Big data matching recommendation system and method based on technology transfer platform

Abstract

The invention discloses a big data matching recommendation system and method based on a technology transfer platform, and relates to the technical field of artificial intelligence, and the system comprises a data processing module, a knowledge graph construction module, an intention analysis module, a similarity calculation module and a recommendation sorting module. Wherein the data processing module adopts a field adaptive encoder to realize heterogeneous data standardization; the knowledge graph construction module extracts technical entities and relationships based on a pre-training language model; the intention analysis module extracts core features of a supply party and a demand party through a double-decoder mechanism; the similarity calculation module fuses graph structure analysis and differentiation weight distribution and realizes dynamic compensation; and the recommendation sorting module generates a personalized recommendation result in combination with the market potential evaluation. According to the invention, the intelligent matching precision of the technology transfer platform is obviously improved, the precise docking of cross-field technology supply and demand is realized, and the method has important practical value.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a big data matching and recommendation system and method based on a technology transfer platform. Background Art

[0002] The technology transfer platform is an infrastructure to promote the collaborative innovation of industry, university and research. Existing technology transfer platforms use methods such as keyword matching and text similarity calculation for supply and demand docking.

[0003] At present, the technology transfer platform has the following deficiencies in actual applications: In actual applications, since the technology information comes from multiple entities such as universities, research institutes, and enterprises, the formats and standards of the technology information provided by different entities are inconsistent, resulting in low information integration efficiency. At the same time, conventional text processing methods are difficult to accurately understand the professional terms and implicit knowledge in technical documents, reducing the matching accuracy between technology supply and demand sides. In addition, the existing technology transfer platforms lack an evaluation mechanism for the potential of the technology market, affecting the practicality of the recommendation results. The above problems limit the service effect of the technology transfer platform and affect the efficiency of technology achievement transformation. Summary of the Invention

[0004] The present invention provides a big data matching and recommendation system and method based on a technology transfer platform, and it is difficult to connect supply and demand across different fields.

[0005] In view of this, the first aspect of the present invention provides a big data matching and recommendation system based on a technology transfer platform, including: a data processing module, configured to convert a technology information set from distributed heterogeneous data sources into a standardized vector representation by using a domain adaptive encoder; a knowledge graph construction module, configured to extract technology entities and relationships in the standardized vector representation through a pre-trained large-scale language model, and construct a multi-domain fusion knowledge graph; an intention parsing module, configured to parse the intentions of both technology supply and demand sides based on the multi-domain fusion knowledge graph, extract core features and generate an intention representation vector by using a dual decoder mechanism; a similarity calculation module, configured to calculate the multi-dimensional similarity between intention representation vectors by using a multi-level heterogeneous graph network with a fusion graph structure analysis and differential weight allocation mechanism, and perform dynamic similarity compensation on the feature sparse regions; a recommendation ranking module, configured to identify the matching relationship between the technology supplier and the demander according to the multi-dimensional similarity, perform comprehensive scoring and ranking in combination with market potential evaluation indicators, and generate personalized recommendation results.

[0006] Optionally, the processing flow of the data processing module includes: Preprocess the technology information set in the distributed heterogeneous data sources, and construct a multi-level technology ontology index structure; Process structured, semi-structured, and unstructured data in the technical information set, and extract the technical term set, function description set, and performance metric set; Construct a domain adaptive encoder, including a domain weight parameter matrix, a cross-domain mapping function, and a semantic preservation loss function; Train the domain adaptive encoder to encode the technical term set, function description set, and performance metric set into an initial vector representation; Perform dimensionality reduction on the initial vector representation to obtain a standardized vector, and at the same time establish a bidirectional traceable mapping mechanism from the technical description text to the standardized vector.

[0007] Optionally, the processing flow of the knowledge graph construction module includes: Convert the standardized vector into a technical description text through the bidirectional traceable mapping mechanism, and perform technical entity recognition through a pre-trained large-scale language model to extract the technical entity set; Construct a domain term correspondence structure, perform domain adaptation transformation on the technical entity set, and form a cross-domain technical term mapping relationship; Through multi-level technical relationship analysis, extract the explicit and implicit relationships between technical entities from the technical entity set to form a technical relationship set; Use the technical entity set as nodes and the technical relationship set as edges to construct an initial graph structure, and assign weight values to the nodes and edges in the initial graph structure; Adjust the initial graph structure according to the feedback information of the preset technical matching examples, and add derivation relationships in the low-density area of the initial graph structure to form a multi-domain fusion knowledge graph.

[0008] Optionally, the processing flow of the intention parsing module includes: Construct an intention-capability double decoder, where the intention-capability double decoder includes a technical provider decoding unit and a demander decoding unit; Extract the first sub-graph structure associated with the technical provider from the multi-domain fusion knowledge graph, and input the first sub-graph structure into the technical provider decoding unit to generate a provider capability representation vector; Extract the second sub-graph structure associated with the demander from the multi-domain fusion knowledge graph, and input the second sub-graph structure into the demander decoding unit to generate a demander intention representation vector.

[0009] Optionally, the processing flow of the intention parsing module further includes: Use the technical function decomposition and recombination algorithm to decompose the provider capability representation vector to obtain a set of basic function units, and construct a function unit dependency graph; Decompose the demander intention representation vector to obtain a set of basic demand units, and construct a demand unit priority graph; Calculate the mapping relationship between the set of basic functional units and the set of basic requirement units to obtain a function-requirement matching degree matrix, and based on the function-requirement matching degree matrix, fuse and reorganize the supplier ability representation vector and the demander intention representation vector to generate an intention representation vector.

