Identification method for experts and industry-university-research institutions in key core technology attack field

Abstract

The invention discloses an identification method for experts and industry-university-research institutions in the key core technology breakthrough field. The method comprises the following steps: acquiring a scientific research literature lexicon in the key core technology field; based on the scientific research literature lexicon, constructing a domain expert knowledge network oriented to the key core technology to obtain domain expert data of key core technology attacks; constructing an identification network of an industry-university-research cooperation mechanism based on the key core technical field to select industry-university-research mechanism data to be cooperated; and obtaining a matching similarity between the field expert data and the to-be-cooperated industry-university-research institution data, and determining a field expert selected by the to-be-cooperated industry-university-research institution according to the matching similarity. The method solves the problems that an existing domain expert database is single and not detailed enough in data recording of domain experts, or limited to manual collection of personnel or registration of the domain experts, so that the acquired domain expert portraits are not accurate enough due to the lack of personal information, and the user experience is poor. Or the dimension of the field expert portrait cannot cover all aspects, and the technical field features related to the field expert portrait cannot be comprehensively described in a multi-dimensional manner; meanwhile, different subjects of industry-university-research science and technology projects are lack of data support of key core technical fields. An enterprise often encounters technical difficulties in research and development of new products or establishment of industry-university-research institutions and needs to seek field expert collaboration, and the industry-university-research institutions cannot accurately recognize field experts due to the fact that data information (such as internet information) or the information amount of a total data library is large. And the processing efficiency of searching specialists in specific technical fields by industry-university-research is greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of big data information processing technology, and in particular to a method for identifying experts and industry-academia-research institutions in the field of tackling key core technologies. Background Technology

[0003] High complexity is a common characteristic of my country's key core technologies. The industry-university-research collaboration for innovation in key core technologies is also a complex systems engineering project and a typical nonlinear complex system management activity. It should not only be studied from the perspective of systems science, but also fully consider the complexity characteristics of the industry-university-research collaboration network structure of key core technologies.

[0004] However, with the increasing complexity and integration of technologies, industry-academia-research cooperation in China faces problems such as regional administrative barriers and institutional barriers. Currently, the technology transfer rate in my country is less than one-fifth, and the industrialization rate is less than 5%. The phenomenon of technology transfer without commercialization is quite prominent, and the problem of technology and economy being "two separate entities" is serious. The reasons for this are twofold. Domain experts are individuals with outstanding performance and high professional competence in specific fields. On the one hand, domain experts can play an important role in the development of industry-academia-research projects, and many project reviews require their participation, such as bidding. However, current domain expert databases provide limited and insufficiently detailed data, either relying on manual collection or expert self-registration. This lack of personal information easily leads to inaccurate expert profiles or profiles that fail to cover all aspects, failing to comprehensively describe the technical characteristics of the experts involved. On the other hand, there is a lack of data support in key core technology areas among different stakeholders in industry-academia-research projects. Enterprises often encounter technical difficulties when developing new products or establishing industry-academia-research institutions, and need to seek collaboration from experts in the field. However, due to the massive amount of data (such as Internet information) or database information, industry-academia-research institutions cannot accurately identify experts in the field, which greatly reduces the efficiency of finding experts in specific technical fields. Summary of the Invention

[0005] This invention provides a method for identifying experts and industry-academia-research institutions in the field of tackling key core technologies, in order to overcome the above-mentioned technical problems.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] A method for identifying experts and industry-academia-research institutions in the field of tackling key core technologies includes the following steps:

[0008] S1: Obtain a thesaurus of scientific literature in key core technology fields;

[0009] S2: Based on the aforementioned scientific research literature thesaurus, construct a domain expert knowledge network oriented towards key core technologies to obtain domain expert data for tackling key core technologies;

[0010] S3: Construct an identification network for industry-university-research cooperation institutions based on key core technology areas to select data on potential industry-university-research cooperation institutions;

[0011] S4: Obtain the matching similarity between domain expert data and the data of the industry-academia-research institutions to be cooperated with, and confirm the domain experts required by the industry-academia-research institutions to be cooperated with based on the matching similarity.

[0012] Furthermore, S2 specifically includes the following steps;

[0013] S21: Filter the data in the acquired scientific literature thesaurus, and obtain key information of the literature based on the literature in the filtered scientific research thesaurus;

[0014] The screening includes removing non-journal articles and documents lacking key information;

[0015] The non-journal paper data includes conference notices, calls for papers, and topic selection guidelines;

[0016] The key information includes at least the author of the document, the title of the document, and keywords;

[0017] S22: Construct a domain expert knowledge subnetwork based on the key information in the aforementioned documents;

[0018] The domain expert knowledge subnetwork includes the author collaboration subnetwork V. A Document Sub-network V L Sub-network V in the core technology field D and patent sub-network V P ;

[0019] S23: Obtain a domain expert knowledge network oriented towards key core technologies based on the domain expert knowledge sub-network;

[0020] The expression for the domain expert knowledge network is:

