A method, device and medium for matching scientific and technological achievements based on dynamic weight fusion and atlas association

Abstract

The present application relates to a kind of based on dynamic weight fusion and atlas association scientific and technological achievement matching method, equipment and medium, constructs scientific and technological achievement library and corresponding scientific and technological field knowledge graph, constructs multidimensional image for each scientific and technological achievement and technical demand;The semantic matching score between each technical demand and each scientific and technological achievement, atlas association score based on knowledge graph and collaborative filtering score based on historical interaction behavior are calculated, based on the situation characteristics including demand urgency, achievement maturity matching ratio, strategic fit and the correlation between scores, the dynamic fusion weight corresponding to scientific and technological achievement matching score is calculated by pre-trained dynamic weight generation model;The semantic matching score, atlas association score, collaborative filtering score and corresponding dynamic fusion weight are input into nonlinear comprehensive score function, and the final matching score of each scientific and technological achievement is calculated.Compared with prior art, the present application has the advantages of high decision reliability, strong practicality and matching accuracy.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to a method, device and medium for matching scientific and technological achievements based on dynamic weight fusion and graph association. Background Technology

[0002] In the field of science and technology innovation management, especially in large institutions such as power grid companies, a large number of scientific and technological achievements have been accumulated from science and technology projects, such as patents, technical reports, and software systems, as well as technical needs from business departments. How to efficiently and accurately match the massive and diverse scientific and technological achievements with specific technical needs is a key link in improving the efficiency of achievement transformation and driving business innovation.

[0003] Currently, there are two main technical approaches in the industry for achieving this type of matching. The first is keyword or tag-based retrieval matching. This method segments the text descriptions of achievements and needs, extracts keywords, and calculates word frequency or Boolean matching degree. This method is simple to implement, but its drawback is that it heavily relies on the selection and consistency of keywords, making it difficult to understand the deeper semantics behind the text. It also cannot identify semantically similar but differently expressed concepts such as "distributed energy" and "decentralized power generation," resulting in limited matching accuracy and frequent missed and incorrect matches. To improve semantic understanding, the second approach introduces semantic vector matching technology based on pre-trained language models (such as BERT). This method transforms the text descriptions of achievements and needs into high-dimensional semantic vectors and evaluates the matching degree by calculating the cosine similarity between vectors. This method overcomes the semantic limitations of keyword matching to some extent. However, its matching logic is still single-dimensional and linear. It cannot utilize the complex network of relationships that objectively exists between scientific and technological achievements, between technologies and experts, and between problems and methods. For example, a breakthrough in cable fault location might be strongly correlated with research on acoustic detection algorithms and a signal processing expert. However, simple semantic vector matching struggles to automatically uncover and utilize such implicit, multi-hop associations, thus limiting the breadth and depth of the matching. Furthermore, in real-world business scenarios, matching priorities and focuses are not static. For instance, the technical needs for emergency fault repair differ drastically from the needs for long-term strategic technology reserves, placing vastly different demands on the technological maturity, response speed, and strategic alignment of the results. Existing matching methods, whether based on keywords or semantic vectors, typically employ static presets or simple rule adjustments, lacking a refined, dynamic decision-making mechanism capable of adapting to different business contexts.

[0004] In summary, existing technologies suffer from the following main problems: First, the matching dimensions are limited, focusing either on surface text or a single semantic element, failing to systematically integrate multi-dimensional information such as semantics, relationships, and historical preferences. Second, the matching strategies are rigid, unable to dynamically adjust and optimize based on real-time and changing business contexts, such as urgency and strategic orientation, leading to a disconnect between recommended results and actual business needs. Finally, the matching process lacks a reasonable explanation for why a particular result is recommended, reducing user trust in the system and their willingness to adopt it. Therefore, developing intelligent matching methods that deeply integrate multi-dimensional matching signals, intelligently adapt to complex business contexts, and improve the interpretability of decisions are technical problems that need to be addressed. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a method, device and medium for matching scientific and technological achievements based on dynamic weight fusion and graph association. By combining a dynamic weight generation model based on contextual features with a nonlinear comprehensive scoring function including geometric mean interaction terms, the matching system can intelligently adapt to changing business scenarios and accurately identify high-value achievements that perform well in multiple dimensions such as semantics, association and preference, thereby improving the business fit and decision reliability of the recommendation results.