[0010] Optionally, the processing flow of the similarity calculation module includes: Construct a multi-level heterogeneous graph network, and map the supplier ability representation vector and the demander intention representation vector to the corresponding nodes; Based on the characteristics of the technical field, matching historical data, and technical maturity indicators, dynamically allocate the weight coefficients of each feature dimension to form an adaptive weight matrix; Calculate the similarity in three dimensions of direct similarity, indirect relevance, and complementary adaptability between the mapped nodes in the multi-level heterogeneous graph network to form a multi-dimensional similarity vector; Identify the feature sparse regions in the intention representation vector space and generate a multi-dimensional similarity compensation vector; Integrate the multi-dimensional similarity vector and the multi-dimensional similarity compensation vector to generate a comprehensive similarity score. Optionally, the processing flow of the recommendation ranking module includes: Based on the comprehensive similarity score, determine the set of preliminary matching relationships between the technology suppliers and demanders, and establish a technical transfer feasibility evaluation system; Quantitatively evaluate the technical transfer feasibility of each pair of suppliers and demanders in the set of preliminary matching relationships based on the technical transfer feasibility evaluation system; Adopt a weighted fusion method to combine the comprehensive similarity score with the technical transfer feasibility score to generate a final matching score; According to the final matching score, perform group ranking within the technical field for the set of preliminary matching relationships to generate a personalized recommendation result.

[0011] The second aspect of the present invention provides a big data matching and recommendation method based on a technology transfer platform, including: using a domain adaptive encoder to convert a technology information set from a distributed heterogeneous data source into a standardized vector representation; Extract technical entities and relationships from the standardized vector representation through a pre-trained large-scale language model to construct a multi-domain fusion knowledge graph; Based on the multi-domain fusion knowledge graph, adopt a dual decoder mechanism to analyze the intentions of both technology suppliers and demanders, extract core features, and generate an intention representation vector; Use a multi-level heterogeneous graph network with a fusion graph structure analysis and differential weight allocation mechanism to calculate the multi-dimensional similarity between intention representation vectors, and perform dynamic similarity compensation on the feature sparse regions; Identify the matching relationship between technology suppliers and demanders based on multi-dimensional similarity, and conduct comprehensive scoring and ranking in combination with technology transfer feasibility indicators to generate personalized recommendation results.

[0012] The beneficial effects of the present invention are as follows: The present invention realizes the unified expression and organization of heterogeneous data through a domain adaptive encoder and a multi-level ontology index structure, improving the accuracy and efficiency of technology information processing; The multi-domain fusion knowledge graph constructed based on the dual-threshold determination and dynamic optimization mechanism enhances the ability to discover technology associations and improves the integrity of knowledge representation; The dual-decoder structure and hierarchical decomposition and recombination algorithm are adopted to achieve accurate parsing and feature alignment of technology supply and demand intentions, enhancing the interpretability of the matching process; The designed three-dimensional similarity calculation model and dynamic compensation mechanism effectively improve the accuracy and robustness of technology matching and enhance the adaptability of the system to technological development and changes; Through multi-dimensional feasibility evaluation and adaptive ranking mechanism, the accuracy and practicality of the recommendation results are improved, providing a reliable basis for technology transfer decisions. Brief Description of the Drawings

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 It is a flowchart of a big data matching and recommendation system based on a technology transfer platform.

[0015] Figure 2 It is a flowchart of the intention parsing module processing of a big data matching and recommendation system based on a technology transfer platform.

[0016] Figure 3 It is a flowchart of the similarity calculation module processing of a big data matching and recommendation system based on a technology transfer platform. Detailed Embodiments

[0017] In order to make the invention objectives, features, and advantages of the present invention more obvious and understandable, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0018] Example 1, refer to Figures 1 to 3, which is the first embodiment of the present invention. This embodiment provides a big data matching and recommendation system based on a technology transfer platform. The flowchart of the system is as shown in Figure 1 shown, and the system includes: A data processing module, which is used to convert the technology information set from distributed heterogeneous data sources into a standardized vector representation through a domain adaptive encoder.

[0019] In a preferred embodiment of the present invention, the data processing module includes the following steps: preprocess the technology information set in the distributed heterogeneous data sources, and construct a multi-level technology ontology index structure according to the information source type, data format, and domain identifier. Specifically, it includes: obtaining the technology information set in the distributed heterogeneous data sources, including but not limited to patent documents, scientific and technological papers, technical reports, product manuals, and industry standards, and performing data cleaning operations on the technology information set; classifying and labeling the technology information set according to the information source type; dividing the technology information set into structured data, semi-structured data, and unstructured data according to the data format; extracting domain identifier information from the technology information set, including technology field classification codes, keyword sets, and topic tags; constructing a multi-level technology ontology index structure based on the information source type, data format, and domain identifier. This structure includes a domain layer (reflecting the relationships between different technology fields), a concept layer (representing the core technology concepts and their associations within each field), an attribute layer (describing the specific attributes and characteristics of technology concepts), and an instance layer (containing specific technology instances and their attribute values).

[0020] Furthermore, a cross-modal coding unit is used to process the structured data, semi-structured data, and unstructured data in the technology information set, and extract a technology term set, a function description set, and a performance index set. Specifically, it includes: constructing a cross-modal coding unit containing multiple processing modules, and using the cross-modal coding unit to process the structured data, semi-structured data, and unstructured data respectively; extracting technology terms and performance indicators from the structured data, parsing and extracting the technology information in the nested structure from the semi-structured data, and extracting the technology information contained in the text and graphics from the unstructured data; through cross-modal feature alignment, integrating the information extracted from different modal data to form a unified technology term set, function description set, and performance index set.

