorted to form a second related queue. Step S410, vectorizing the result matching data to form a result matching vector; Step S420, preliminarily sorting the current retrieval needs according to the similarity; Step S430, optimizing the preliminary sorting using the Rerank technology to generate a second related queue.

[0044] Judge whether user matching data is demand matching data or achievement matching data to determine whether the user is a demander or a supplier, and provide a shared knowledge base for suppliers and demanders, so as to facilitate suppliers and demanders to quickly match needs, promote the rapid transformation and application of scientific and technological achievements, increase the deep integration of industry, academia and research, and improve the efficiency of marketization and commercialization of scientific and technological achievements.

[0045] Step S500: In combination with the first related queue or the second related queue, matching suggestions for user matching data are generated based on the large language model.

[0046] When a user queries for scientific and technological achievements, the first related queue generates concise and user-friendly summary language through the Large Language Models (LLM). When a user queries for candidate needs, the second related queue responds to the user through LLM. LLM builds a generation model that can summarize based on contextual information and identify key features to generate answers that better meet user needs and provide matching suggestions to users.

[0047] For example, when a user enters a brief description of the achievement in the demand query module: "This project uses silane coupling agent as raw material, and under certain conditions, uses hydrolysis condensation to synthesize siloxane resin containing active groups. The siloxane resin is not only resistant to high temperature and chemical corrosion, but its active groups can react with other resins to form high temperature resistant resins with good interfacial bonding properties and even homogeneous phases. The resin system has excellent high temperature resistance and weight retention." NLP automatically extracts the above keywords and retrieves the demand title, demand details, technical field, etc. corresponding to the achievement. It also generates matching suggestions for the company information, cooperation methods, and contact information of the person who proposed the demand.

[0048] That is, by comprehensively considering the current context information, the search results of the first related queue or the second related queue, and the knowledge base, more accurate and diverse matching suggestions can be generated for user reference.

[0049] In some embodiments, step S100, the knowledge base also includes an index structure created by multidimensional information on scientific and technological achievements and current candidate needs, including: vectorizing the multidimensional information on scientific and technological achievements to generate a multidimensional vector of scientific and technological achievements; setting a primary index based on the multidimensional information on scientific and technological achievements, and filling the multidimensional vector of scientific and technological achievements into the primary index according to the dimension; setting a secondary index based on the classification of the primary index, and filling the multidimensional vector of scientific and technological achievements into the secondary index according to the classification; setting a tertiary index based on the sub-classification of the secondary index, and filling the multidimensional vector of scientific and technological achievements into the tertiary index according to the sub-classification.

[0050] In the embodiment of the present application, the structure of the above-mentioned multi-level index is designed to facilitate the processing of complex data in the knowledge base. By assigning the multidimensional vector of scientific and technological achievements to the corresponding index entries according to the attributes, it is possible for users to locate relevant scientific and technological achievements more quickly during retrieval. Exemplarily, the first-level index, the second-level index, and the third-level index can all be used as the search entry for the next-level index to reduce the disk I / O operation during the retrieval process and increase the retrieval speed. Among them, when there are more scientific and technological achievement data and more dimensions, a fourth-level index and a fifth-level index can also be added, and this application does not limit this.

[0051] In some implementation schemes, multidimensional information of scientific and technological achievements is vectorized to generate a multidimensional vector of scientific and technological achievements, including: extracting multidimensional information of scientific and technological achievements from a knowledge base, wherein the multidimensional information of scientific and technological achievements includes at least the name of the scientific and technological achievements, a brief introduction to the scientific and technological achievements, and application scenarios of the scientific and technological achievements; extracting candidate demand data from the knowledge base, wherein the candidate demand data includes a demand title, a demand description, and a demand technical field.

[0052] Pre-process the multi-dimensional information of scientific and technological achievements and candidate demand data; vectorize the processed multi-dimensional information of scientific and technological achievements and candidate demand data to generate multi-dimensional information vectors of scientific and technological achievements and candidate demand vectors.

[0053] In the embodiment of the present application, preprocessing the multi-dimensional information of scientific and technological achievements and candidate demand data includes cleaning the data (removing irrelevant characters, special symbols, etc.), converting it into the same format and then extracting keywords, etc., so as to convert the text content into words or phrases. Figure 2 As shown in the figure, the multi-dimensional information of scientific and technological achievements and candidate demand data (text data) can be converted into numerical vectors by using the TF-IDF or Word2Vec model, and BERT Embedding can learn the vector representation of words through contextual relationships.

[0054] In some embodiments, step S200, obtaining user matching data, and analyzing the user matching data into demand matching data or outcome matching data, includes: using natural language processing technology to analyze the user matching data; and determining the index level of the index structure corresponding to the user matching data according to a routing model, wherein the routing model is pre-trained by user query history records and user feedback.