[0006] The objective of this invention can be achieved through the following technical solutions: According to one aspect of the present invention, a method for matching scientific and technological achievements based on dynamic weight fusion and graph association is provided, comprising the following steps: S1, collecting scientific and technological achievement information from the project completion materials and user-submitted technical requirement information; S2, constructing a scientific and technological achievement database and a corresponding scientific and technological field knowledge graph, and constructing multi-dimensional profiles for each scientific and technological achievement and technical requirement; S3, calculating a matching score for the scientific and technological achievement, including semantic matching scores between each technical requirement and each scientific and technological achievement, graph association scores based on the knowledge graph, and collaborative filtering scores based on historical interaction behavior; S4, calculating dynamic fusion weights corresponding to the matching scores of the scientific and technological achievements based on contextual features including demand urgency, achievement maturity matching ratio, strategic fit, and correlation between scores through a pre-trained dynamic weight generation model; S5, inputting the semantic matching score, graph association score, collaborative filtering score, and corresponding dynamic fusion weights into a nonlinear comprehensive scoring function to calculate the final matching score of each scientific and technological achievement; sorting all scientific and technological achievements according to the final matching scores, generating a matching recommendation list, and outputting it.

[0007] Furthermore, the collection of scientific and technological achievement information in S1 specifically includes: The process involves OCR recognition or structured reading of PDF and Word documents in the project completion materials, converting them into text in a unified format; cleaning, word segmentation, and stop word removal of the converted text; and using named entity recognition technology to extract structured scientific and technological achievement information from the text, including achievement name, technical field, innovation points, patent number, application scenario, and team members.

[0008] Furthermore, the construction of a knowledge graph in the technology field in S2 specifically includes: defining entity types including scientific and technological achievements, technical terms, researchers, application departments, and pain points solved; defining the types of relationships between entities including affiliation, application, solution, and cooperation; and storing the scientific and technological achievement information and semantic relationships between entities extracted in S1 into a graph database to form a knowledge graph.

[0009] Furthermore, in step S3, the specific calculation steps for the semantic matching score include: Using a pre-trained language model, the textual descriptions of technical requirements and scientific and technological achievements are encoded into high-dimensional semantic vectors, respectively. The cosine similarity between the high-dimensional semantic vectors of technical requirements and scientific and technological achievements is calculated to obtain an initial semantic similarity. Based on the overlapping technical field labels in the multi-dimensional profiles of the technical requirements and scientific and technological achievements, the initial semantic similarity is weighted and corrected to obtain the semantic matching score.

[0010] Furthermore, in step S3, the specific calculation steps for the map association score include: In the knowledge graph, starting with the entity corresponding to the technical requirement and the entity corresponding to the scientific and technological achievement, a multi-hop path query is performed to find all relational paths connecting the two entities; based on the length of the relational path and the preset weight of the relation type on the relational path, the association strength of each path is calculated; the association strength of all relational paths is aggregated to obtain the graph association score of the scientific and technological achievement.

[0011] Furthermore, in step S3, the specific calculation steps for the collaborative filtering score include: Collect historical behavioral data of users towards scientific and technological achievements, including browsing, collection, consultation, and successful connection records, and construct a user-scientific and technological achievement interaction matrix; use a matrix factorization model to train the user-scientific and technological achievement interaction matrix to obtain user latent semantic vectors and scientific and technological achievement latent semantic vectors; based on the target user latent semantic vector corresponding to the technical needs, calculate the cosine similarity with the latent semantic vectors of each scientific and technological achievement to obtain the collaborative filtering score of the scientific and technological achievement.

[0012] Furthermore, the dynamic weight generation model in S4 is a multilayer perceptron, with the number of input layer nodes being the same as the dimension of the context features, and the output layer having three nodes using the Softmax activation function. The sum of the three dynamically fused weights output is 1.

[0013] Furthermore, the nonlinear comprehensive scoring function in S5 includes a nonlinear transformation term based on the Sigmoid function and an interaction term based on the geometric mean of multiple matching scores, which calculates the final matching score. The expression is: , in, For semantic matching scores, For the graph association score, For collaborative filtering scores, , and For the corresponding dynamic fusion weights, The preset interaction item coefficients, k The parameter controls the steepness of the function curve. This is the preset score benchmark threshold.