[0021] Furthermore, a domain adaptive encoder is constructed, including a domain weight parameter matrix, a cross-domain mapping function, and a semantic preservation loss function. Specifically, it includes: setting a domain weight parameter matrix for adjusting the importance of different domain features; designing a cross-domain mapping function that can map the input features to a specific domain representation space , specifically as follows: ; where x is the input data, and d is the target domain identifier. is the basic encoding function, is the domain weight parameter matrix, is the domain bias term, is the non-linear activation function.

[0022] Construct a semantic preservation loss function \(L\) for maintaining the semantic content of the original technical information sem , and the formula is as follows: ; where, L content is the content preservation loss, L structure is the structure preservation loss, L domain is the domain consistency loss, is the hyperparameter for weighing different loss components.

[0023] Furthermore, train the encoding parameters of the domain adaptive encoder through the contrastive learning method, and use the trained domain adaptive encoder to encode the technical term set, function description set and performance metric set into the initial vector representation, specifically including: establishing a contrastive learning framework containing the positive sample pair generation strategy and the negative sample selection strategy, where different expression forms of the same technical information or the expressions of the same technology in different domains are regarded as positive sample pairs, and combining the random sampling and hard example mining methods to select representative negative samples; defining the contrastive loss function as introducing the margin mechanism to improve InfoNCE, and combining it with the semantic preservation loss function to form the overall training objective \(L\) total , and the formula is as follows: ; ; where, is the weight coefficient for balancing the two losses, is the contrastive loss, is the semantic preservation loss, is the similarity gap between the negative sample and the positive sample, is the temperature parameter (used to control the scaling of the similarity score), and \(m\) is the margin hyperparameter (used to explicitly enforce the minimum similarity gap between the positive and negative samples).

[0024] It should be noted that the positive sample similarity is , and the negative sample similarity is (where ), where and are the encoding representations of the positive sample pair, is the cosine similarity function (used to calculate the similarity between two vectors and the similarity between) is the encoded representation of negative samples. Substantially, the contrastive loss function, by introducing a margin mechanism, forces the similarity gap between positive and negative samples to be at least m, while the temperature parameter adjusts the model's sensitivity to difficult samples, and the two work together to enable the encoder to learn a more discriminative feature representation.

[0025] The batch gradient descent algorithm is used to optimize the encoding parameters, including an adaptive learning rate adjustment strategy, gradient clipping to prevent gradient explosion, and an early stopping strategy to prevent overfitting; using the trained domain adaptation encoder, the technical term set, function description set, and performance metric set are converted into vector representations, and the above three vector representations are integrated into an initial vector representation.

[0026] Furthermore, dimensionality reduction is performed on the initial vector representation to obtain a standardized vector, and at the same time, a bidirectional traceable mapping mechanism from technical description text to the standardized vector is established, specifically including: using a non-linear dimensionality reduction technique to perform dimensionality reduction on the initial vector representation while retaining the structural information of the data; performing standardization processing on the dimensionality-reduced vector to ensure the stability of the vector distribution; constructing a forward mapping mechanism from technical description text to the standardized vector to map key technical terms to the vector space; constructing a reverse parsing mechanism from the standardized vector to the technical description text to decompose the standardized vector into a linear combination of technical features, establishing a feature-text template library, converting the decomposed features into natural language descriptions, and generating smooth technical description text.

[0027] Optionally, a consistency verification mechanism for the bidirectional mapping is implemented to ensure semantic preservation during the conversion process; a mapping relationship database is constructed to support the accurate retrieval and traceability analysis of technical information.

[0028] Through the data processing module, the conversion from distributed heterogeneous data to standardized vector representation is realized. This module has the following technical advantages: semantic preservation and feature extraction of technical information are achieved through the domain adaptation encoder; a standardized information organization framework is provided based on a multi-level technical ontology index structure; the accurate retrieval and analysis of technical information are supported in combination with the bidirectional traceable mapping mechanism. These advantages provide a high-quality data foundation for the matching recommendation of the technology transfer platform.

[0029] The knowledge graph construction module is used to extract technical entities and relationships in the standardized vector representation through a pre-trained large-scale language model, and construct a multi-domain fusion knowledge graph.

[0030] In another preferred embodiment of the present invention, the processing flow of the knowledge graph construction module includes: converting the standardized vector into a technical description text through a bidirectional traceable mapping mechanism, and performing technical entity recognition through a pre-trained large-scale language model to extract a technical entity set; constructing a corresponding structure of domain terms, performing domain adaptive transformation on the technical entity set, and forming a cross-domain technical term mapping relationship, specifically including: establishing a domain classification system, and performing multi-dimensional classification and labeling on technical entities; constructing a feature vector of a technical entity based on technical function characteristics and application scenario characteristics; calculating a similarity matrix of feature vectors to obtain the association strength between technical entities; constructing a cross-domain term mapping relationship based on a first preset threshold and a second preset threshold, constructing a technical entity pair with an association strength higher than the first preset threshold as an equivalent term mapping relationship, and constructing a technical entity pair with an association strength between the two preset thresholds as a related term mapping relationship; based on the constructed term mapping relationship, converting the technical entity into a unified standard representation. Wherein, the first preset threshold and the second preset threshold are determined by historical matching sample analysis and expert evaluation, and the first preset threshold is greater than the second preset threshold.