[0055] In the embodiments of the present application, the combination of NLP technology and routing model can more accurately understand the user's query intention. For example, the above-mentioned introduction of multi-level indexing and the routing model pre-trained by user query history records and user feedback can route the query to the most appropriate index structure, thereby further improving retrieval efficiency and accuracy.

[0056] Combination Figure 5 As shown, Figure 5 The schematic diagram shows a process of an embodiment of a RAG-based dynamic matching method for scientific and technological achievements provided by the present application. Figure 4 .

[0057] In some embodiments, step S500, combining the first related queue or the second related queue, and generating a response of the user matching data according to the large language model, includes: Step S510: collecting satisfaction feedback, where the satisfaction feedback includes response scores and comments on user matching data.

[0058] Step S520: adjusting the relevant matching parameters of the index structure and the user matching data according to the satisfaction feedback.

[0059] For example, user satisfaction surveys on matching results may be collected in the user interface, and the similarity thresholds mentioned above may be adjusted and the Rerank sorting model may be optimized based on user satisfaction feedback.

[0060] Step S530, adjusting the knowledge base according to the satisfaction feedback, including updating the description of scientific and technological achievements in the knowledge base.

[0061] For example, when user satisfaction feedback involves knowledge base storage data, the content of the knowledge base can also be updated according to the satisfaction feedback, such as adding missing or latest scientific and technological achievements, updating scientific and technological achievement descriptions, etc. Feedback optimization improves adaptability, thereby improving the matching degree with user needs.

[0062] In some implementation schemes, step S100, pre-building a knowledge base, includes: storing multi-dimensional information of scientific and technological achievements in a structured manner according to dimensions; automatically capturing the latest scientific and technological achievement information from a data source according to a periodic threshold; cleaning the latest scientific and technological achievement information, and storing the cleaned latest scientific and technological achievement information in a structured manner according to dimensions in the knowledge base; and updating the index structure according to the latest scientific and technological achievement information.

[0063] In the embodiment of the present application, structured storage can facilitate storage according to the key dimensions of multi-dimensional information of scientific and technological achievements, so as to facilitate subsequent matching suggestion analysis and improve retrieval matching efficiency. Among them, by setting a cycle threshold to automatically capture the latest scientific and technological achievement information, and convert it into the same format as the scientific and technological achievements stored in the knowledge base for subsequent retrieval.

[0064] For example, the cycle threshold can be set according to the frequency of hours, days, etc. The data source can be a patent library, an online database, a journal publishing platform, etc. The latest scientific and technological achievement data can be automatically acquired by configuring a web crawler or an API interface.

[0065] In some embodiments, step S500, in combination with the first relevant queue or the second relevant queue, generates matching suggestions for user matching data based on the large language model, including: connecting to an external Internet knowledge source, and the large language model generates matching suggestions for user matching data based on the first relevant queue and the second relevant queue and the external Internet knowledge source.

[0066] By connecting with external Internet knowledge sources, it is possible to dynamically obtain the latest information from external Internet knowledge sources, thereby enhancing the input of the large language model, so that it can structure the first relevant queue or the second relevant queue to generate more accurate and timely matching suggestions.

[0067] In the embodiment of the present application, the method of close integration with external Internet knowledge sources makes it possible for the search scope of users when they query to not be limited to the internal knowledge base, but also to dynamically obtain the latest news from external Internet knowledge sources, such as from news websites, social media, etc. Exemplarily, based on the analysis results of QFS, obtaining information from external Internet knowledge sources can ensure the timeliness of the user's current matching needs and provide the accuracy of matching suggestions generated by the large language model.

[0068] In some embodiments, the present application also provides a computer program product, including a computer program / instruction, which implements the above-mentioned RAG-based dynamic matching method for scientific and technological achievements when executed by a processor.

[0069] It should be understood by those skilled in the art that the embodiments of the present invention can be provided as methods, systems or computer program products. Those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0070] The above description is only an implementation mode of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly used in other related technical fields, should be included in the protection scope of the present invention.

Claims

1. A RAG-based dynamic matching method for scientific and technological achievements, characterized in that: include: Pre-constructing a knowledge base, the knowledge base storing multi-dimensional information of scientific and technological achievements and current candidate requirements, wherein the knowledge base also includes an index structure created by the multi-dimensional information of scientific and technological achievements and the current candidate requirements; Acquire user matching data, and determine whether the user matching data is demand matching data or outcome matching data; When the user matching data is demand matching data, searching the index structure for scientific and technological achievement data related to the demand matching data, and sorting the retrieved scientific and technological achievement data to form a first related queue; When the user matching data is outcome matching data, retrieving candidate demand data related to the outcome matching data in the index structure, and sorting the retrieved candidate demand data to form a second related queue; In combination with the first related queue or the second related queue, matching suggestions for the user matching data are generated based on the large language model.