[0014] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0015] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) It realizes the deep integration of multi-dimensional matching information and dynamic decision-making with context adaptation, which significantly improves the matching accuracy and business fit: This invention calculates the matching relationship from three dimensions: text, knowledge and behavior by synchronously calculating semantic matching score, graph association score and collaborative filtering score; and introduces a dynamic weight generation model driven by context features, so that the system can automatically and finely adjust the fusion weight of different score dimensions according to real-time business parameters such as urgency of demand and strategic fit, and apply the matching strategy to the management of scientific and technological achievements, so that the recommendation results can not only more accurately hit the core of technology, but also highly fit different business scenarios, such as the priority requirements of emergency repair and strategic research and development, and solve the problem of the disconnect between the recommendation results and the actual business.

[0017] (2) By introducing nonlinear comprehensive scoring and collaborative interaction terms, the robustness of matching decisions and the ability to identify high-value results are enhanced: In the fusion stage, a comprehensive scoring function containing Sigmoid nonlinear transformation and geometric mean interaction terms is designed. The threshold effect in matching decisions is simulated by the Sigmoid function, which enhances the discriminative power of the results. Its core geometric mean interaction term captures the synergistic effect of multiple evidence consistency, which can effectively suppress the influence of single-dimensional outliers and improve the final score of high-value results in the three dimensions of semantics, association and preference. This enables the scientific and technological achievement management system using this method to more reliably identify and recommend core achievements that are technically sound, widely associated and recognized by users, thereby improving the intelligence level of the matching system and the accuracy of the recommendation results.

[0018] (3) An interpretable matching link from data to knowledge to decision-making was constructed, which improved the credibility and practicality of the system: The present invention constructs a knowledge graph containing rich entities and relationships from structured data, performs multi-hop path query and association strength calculation based on the graph, considers the correlation between scores as contextual features in the dynamic weight model, and finally clearly reflects the contribution and synergistic effect of each item in the scoring function. Each matching recommendation has a clear and traceable decision basis, which can be explained as follows: technology A and demand B have high semantic similarity, and there is a two-hop association in the knowledge graph through expert C. At the same time, users with similar history also prefer this technology. The deep interpretability enhances the trust and willingness of business experts and managers to adopt the system recommendation results, resulting in a transparent platform for assisting decision-making and knowledge discovery, and improving the user experience. Attached Figure Description

[0019] Figure 1 This is a flowchart of a scientific and technological achievement matching method based on dynamic weight fusion and graph association. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] Example 1 like Figure 1 The image shows a method for matching scientific and technological achievements based on dynamic weight fusion and graph association. The specific steps include: S1. Collect information on scientific and technological achievements from the project completion materials and information on technical requirements submitted by users; S2. Construct a database of scientific and technological achievements and a corresponding knowledge graph of scientific and technological fields, and build multi-dimensional profiles for each scientific and technological achievement and technical need. S3: Calculate the matching score of scientific and technological achievements, including the semantic matching score between various technical requirements and scientific and technological achievements, the graph association score based on knowledge graph, and the collaborative filtering score based on historical interaction behavior. S4. Based on contextual features including demand urgency, achievement maturity matching ratio, strategic fit and correlation between scores, a pre-trained dynamic weight generation model is used to calculate the dynamic fusion weights corresponding to the matching scores of scientific and technological achievements. S5. Input the semantic matching score, graph association score, collaborative filtering score, and corresponding dynamic fusion weights into the nonlinear comprehensive scoring function to calculate the final matching score of each scientific and technological achievement; sort all scientific and technological achievements according to the final matching score, generate a matching recommendation list and output it.

[0022] The specific steps for collecting scientific and technological achievement information in S1 are as follows: OCR recognition or structured reading of PDF and Word documents in the project completion materials, converting them into text in a unified format; cleaning, segmenting, and removing stop words from the converted text; and using named entity recognition technology to extract structured scientific and technological achievement information from the text, including achievement name, technical field, innovation points, patent number, application scenario, and team members.

[0023] The sources of information on scientific and technological achievements mainly include documents from completed scientific and technological projects in the internal scientific and technological project management system, such as project completion reports and technical summaries, usually in PDF or Word format; and technical requirement forms submitted by users through the platform's front end, which include structured or semi-structured information such as text descriptions and category tags. This raw data forms the basis for subsequent processing.