[0031] Among them, the pre-trained large-scale language model adopts a bidirectional encoder structure based on the Transformer architecture, and processes the semantic associations in technical texts through a self-attention mechanism. The model is pre-trained on a corpus containing multi-domain technical literature and patent documents, and uses masked language modeling training objectives. To perform technical entity recognition tasks, the model is fine-tuned in a targeted manner, and the sequence annotation effect is enhanced with conditional random fields, and the results are constrained by domain ontology knowledge. The model supports the recognition of composite entities and nested entities in technical texts, and can capture the accurate meaning of professional terms through context-enhanced semantic representations. The model also integrates adversarial training and multi-task learning strategies to improve cross-domain recognition capabilities while maintaining the contextual relevance of entity representations, providing basic data support for subsequent term mapping and association analysis. The hierarchical attention mechanism and boundary detection algorithm are used in the processing process to improve the recognition accuracy of complex technical concepts, and the output results retain the original context information to support bidirectional traceability.

[0032] Furthermore, through multi-level technical relationship analysis, explicit and implicit relationships between technical entities are extracted from the technical entity set to form a technical relationship set, which specifically includes: building a basic technical knowledge base, which contains predefined technical relationship types and relationship patterns; extracting explicit technical relationships based on rule templates and semantic analysis, including component composition relationships, function realization relationships, and technical evolution relationships; using knowledge reasoning and deep learning methods to mine implicit technical relationships, including technical similarity relationships, technical dependency relationships, and technical fusion relationships; and conducting credibility assessment and hierarchical classification of the identified relationships to form a structured technical relationship set.

[0033] Furthermore, an initial graph structure is constructed using the technical entity set as nodes and the technical relationship set as edges, and weight values are assigned to the nodes and edges in the initial graph structure. The initial graph structure is adjusted according to the feedback information of the preset technical matching examples, and derivation relationships are added to the low-density regions of the initial graph structure to form a multi-domain fusion knowledge graph, which specifically includes: extracting the association path features between technical entity nodes using the pre-annotated technical matching example set; updating the weight values of the corresponding nodes and edges in the initial graph structure based on the frequency and intensity of technical entity associations; calculating the node connection density of each region in the initial graph structure, and determining the regions where the node connection density is lower than the preset density threshold as low-density regions; in the low-density regions, based on the semantic relevance of the existing nodes and edges, generating new derivation relationships through transitive rules and analogical reasoning methods; evaluating the credibility of the newly generated derivation relationships, and adding the derivation relationships with credibility higher than the preset relationship threshold to the initial graph structure; repeating the above process until the initial graph structure converges to form the final multi-domain fusion knowledge graph. Among them, the node connection density is calculated by calculating the average number of connections between nodes in a unit region, the preset density threshold is determined through statistical analysis of historical data, and the preset relationship threshold is determined through expert evaluation and historical case verification.

[0034] The knowledge graph construction module of the present invention has the following technical advantages: improving the discrimination of cross-domain technical term mapping through a double-threshold determination mechanism; expanding the association types between technical entities by adopting multi-level analysis of explicit and implicit relationships; realizing the dynamic optimization of the knowledge structure based on the derivation relationship supplement mechanism in low-density regions. These improvements enhance the support effect of the knowledge graph on the technology transfer platform.

[0035] An intention parsing module, which is used to parse the intentions of both technology supply and demand sides based on the multi-domain fusion knowledge graph, extract core features, and generate an intention representation vector using a double-decoder mechanism.

[0036] In a preferred embodiment of the present invention, the processing flow chart of the intention parsing module is as Figure 2As shown in the figure, it includes: constructing an intention-capability dual decoder, where the intention-capability dual decoder includes a technology provider decoding unit and a demander decoding unit. Each decoding unit contains an attention guidance layer, a semantic aggregation layer, and a feature mapping layer. The specific design is as follows: Construct a dual decoder structure, including a technology provider decoding unit and a demander decoding unit. The two decoding units share the basic network architecture but adopt independent parameter spaces, which are used to process the subgraph structures associated with the technology provider and the demander respectively, and map the node features of the multi-domain fusion knowledge graph and their associated subgraph structures to a unified semantic space; Construct an attention guidance layer, which adopts a multi-head self-attention mechanism to adaptively select technology capability features and demand intention features through a query matrix, a key matrix, and a value matrix. Among them, the technology provider decoding unit extracts core capabilities, technology advantages, and application domain features based on the first subgraph structure, and the demander decoding unit extracts key demands, technology constraints, and application scenario features based on the second subgraph structure; Construct a semantic aggregation layer, which adopts a multi-layer perceptron structure and a residual connection mechanism to perform local-global non-linear transformation and fusion on the features in the subgraph structure, and maintains the stability of the feature distribution through layer normalization; Construct a feature mapping layer, which uses a learnable projection matrix to map the supply-side capability representation vector and the demand-side intention representation vector after feature fusion to a unified semantic space of a predetermined dimension.

[0037] Preferably, the intention-capability dual decoder of the present invention has the following technical advantages: First, it uses a dual decoder structure to replace a single decoder, and improves the adaptability to the differences in technology supply and demand characteristics through the design method of sharing the basic architecture but maintaining independent parameter spaces; Second, it designs a differential feature extraction strategy, and uses the attention mechanism to respectively strengthen the core capability features of the supply side and the key demand features of the demand side; Third, it integrates the associated subgraph structure information into the feature extraction process, and enhances the expression ability of technology relevance through the local-global feature fusion method.