[0021] KVN={V A V L V D V P SE A-L-D-P}

[0022]

[0023] In the formula: KVN represents a domain expert knowledge network; SE A-L-D-P This indicates that the subnetwork V represents the author collaboration relationship.A Document Sub-network V L Sub-network V in the core technology field D and patent sub-network V P Defined hyperedge; A i L represents the i-th author node; j D represents the j-th document node; k P represents the k-th core technology node; l Indicates the l-th existing patent node; α(A i ,L j ) = 1 indicates that the author node A i and document node L j There is a correlation between them; α(A) i D k ) = 1 indicates that the author node A i and Node D in the core technology field k There is a correlation between them; α(A) i ,P l ) = 1 indicates that the author node A i and patent node P l There is a correlation between them; α(L) j D k ) = 1 indicates that the document node L j and Node D in the core technology field k There is a correlation between them; α(D) k ,P l ) = 1 indicates that node D is in the core technology field. k and patent node P l There is a relationship between them;

[0024] S24: Based on the domain expert knowledge network, construct a domain expert recommendation strategy to obtain domain expert data for tackling key core technologies.

[0025] Furthermore, the domain expert recommendation strategy constructed in S24 is specifically as follows:

[0026] S241: Define the author collaboration subnetwork V A The main attributes of the nodes that participate in forming the hyperedges are used to obtain the number of hyperedges in the domain expert knowledge network;

[0027] The node's main attributes include node degree SD and hyperedge connectivity L(SE). i );

[0028] S242: Obtain the document subnetwork V based on the number of hyperedges of the domain expert knowledge network. L Influence of Chinese literature nodes;

[0029] The formula for calculating the influence of the document node is as follows:

[0030]

[0031] In the formula: I(l) i ) represents the influence degree of the document node; F(l i ) indicates document node l i The total number of hyperedges composed of cited literature in the domain expert knowledge hypernetwork model; A(l i ) represents the relationship between the author subnetwork and the document node l i The total number of authors whose cited references together form the hyperedge; N represents the total number of hyperedges in the hypernetwork model; N a This represents the total number of author nodes in the author subnetwork;

[0032] S243: Obtain the core technology sub-network V based on the number of hyperedges of the domain expert knowledge network. D Similarity of nodes in the domain;

[0033] The formula for calculating the similarity of the domain nodes is as follows:

[0034]

[0035] Where: Sim didj V represents the sub-network in the core technology field. D Middle domain node d i With domain node d j Similarity; Let P and Q represent the KL distance of the feature word probability distribution, and let P and Q represent the sub-network V in the core technology field, respectively. D All terms in the domain are defined by the domain node d. i With domain node d j The events that occur in the topic distribution; P(i) represents the domain node d i Feature word probability distribution; Q(j) represents the domain node d j The feature word probability distribution;

[0036] S244: Obtain the patent subnetwork V based on the number of hyperedges of the domain expert knowledge network. P The impact of patent nodes in China;

[0037] The formula for calculating the influence of the patent node is as follows:

[0038]

[0039] In the formula: K(p) i ) represents the patent node p i Influence degree; R(p) i ) represents the patent node p iThe total number of hyperedges composed of cited patents in the domain expert knowledge hypernetwork model; A(p i ) represents the author's subnetwork and the patent node p i The cited patents together form the total number of author nodes in the hyperedge; N represents the total number of hyperedges in the domain expert knowledge hypernetwork model; N a This represents the total number of author nodes in the author subnetwork;

[0040] S245: Based on the influence degree of the document nodes, the similarity of the domain nodes, and the influence degree of the patent nodes, construct a hyperedge scoring model for the domain expert knowledge network of key core technologies to obtain the hyperedge scoring value of the domain experts of key core technologies.

[0041] The expression for the hyperedge scoring model is:

[0042]

[0043]

[0044] In the formula: SER(SE) i ) represents the hyperedge score of the domain expert knowledge network; N represents the total number of hyperedges in the domain expert knowledge network; I li Indicates the influence degree of document nodes; Sim didj Represents the domain node d i With domain node d i Similarity between them; K pi Indicates the influence of the patent node; β represents the set weight parameter; L(SE) i ) represents the superedge SE i Hyperedge connectivity; SER(SE) ai ) represents the author node a i The score of the hyperedges formed; SD ai Author node a i The node superdegree value; S(a) represents the superedge score value of author node a;

[0045] S246: Based on the hyperedge score of the domain experts in key core technologies, sort the author nodes in the domain expert knowledge network in descending order to obtain the domain expert recommendation order list;

[0046] Based on the aforementioned domain expert recommendation order list, domain expert data for tackling key core technologies is obtained.