[0024] Named entity recognition (NER) technology involves inputting text into a NER model. This model is based on sequence labeling architecture and has been fine-tuned for use with scientific and technological texts. The model predicts the entity category label word by word, such as achievement name, technical field, innovation point, patent number, application scenario, and team members. Based on the predicted label sequence, the corresponding text fragments are extracted and organized into structured scientific and technological achievement information in key-value pair format.

[0025] The construction of a knowledge graph in the technology field in S2 specifically includes: defining entity types including scientific and technological achievements, technical terms, researchers, application departments, and pain points solved; defining the types of relationships between entities including affiliation, application, solution, and cooperation; and storing the scientific and technological achievement information and semantic relationships between entities extracted in S1 into a graph database to form a knowledge graph.

[0026] First, natural language processing techniques are used to extract structured information from unstructured documents, forming a standardized repository of scientific and technological achievements. Second, based on the extracted entities and relationships, a structured knowledge graph of the scientific and technological domain is constructed. For each scientific and technological achievement and each technological requirement in the repository, a structured multidimensional profile data object is constructed according to its multidimensional attributes, including technical field, application scenario, and maturity level.

[0027] In S3, for a specific target technology requirement, three types of technology achievement matching scores are calculated in parallel between the requirement and each candidate achievement in the technology achievement database. The semantic matching score is obtained by analyzing the closeness of the text descriptions of the requirement and the achievement in the semantic space; the graph association score is obtained by querying and calculating the association path strength between two corresponding entities in the technology domain knowledge graph; and the collaborative filtering score is obtained by analyzing historical interaction data, i.e., the user-achievement behavior matrix, to predict the preference strength of the target user, i.e., the requirement submitter, for the candidate achievements.

[0028] Specifically, the specific steps for calculating the semantic matching score include: Using a pre-trained language model, the textual descriptions of technical requirements and scientific and technological achievements are encoded into high-dimensional semantic vectors, respectively. The cosine similarity between the high-dimensional semantic vectors of technical requirements and scientific and technological achievements is calculated. The cosine similarity measures the directional consistency of the two vectors by measuring the cosine of the angle between them in space. The value range is [-1,1], which is usually mapped to the [0,1] interval after processing in this scenario to obtain the initial semantic similarity. Based on the overlapping technical field labels in the multi-dimensional profiles of technical requirements and scientific and technological achievements, the initial semantic similarity is weighted and corrected to obtain the semantic matching score.

[0029] The specific steps of the correction include: extracting a set of technical field labels from the multi-dimensional profiles of the target technical needs and candidate scientific and technological achievements; calculating the overlap between these two label sets; and making corrections based on the magnitude of the overlap. If the overlap is high, the initial semantic similarity is multiplied by a coefficient greater than 1 for upward correction; if the overlap is low but the initial similarity is high, it may mean that the semantics are related but the domains do not match, in which case it can be appropriately downward corrected or kept unchanged. Through this weighted correction step, the final semantic matching score is obtained. This makes the matching not only consider textual semantics but also incorporate structured domain knowledge, resulting in greater accuracy.

[0030] The specific steps for calculating the map association score include: In the knowledge graph, starting with the entity corresponding to the technical requirement and the entity corresponding to the scientific and technological achievement, a multi-hop path query is performed to find all relational paths connecting the two entities. The path consists of a series of alternating nodes and edges, and the path length, i.e. the number of hops, may be 1 (directly connected), 2, 3, etc. Based on the length of the relational path and the preset weight of the relation type on the relational path, the association strength of each path is calculated. Among them, the longer the path, the greater the strength decay. The association strength of all relational paths is aggregated to obtain the graph association score of the scientific and technological achievement.

[0031] In S3, the specific steps for calculating the collaborative filtering score include: Collect historical user behavior data related to scientific and technological achievements, including browsing, collection, consultation, and successful connection records. For each user-achievement pair, calculate the comprehensive interaction intensity value using a weighted attenuation formula based on all historical behavior types, frequencies, and times. Organize all user interaction intensity values ​​into a user-scientific and technological achievement interaction matrix R, where the row index represents the user and the column index represents the scientific and technological achievement. R[ i ][ j The value of ] is the user i Results j The interaction strength is determined by setting most cells to 0, forming a sparse user-technology achievement interaction matrix. A matrix factorization model is used to train the interaction matrix, finding two low-dimensional matrices whose elements are user latent semantic vectors and technology achievement latent semantic vectors, respectively, such that their product approximates the original interaction matrix as closely as possible. After training, for the target user corresponding to the target technology need, their corresponding user latent semantic vector can be obtained from the user latent semantic vector matrix; for each candidate technology achievement, its corresponding technology achievement latent semantic vector can be obtained from the technology achievement latent semantic vector matrix. These vectors encode user preferences and achievement characteristics in a low-dimensional "latent space." Based on the target user latent semantic vector corresponding to the technology need and the latent semantic vectors of each technology achievement, cosine similarity is calculated to obtain the collaborative filtering score of the technology achievement, reflecting the degree of preference of the target user for the achievement based on group behavior patterns.