[0038] Furthermore, extract the first subgraph structure associated with the technology provider from the multi-domain fusion knowledge graph, input the first subgraph structure into the technology provider decoding unit, identify the core capabilities, technology advantages, and application domains of the technology provider, and generate a supply-side capability representation vector; Extract the second subgraph structure associated with the demander from the multi-domain fusion knowledge graph, input the second subgraph structure into the demander decoding unit, identify the key demands, technology constraints, and application scenarios of the demander, and generate a demand-side intention representation vector.

[0039] Furthermore, the supply - side capability characterization vector is decomposed using the technology function decomposition and recombination algorithm to obtain a set of basic function units, and a function unit dependency graph is constructed, which specifically includes: using a recursive hierarchical algorithm to divide the function levels according to the principle of coarser - to - finer function granularity, extracting basic function units including core functions, performance functions, scenario functions, and extended functions, where the recursive hierarchical algorithm uses a top - down hierarchical clustering method; extracting characteristic parameters for each basic function unit, constructing a function unit feature matrix and mapping it to a unified representation space; analyzing the hierarchical relationships, pre - conditions, and mutual exclusion constraints between basic function units, establishing a directed acyclic graph structure to represent the function unit dependency relationship; performing a topological sort based on the function unit dependency graph, identifying the critical path, and determining the set of minimum function units, where the set of minimum function units refers to the set of all function units on the critical path.

[0040] Furthermore, the demand - side intention characterization vector is decomposed to obtain a set of basic demand units, and a demand unit priority graph is constructed, which specifically includes: using a hierarchical decomposition algorithm to divide the hierarchical structure according to demand types, extracting basic demand units including core demands, performance demands, scenario demands, and extended demands; extracting attribute features for each basic demand unit, constructing a demand unit feature matrix and mapping it to a unified representation space; analyzing the priority relationships, dependency conditions, and conflict constraints between basic demand units, establishing a directed weighted graph structure to represent the demand unit priority relationship; performing a hierarchical sort based on the demand unit priority graph, identifying the critical demand path, and determining the set of core demand units, where the set of core demand units refers to the set of all demand units on the critical demand path.

[0041] Preferably, the decomposition and recombination algorithm of the present invention has the following technical effects: through the recursive hierarchical function decomposition algorithm and the construction of the function unit dependency graph, the refined expression and structured description of technical capabilities are realized, and the recognition accuracy of technical features is improved; by using a demand priority modeling method based on a weighted graph, the critical demand path is identified through hierarchical sorting, providing a quantitative basis for demand analysis and enhancing the accuracy of demand matching; the supply - side function decomposition and the demand - side demand decomposition adopt a unified modeling framework, which is convenient for subsequent alignment and matching of supply - demand characteristics.

[0042] Furthermore, calculate the mapping relationship between the set of basic functional units and the set of basic requirement units to obtain a function-requirement matching degree matrix. Based on the function-requirement matching degree matrix, fuse and reorganize the supplier ability representation vector and the demander intention representation vector to generate an intention representation vector, which specifically includes: calculating the semantic similarity and structural similarity between the basic functional units and the basic requirement units in the unified representation space, and obtaining the comprehensive matching degree by weighting the semantic similarity and the structural similarity based on a preset weight; constructing a function-requirement matching degree matrix based on the comprehensive matching degree, normalizing the function-requirement matching degree matrix, and filtering out the low-correlation mapping relationships in the function-requirement matching degree matrix according to a preset matching threshold; using the function-requirement matching degree matrix to group the set of minimum functional units and the set of core requirement units, extracting the high-matching function-requirement pairs, and using an attention mechanism to perform weighted fusion on the function-requirement pairs to generate an intention representation vector; constructing an interpretability path for the intention representation vector, including recording the mapping relationship corresponding to the function-requirement matching degree matrix and the weight information in the weighted fusion process.

[0043] Among them, the semantic similarity is obtained by calculating the cosine distance of the unit feature vectors in the unified representation space, and the structural similarity is obtained by calculating the weighted sum of the hierarchical distance and path coincidence degree of the corresponding nodes in the functional unit dependency graph and the requirement unit priority graph; the preset weight is set according to the importance of the semantic similarity and the structural similarity in the technical feature expression, and the matching threshold is determined based on the significance level of the function-requirement mapping relationship; the extraction process of the high-matching function-requirement pairs is realized by executing the bipartite graph maximum weight matching algorithm on the function-requirement matching degree matrix; the attention mechanism calculates the attention weight based on the comprehensive matching degree of the function-requirement pairs, and performs weighted summation on the functional unit feature vectors and the requirement unit feature vectors; the interpretability path includes the mapping link from the functional unit to the requirement unit, the corresponding matching degree value, and the attention weight distribution in the fusion process.

[0044] A similarity calculation module, configured to use a multi-level heterogeneous graph network with a fused graph structure analysis and differential weight allocation mechanism to calculate the multi-dimensional similarity between intention representation vectors and perform dynamic similarity compensation on the feature sparse regions.