2. The RAG-based dynamic matching method for scientific and technological achievements according to claim 1 is characterized in that: The knowledge base also includes an index structure created by the multi-dimensional information of scientific and technological achievements and the current candidate requirements, including: Vectorizing the multi-dimensional information of the scientific and technological achievements to generate a multi-dimensional vector of scientific and technological achievements; A primary index is set according to the multidimensional information of the scientific and technological achievements, and the multidimensional vector of the scientific and technological achievements is filled into the primary index according to the dimension; Set a secondary index according to the classification of the primary index, and fill the multidimensional vector of scientific and technological achievements into the secondary index according to the classification; A tertiary index is set according to the sub-categories of the secondary index, and the multi-dimensional vector of scientific and technological achievements is filled into the tertiary index according to the sub-categories.

3. The RAG-based dynamic matching method for scientific and technological achievements according to claim 2 is characterized in that: The step of vectorizing the multi-dimensional information of the scientific and technological achievements to generate a multi-dimensional vector of scientific and technological achievements includes: Extracting multi-dimensional information of scientific and technological achievements from the knowledge base, wherein the multi-dimensional information of scientific and technological achievements at least includes the name of the scientific and technological achievements, a brief introduction to the scientific and technological achievements, and an application scenario of the scientific and technological achievements; Extract candidate demand data from the knowledge base, wherein the candidate demand data includes demand title, demand description, and demand technical field; Preprocessing the multi-dimensional information of scientific and technological achievements and the candidate demand data; The processed multi-dimensional information of scientific and technological achievements and the candidate demand data are vectorized to generate a multi-dimensional information vector of scientific and technological achievements and a candidate demand vector.

4. The RAG-based dynamic matching method for scientific and technological achievements according to claim 1 is characterized in that: When the user matching data is demand matching data, searching the index structure for scientific and technological achievement data related to the demand matching data, and sorting the retrieved scientific and technological achievement data to form a first related queue, including: Vectorizing the demand matching data to form a demand matching vector; Calculating the similarity between the demand matching vector and the multi-dimensional information vector of scientific and technological achievements in the index structure; Preliminarily sorting the retrieved scientific and technological achievement data according to the similarity; The preliminary sorting is optimized using the Rerank technology to generate the first related queue.

5. The RAG-based dynamic matching method for scientific and technological achievements according to claim 1 is characterized in that: The obtaining of user matching data and analyzing the user matching data into demand matching data or outcome matching data includes: Analyze the user matching data using natural language processing technology; An index level of the index structure corresponding to the user matching data is determined according to a routing model, wherein the routing model is pre-trained by user query history records and user feedback.

6. The RAG-based dynamic matching method for scientific and technological achievements according to claim 1 is characterized in that: The step of combining the first related queue or the second related queue and summarizing and generating a response to the user matching data according to a large language model includes: Collecting satisfaction feedback, wherein the satisfaction feedback includes response scores and comments on the user matching data; Adjusting the relevant matching parameters of the index structure and the user matching data according to the satisfaction feedback; The knowledge base is adjusted according to the satisfaction feedback, wherein the adjusting the knowledge base includes updating the description of scientific and technological achievements in the knowledge base.

7. The RAG-based dynamic matching method for scientific and technological achievements according to claim 1 is characterized in that: The pre-built knowledge base includes: Structurally storing the multi-dimensional information of the scientific and technological achievements according to the dimensions; Automatically capture the latest scientific and technological achievements information from the data source according to the cycle threshold; Cleaning the latest scientific and technological achievement information, and storing the cleaned latest scientific and technological achievement information in a structured manner according to dimensions in the knowledge base; The index structure is updated according to the latest scientific and technological achievement information.

8. The RAG-based dynamic matching method for scientific and technological achievements according to claim 1, characterized in that: When the user matching data is outcome matching data, retrieving candidate demand data related to the outcome matching data in the index structure, and sorting the retrieved candidate demand data to form a second related queue, including: Vectorizing the result matching data to form a result matching vector; Preliminarily sorting the retrieved current demands according to the similarity; The preliminary sorting is optimized using the Rerank technology to generate the second related queue.

9. The RAG-based dynamic matching method for scientific and technological achievements according to claim 1, characterized in that: The combining the first related queue or the second related queue and generating a matching suggestion for the user matching data according to the large language model inductively includes: An external Internet knowledge source is connected, wherein the large language model generates a matching suggestion for the user matching data according to the first relevant queue, the second relevant queue and the external Internet knowledge source.

10. A computer program product, characterized in that It includes a computer program / instruction, which, when executed by a processor, implements the RAG-based dynamic matching method for scientific and technological achievements as described in any one of claims 1 to 9.

Images