[0032] The dynamic weight generation model in S4 is a multilayer perceptron. The number of nodes in the input layer is the same as the dimension of the contextual features. The output layer has three nodes and uses the Softmax activation function. Nonlinear feature combination is performed through the first-layer linear transformation and the ReLU activation function. The sum of the three dynamic fusion weights output is 1.

[0033] In S5, the matching score of the scientific and technological achievement calculated in S3 and the dynamic fusion weight generated in S4 are input into a nonlinear comprehensive scoring function to calculate the final matching score for each candidate achievement. The system then sorts all candidate achievements in descending order according to the final matching score, generates a structured matching recommendation list, and outputs it to the user through the user interface.

[0034] The nonlinear comprehensive scoring function in S5 includes a nonlinear transformation term based on the Sigmoid function and an interaction term based on the geometric mean of multiple matching scores, which calculates the final matching score. The expression is: , in, For semantic matching scores, For the graph association score, For collaborative filtering scores, , and For the corresponding dynamic fusion weights, The preset interaction item coefficients, k The parameter controls the steepness of the function curve. This is the preset score benchmark threshold.

[0035] Example 2 According to the method in Example 1, this example provides a smart matching system for scientific and technological achievements based on dynamic weight fusion and graph association, including: a data acquisition and processing module, a knowledge construction and profiling module, a multi-dimensional matching calculation module, a dynamic weight generation module, and a comprehensive scoring and output module.

[0036] Among them, the knowledge construction and profiling module is used to build a science and technology achievement database and a corresponding science and technology field knowledge graph based on science and technology achievement information and technology demand information, and to build multi-dimensional profiles for each science and technology achievement and technology demand.

[0037] The multi-dimensional matching calculation module is used to calculate the matching score of scientific and technological achievements, including: a semantic matching calculation unit, which is used to calculate the semantic matching score between each technical requirement and each scientific and technological achievement; a graph association calculation unit, which is used to calculate the graph association score between each technical requirement and each scientific and technological achievement based on the knowledge graph of the scientific and technological field; and a collaborative filtering calculation unit, which is used to calculate the collaborative filtering score between each technical requirement and each scientific and technological achievement based on the user's historical interaction behavior with scientific and technological achievements.

[0038] The dynamic weight generation module is used to calculate dynamic fusion weights corresponding to semantic matching scores, graph association scores, and collaborative filtering scores based on contextual features and through a pre-trained dynamic weight generation model. Contextual features include demand urgency, achievement maturity matching ratio, strategic fit, and correlation between at least two matching scores.

[0039] The comprehensive scoring and output module is used to input semantic matching scores, graph association scores, collaborative filtering scores, and corresponding dynamic fusion weights into a nonlinear comprehensive scoring function to calculate the final matching score of each scientific and technological achievement, and to sort all scientific and technological achievements according to the final matching scores, generating and outputting a matching recommendation list.

[0040] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0041] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0042] Multiple components in the device are connected to an I / O interface, including: input units such as a keyboard, mouse, etc.; output units such as various types of displays, speakers, etc.; storage units such as disks, optical disks, etc.; and communication units such as network interface cards, modems, wireless transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks. The processing unit performs the various methods and processes described above, such as the method of the present invention. For example, in some embodiments, the method of the present invention may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or the communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the method of the present invention described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute the method of the present invention by any other suitable means (e.g., by means of firmware).