[0045] Specifically, the processing flow chart of the similarity calculation module is as Figure 3As shown in the figure, it includes: constructing a multi-level heterogeneous graph network, mapping the supply-side ability representation vector and the demand-side intention representation vector to the corresponding nodes, specifically including: constructing a multi-level heterogeneous graph network using a three-layer heterogeneous network structure, including a technology field layer, a technology function layer, and a technology parameter layer. The nodes in each layer represent technical features at different levels, and connections are established between nodes through corresponding association relationships; in the technology field layer, a domain node set is constructed based on the technology field ontology, an edge set is established according to the association relationships between domains, and the domain features of the supply side and the demand side are mapped to the corresponding nodes; in the technology function layer, a function node set is constructed based on the function unit dependency graph and the demand unit priority graph, an edge set is established according to the dependency relationships between functions, and the function features of the supply side and the demand side are mapped to the corresponding nodes; in the technology parameter layer, a parameter node set is constructed based on function attributes, technical indicators, and constraint conditions, an edge set is established according to the association relationships between parameters, and the parameter features of the supply side and the demand side are mapped to the corresponding nodes; constructing inter-layer connection relationships, analyzing the corresponding relationships and influence relationships between nodes at different levels in the three-layer network, establishing an inter-layer edge set, and assigning an initial weight based on technical knowledge to the inter-layer edges. The inter-layer connection relationships include mapping edges from the technology field layer to the technology function layer and mapping edges from the technology function layer to the technology parameter layer, where the weights of the mapping edges are determined based on the statistical co-occurrence frequencies in the professional knowledge base.

[0046] Preferably, the technical features are represented by a multi-level heterogeneous graph network, and the technical features of both the supply and demand sides are mapped layer by layer in the technology field layer, the technology function layer, and the technology parameter layer and inter-layer associations are established, breaking through the limitation of the expression of a single feature dimension in the prior art and improving the integrity of the expression of technical features.

[0047] Furthermore, based on the technology field characteristics, matching historical data, and technology maturity indicators, combined with the technology life cycle evaluation function, the weight coefficients of each feature dimension are dynamically allocated to form an adaptive weight matrix. Among them, the technology field characteristics include technology innovation attributes, application scenario scope, and technology development direction, which are extracted according to the node attributes in the technology field layer of the multi-level heterogeneous graph network; the matching historical data includes the distribution law and importance scoring of technical features in historical successful matching cases, and is statistically analyzed based on the historical matching records in the technology function layer and the technology parameter layer; the technology maturity indicators include the degree of technology implementation, verification completeness, and application popularity, which are obtained by analyzing the node correlation degrees in the multi-level heterogeneous graph network; the technology life cycle evaluation function is constructed based on the S-shaped curve of technology development, mapping the technology maturity indicators to temporal weight factors; a weighted combination method is used to fuse the above features, calculating the adaptive weight coefficients of each dimension feature according to the feature importance and temporal weight factors, and organizing the weight coefficients into a weight matrix corresponding to the structure of the multi-level heterogeneous graph network.

[0048] Further, calculate the similarity degrees of three dimensions, namely, direct similarity, indirect relevance, and complementary adaptability, among the mapped nodes in the multi-level heterogeneous graph network to form a multi-dimensional similarity vector, which specifically includes: calculating the direct similarity, based on the feature vectors of the corresponding nodes in the multi-level heterogeneous graph network, and using cosine similarity to calculate the matching degree of the node features of the supply and demand sides at the technical field layer, technical function layer, and technical parameter layer; calculating the indirect relevance, based on the path connection relationship between the nodes in the multi-level heterogeneous graph network, calculating the reachability intensity between the supply and demand side nodes through weighted path distance, and combining the inter-layer connection weights to obtain the association degree; calculating the complementary adaptability, based on the combined features of the supply and demand side nodes in the multi-level heterogeneous graph network, evaluating the technical matching potential of the supply and demand sides by analyzing the completeness of the dependency relationship of the function nodes and the satisfaction degree of the constraint conditions of the parameter nodes; normalizing the direct similarity, indirect relevance, and complementary adaptability, and combining them to form a multi-dimensional similarity vector representing the technical matching degree of the supply and demand sides.

[0049] Further, use the local density clustering method to identify the feature sparse regions in the vector space of the intention representation, and generate a multi-dimensional similarity compensation vector based on the distribution characteristics of historical successful matching cases, which specifically includes: calculating the local density values of each vector point in the vector space of the intention representation, including calculating the number of samples within a specified radius centered on each vector point, and determining the feature distribution density of this region based on the number of samples; setting a feature distribution density threshold to divide the vector space into regions, marking the regions with local density values less than or equal to the feature distribution density threshold as feature sparse regions, and marking the regions with local density values greater than the feature distribution density threshold as feature dense regions; screening historical successful matching cases in the feature dense regions, and extracting the feature distribution rules of these cases, including statistical features such as feature mean, variance, and correlation; based on the feature distribution rules, perform feature compensation on the vector points in the feature sparse regions to generate a multi-dimensional similarity compensation vector, where the generation of the compensation vector comprehensively considers the regional density difference, feature distribution variance, and feature distribution of adjacent dense regions.

[0050] It should be noted that when the local density values of all regions in the vector space are greater than the feature distribution density threshold, the entire vector space is marked as a feature dense region. At this time, the generated multi-dimensional similarity compensation vector is a zero compensation vector (all element values are zero). This indicates that no feature compensation is required, and the zero compensation vector will not change the original similarity value when integrated with the multi-dimensional similarity vector, but maintains the consistency and integrity of the processing flow.

[0051] Furthermore, adopt a non-linear weighted fusion method to integrate the multi-dimensional similarity vector and the multi-dimensional similarity compensation vector, where the weight coefficient is determined by minimizing the error function of historical matching samples to generate a comprehensive similarity score.

[0052] Preferably, the similarity calculation module of the present invention achieves the following technical effects: designing a dynamic weight allocation mechanism based on the technology life cycle, enhancing the adaptability of similarity calculation to the technology development trend; constructing a three-dimensional similarity calculation model, reflecting the explicit matching degree of technical features through direct similarity, mining potential technical associations through indirect relevance, and evaluating the technical integration potential through complementary adaptability, improving the recognition accuracy of technical matching; proposing a dynamic compensation method for feature sparse regions, solving the problem of similarity calculation deviation caused by uneven feature distribution, and improving the accuracy of similarity calculation.

[0053] The recommendation and ranking module is used to comprehensively score and rank the supply-demand matching pairs according to the multi-dimensional similarity, combined with the technology maturity and market potential evaluation indicators, and output a personalized recommendation list.

[0054] Specifically, the processing flow of the recommendation and ranking module includes: calculating the similarity mean and standard deviation according to the comprehensive similarity distribution of historical matching samples, setting the matching recognition benchmark threshold, and dividing the similarity interval into three matching degree intervals of high, medium, and low based on this threshold.

[0055] Furthermore, based on the comprehensive similarity score, determine the set of preliminary matching relationships between the technology supplier and the demander. Specifically: perform normalization calculation on the comprehensive similarity scores of the technology supply and demand sides, use the set matching recognition benchmark threshold for preliminary screening, and construct a candidate matching set; based on the matching distribution variance and skewness of the supply-demand matching pairs in each dimension in the candidate matching set, calculate the adaptive adjustment coefficient, and perform real-time adjustment and optimization on the matching recognition benchmark threshold; apply the adjusted matching recognition benchmark threshold to perform secondary screening on the candidate matching set, and finally determine the set of preliminary matching relationships between the supply and demand sides.

[0056] Furthermore, use the historical data of technology transfer to establish a technology transfer feasibility evaluation system including transformation difficulty indicators, implementation cycle indicators, and investment scale indicators; quantitatively score the technology transfer feasibility of the supply-demand matching pairs in the set of preliminary matching relationships based on the evaluation system, specifically including: statistically analyzing the transformation difficulty, implementation cycle, and investment scale of historical technology transfer projects, and establishing a hierarchical scoring standard based on the technology field and scale level; combining the actual situations of the technology supply and demand sides, evaluating the transformation difficulty from dimensions such as technical complexity, resource equipment matching degree, and technical team reserve; evaluating the implementation cycle indicator based on the project cycle and milestone plan; comprehensively considering the manpower, material resources, and capital requirements, evaluating the investment scale indicator; determining the weight coefficients of each indicator according to the success rate distribution of historical projects, and performing weighted calculation to obtain the comprehensive score of technology transfer feasibility.

[0057] Furthermore, a weighted fusion method is adopted to combine the comprehensive similarity score with the technical transfer feasibility score to generate a final matching score; the preliminary matching relationship set is grouped and sorted within the technical field according to the final matching score to generate a personalized recommendation result.

[0058] Further, this embodiment also provides a big data matching and recommendation method based on a technology transfer platform, including: using a domain adaptive encoder to convert a technical information set from distributed heterogeneous data sources into a standardized vector representation; extracting technical entities and relationships in the standardized vector representation through a pre-trained large-scale language model to construct a multi-domain fusion knowledge graph; based on the multi-domain fusion knowledge graph, using a dual decoder mechanism to analyze the intentions of both technology supply and demand sides, extract core features and generate intention representation vectors; using a multi-level heterogeneous graph network with a fusion graph structure analysis and differential weight allocation mechanism to calculate the multi-dimensional similarity between intention representation vectors and implement dynamic similarity compensation for feature sparse regions; identifying the matching relationship between the technology supplier and the demander according to the multi-dimensional similarity, performing comprehensive scoring and ranking in combination with technical transfer feasibility indicators to generate a personalized recommendation result.

[0059] In summary, the present invention realizes the unified expression and organization of heterogeneous data through a domain adaptive encoder and a multi-level ontology index structure, improving the accuracy and efficiency of technical information processing; constructing a multi-domain fusion knowledge graph based on a dual-threshold determination and dynamic optimization mechanism enhances the ability to discover technical associations and improves the integrity of knowledge representation; adopting a dual decoder structure and a hierarchical decomposition and recombination algorithm realizes the accurate parsing and feature alignment of technology supply and demand intentions, enhancing the interpretability of the matching process; designing a three-dimensional similarity calculation model and a dynamic compensation mechanism effectively improves the accuracy and robustness of technology matching and enhances the adaptability of the system to technological development and changes; through multi-dimensional feasibility evaluation and an adaptive ranking mechanism, the accuracy and practicality of the recommendation result are improved, providing a reliable basis for technology transfer decisions.

[0060] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A big data matching and recommendation system based on a technology transfer platform, characterized in that, Including: A data processing module, which is used to convert a technical information set from distributed heterogeneous data sources into a standardized vector representation by using a domain adaptive encoder; A knowledge graph construction module, which is used to extract technical entities and relationships in the standardized vector representation through a pre-trained large-scale language model to construct a multi-domain fusion knowledge graph; An intention parsing module, which is used to parse the intentions of both technical supply and demand sides based on the multi-domain fusion knowledge graph, adopt a dual decoder mechanism to extract core features and generate an intention representation vector; A similarity calculation module, which is used to calculate the multi-dimensional similarity between intention representation vectors by using a multi-level heterogeneous graph network with a fusion graph structure analysis and a differential weight allocation mechanism, and implement dynamic similarity compensation for feature sparse regions; A recommendation ranking module, which is used to identify the matching relationship between technical suppliers and demanders according to the multi-dimensional similarity, perform comprehensive scoring and ranking in combination with market potential evaluation indicators, and generate personalized recommendation results.