[0043] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0044] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0045] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0046] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for matching scientific and technological achievements based on dynamic weight fusion and graph association, characterized in that, The specific steps include: S1, collecting scientific and technological achievement information from the project completion materials and user-submitted technical requirement information; S2, constructing a scientific and technological achievement database and a corresponding knowledge graph of scientific and technological fields, and building multi-dimensional profiles for each scientific and technological achievement and technical requirement; S3, calculating the matching score of scientific and technological achievements, including the semantic matching score between each technical requirement and each scientific and technological achievement, the graph association score based on the knowledge graph, and the collaborative filtering score based on historical interaction behavior; S4, calculating the dynamic fusion weight corresponding to the matching score of the scientific and technological achievement through a pre-trained dynamic weight generation model based on contextual features including demand urgency, achievement maturity matching ratio, strategic fit, and correlation between scores; S5, inputting the semantic matching score, graph association score, collaborative filtering score, and corresponding dynamic fusion weight into a nonlinear comprehensive scoring function to calculate the final matching score of each scientific and technological achievement; sorting all scientific and technological achievements according to the final matching score, generating a matching recommendation list, and outputting it.

2. The method for matching scientific and technological achievements based on dynamic weight fusion and graph association according to claim 1, characterized in that, The specific process of collecting scientific and technological achievement information in S1 is as follows: The process involves OCR recognition or structured reading of PDF and Word documents in the project completion materials, converting them into text in a unified format; cleaning, word segmentation, and stop word removal of the converted text; and using named entity recognition technology to extract structured scientific and technological achievement information from the text, including achievement name, technical field, innovation points, patent number, application scenario, and team members.

3. The method for matching scientific and technological achievements based on dynamic weight fusion and graph association according to claim 1, characterized in that, The construction of a knowledge graph in the technology field in S2 specifically includes: defining entity types including scientific and technological achievements, technical terms, researchers, application departments, and pain points solved; defining the types of relationships between entities including affiliation, application, solution, and cooperation; and storing the scientific and technological achievement information and semantic relationships between entities extracted in S1 into a graph database to form a knowledge graph.

4. The method for matching scientific and technological achievements based on dynamic weight fusion and graph association according to claim 1, characterized in that, In step S3, the specific steps for calculating the semantic matching score include: Using a pre-trained language model, the textual descriptions of technical requirements and scientific and technological achievements are encoded into high-dimensional semantic vectors, respectively. The cosine similarity between the high-dimensional semantic vectors of technical requirements and scientific and technological achievements is calculated to obtain an initial semantic similarity. Based on the overlapping technical field labels in the multi-dimensional profiles of the technical requirements and scientific and technological achievements, the initial semantic similarity is weighted and corrected to obtain the semantic matching score.

5. The method for matching scientific and technological achievements based on dynamic weight fusion and graph association according to claim 1, characterized in that, In step S3, the specific calculation steps for the map association score include: In the knowledge graph, starting with the entity corresponding to the technical requirement and the entity corresponding to the scientific and technological achievement, a multi-hop path query is performed to find all relational paths connecting the two entities; based on the length of the relational path and the preset weight of the relation type on the relational path, the association strength of each path is calculated; the association strength of all relational paths is aggregated to obtain the graph association score of the scientific and technological achievement.

6. The method for matching scientific and technological achievements based on dynamic weight fusion and graph association according to claim 1, characterized in that, In step S3, the specific calculation steps for the collaborative filtering score include: Collect historical behavioral data of users towards scientific and technological achievements, including browsing, collection, consultation, and successful connection records, and construct a user-scientific and technological achievement interaction matrix; use a matrix factorization model to train the user-scientific and technological achievement interaction matrix to obtain user latent semantic vectors and scientific and technological achievement latent semantic vectors; based on the target user latent semantic vector corresponding to the technical needs, calculate the cosine similarity with the latent semantic vectors of each scientific and technological achievement to obtain the collaborative filtering score of the scientific and technological achievement.

7. The method for matching scientific and technological achievements based on dynamic weight fusion and graph association according to claim 1, characterized in that, The dynamic weight generation model in S4 is a multilayer perceptron. The number of nodes in the input layer is the same as the dimension of the context features. The output layer has three nodes and uses the Softmax activation function. The sum of the three dynamically fused weights output is 1.

8. The method for matching scientific and technological achievements based on dynamic weight fusion and graph association according to claim 1, characterized in that, The nonlinear comprehensive scoring function in S5 includes a nonlinear transformation term based on the Sigmoid function and an interaction term based on the geometric mean of multiple matching scores, which calculates the final matching score. The expression is: ， in, For semantic matching scores, For the graph association score, For collaborative filtering scores, , and For the corresponding dynamic fusion weights, The preset interaction item coefficients, k The parameter controls the steepness of the function curve. This is the preset score benchmark threshold.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images