2. The big data matching and recommendation system based on the technology transfer platform according to claim 1, characterized in that The processing flow of the data processing module includes: Preprocessing the technical information set in the distributed heterogeneous data sources to construct a multi-level technical ontology index structure; Processing the structured data, semi-structured data and unstructured data in the technical information set to extract a technical term set, a function description set and a performance index set; Constructing a domain adaptive encoder, including a domain weight parameter matrix, a cross-domain mapping function and a semantic preservation loss function; Training the domain adaptive encoder to encode the technical term set, the function description set and the performance index set into an initial vector representation; Performing dimensionality reduction processing on the initial vector representation to obtain a standardized vector, and at the same time establishing a bidirectional traceable mapping mechanism from technical description text to the standardized vector.

3. The big data matching and recommendation system based on the technology transfer platform according to claim 1, characterized in that, The processing flow of the knowledge graph construction module includes: Converting the standardized vector into a technical description text through the bidirectional traceable mapping mechanism, and performing technical entity recognition through a pre-trained large-scale language model to extract a technical entity set; Constructing a domain term correspondence structure to perform domain adaptation transformation on the technical entity set to form a cross-domain technical term mapping relationship; Through multi-level technical relationship analysis, extracting explicit relationships and implicit relationships between technical entities from the technical entity set to form a technical relationship set; Using the technical entity set as nodes and the technical relationship set as edges to construct an initial graph structure, and assigning weight values to the nodes and edges in the initial graph structure; Adjusting the initial graph structure according to the feedback information of preset technical matching examples, and adding derivation relationships in the low-density region of the initial graph structure to form a multi-domain fusion knowledge graph.

4. The big data matching and recommendation system based on the technology transfer platform according to claim 1, wherein The processing flow of the intention parsing module includes: Constructing an intention-capability dual decoder, where the intention-capability dual decoder includes a technical supplier decoding unit and a demander decoding unit; Extracting a first sub-graph structure associated with the technical supplier from the multi-domain fusion knowledge graph, and inputting the first sub-graph structure into the technical supplier decoding unit to generate a supplier capability characterization vector; Extract the second sub-graph structure associated with the demander from the multi-domain fusion knowledge graph, and input the second sub-graph structure into the demander decoding unit to generate a demander intention representation vector.

5. The big data matching and recommendation system based on a technology transfer platform according to claim 1, wherein The processing flow of the intention parsing module further includes: Decompose the supplier ability representation vector using the technology function decomposition and recombination algorithm to obtain a set of basic function units, and construct a function unit dependency graph; Decompose the demander intention representation vector to obtain a set of basic demand units, and construct a demand unit priority graph; Calculate the mapping relationship between the set of basic function units and the set of basic demand units to obtain a function-demand matching degree matrix, and based on the function-demand matching degree matrix, fuse and recombine the supplier ability representation vector and the demander intention representation vector to generate an intention representation vector.

6. The big data matching and recommendation system based on the technology transfer platform according to claim 1, wherein The processing flow of the similarity calculation module includes: Construct a multi-level heterogeneous graph network, and map the supplier ability representation vector and the demander intention representation vector to corresponding nodes; Based on the characteristics of the technical field, matching historical data, and technical maturity indicators, dynamically allocate the weight coefficients of each feature dimension to form an adaptive weight matrix; Calculate the similarity in three dimensions, namely direct similarity, indirect relevance, and complementary adaptability, between the mapped nodes in the multi-level heterogeneous graph network to form a multi-dimensional similarity vector; Identify the feature sparse region in the intention representation vector space and generate a multi-dimensional similarity compensation vector; Integrate the multi-dimensional similarity vector and the multi-dimensional similarity compensation vector to generate a comprehensive similarity score.

7. The big data matching and recommendation system based on the technology transfer platform according to claim 1, characterized in that The processing flow of the recommendation ranking module includes: Determine a set of preliminary matching relationships between the technology supplier and the demander based on the comprehensive similarity score, and establish a technology transfer feasibility evaluation system; Quantitatively score the technology transfer feasibility of each pair of suppliers and demanders in the set of preliminary matching relationships based on the technology transfer feasibility evaluation system; Use a weighted fusion method to combine the comprehensive similarity score and the technology transfer feasibility score to generate a final matching score; Group and rank the set of preliminary matching relationships within the technical field according to the final matching score to generate a personalized recommendation result.

8. A big data matching and recommendation method based on a technology transfer platform, characterized in that, Including: Use a domain adaptive encoder to convert the technical information set from distributed heterogeneous data sources into a standardized vector representation; Extract technical entities and relationships from the standardized vector representation through a pre-trained large-scale language model to construct a multi-domain fusion knowledge graph; Based on the multi-domain fusion knowledge graph, adopt a dual decoder mechanism to parse the intentions of both technology suppliers and demanders, extract core features, and generate intention representation vectors; Use a multi-level heterogeneous graph network with a fusion graph structure analysis and differential weight allocation mechanism to calculate the multi-dimensional similarity between intention representation vectors, and implement dynamic similarity compensation for the feature sparse region; Identify the matching relationship between the technology supplier and the demander according to the multi-dimensional similarity, and conduct a comprehensive scoring and ranking in combination with the technology transfer feasibility index to generate a personalized recommendation result.

Images