[0047] Furthermore, S3 specifically includes the following steps:

[0048] S31: Identify the industry-academia-research characteristic variables in key core technology areas;

[0049] The industry-academia-research characteristic variables include technology-theme-related characteristic variables, institutional competitiveness characteristic variables, and geographical relationship characteristic variables;

[0050] S32: Based on social network analysis technology, the industry-academia-research characteristic variables are used as a set of actors in the social network to obtain a three-mode network of institution-keyword-region;

[0051] Furthermore, a two-mode network is obtained based on the three-mode network of the institution-keyword-region, in order to expand the multi-mode matrix of industry-academia-research collaboration in key core technology fields;

[0052] The dual-mode network includes at least a dual-mode network of institutional-key core technology themes;

[0053] S33: Using Natural Language Processing (NLP) technology, extract key core technology keywords corresponding to each industry, academia, and research institution based on the unstructured text of scientific and technological literature published by various industry, academia, and research institutions in the key core technology field.

[0054] Based on the LDA topic model, the key core technology keywords are clustered to obtain several industry-academia-research institution clusters related to the technology topics;

[0055] S34: Determine the similarity index between cluster members in each industry-academia-research institution cluster, and establish the core edge network structure between institutions based on the similarity index;

[0056] The similarity metrics include one or more combinations of Jaccard similarity coefficient, cosine similarity, and Euclidean distance;

[0057] The rules for establishing the core edge network structure are as follows:

[0058] The cluster members in the cluster of industry-academia-research institutions are used as network nodes, and the similarity index between each cluster member is used as the edge weight between network nodes.

[0059] Furthermore, each cluster of the aforementioned industry-academia-research institutions corresponds to a core-edge pair;

[0060] The core-edge pair includes core cluster members and edge cluster members, and the core cluster members are cluster members in the industry-academia-research institution cluster whose similarity index is greater than a preset similarity threshold.

[0061] S35: Using Ucinet software, obtain the identification network of industry-university-research cooperation institutions based on the core edge network structure between institutions, and obtain the core degree of each network node in the identification network of industry-university-research cooperation institutions;

[0062] The network nodes are sorted in descending order according to their coreness to obtain a recommended sequence list of network nodes. Based on the recommended sequence list of network nodes, data from industry-academia-research institutions to be cooperated with are selected.

[0063] Furthermore, S4 specifically includes the following steps:

[0064] S41: The acquired domain expert data and the data of the industry-academia-research institutions to be cooperated with are used as feature nodes. The feature nodes obtain the directed network of each node according to the preset link path.

[0065] The preset link paths include domain expert-domain expert link paths, industry-academia-research institution-domain expert link paths, and domain expert-industry-academia-research institution link paths;

[0066] The directed network of nodes includes a domain expert-domain expert directed network, an industry-academia-research institution-domain expert directed network, and a domain expert-industry-academia-research institution directed network;

[0067] S42: Define the target node vector and the neighbor node vectors of the target node according to the directed network of each node;

[0068] The target node vector is used to represent the cooperative relationship between any two nodes in a directed network;

[0069] The neighbor node vector is used to represent the feature information of domain experts or industry-academia-research institutions in the directed network of each node.

[0070] Furthermore, the characteristic information of the domain expert includes at least the total number of documents, the number of citations of the documents, the total number of patents, and the number of citations of the patents.

[0071] The characteristic information of the industry-academia-research institutions includes at least their operating status, enterprise size, and patent achievements;

[0072] Based on a graph convolutional neural network, a heterogeneous graph network is obtained according to the target node vector and the neighbor node vectors, and the heterogeneous graph network is used as the input layer of the R-GCN model;

[0073] S43: Based on the element-wise dot product method, the node distance is obtained according to the target node vector and the neighbor node vectors, and the node distance is used as the predicted score of the cooperation relationship between domain experts and industry-academia-research institutions output by the R-GCN model;

[0074] Furthermore, the target node vector and neighbor node vectors input to the input layer are aggregated using an activation function to obtain the output features of the center node of the R-GCN model;

[0075] The expression for the output feature of the central node of the R-GCN model is as follows:

[0076]

[0077] In the formula: r∈{1,...R}=R represents the type of cooperation relationship; i∈{1,...N}=N represents the type of node; The result represents the l-th level node representation of node i; Represents the l-th level nodes of all neighboring nodes of node j; c represents the set of neighboring nodes of node i whose relation is r; i，r Represents a regularization constant; Represents a linear transformation function;

[0078] S44: Randomly initialize the network weight parameters of the R-GCN model, take the target node vector and neighbor node vector in the obtained heterogeneous graph network as input data, take the predicted score of the cooperation relationship between domain experts and industry-academia-research institutions as output data, train the R-GCN model to obtain the link prediction model between domain experts and industry-academia-research institutions.

[0079] S45: Select HingeLoss as the loss function of the R-GCN model, obtain the loss value of the output features of the center node of the R-GCN model and the output features of the center node of the R-GCN model obtained in the previous model training, and compare the loss value with the preset loss threshold.

[0080] If the loss value is greater than or equal to the preset loss threshold, the network weight parameters of the R-GCN model are adaptively updated, and S44 is executed repeatedly.

[0081] If the loss value is less than the preset loss threshold, then the trained R-GCN model is the link prediction model between domain experts and industry-academia-research institutions.

[0082] S46: Obtain the predicted score of the cooperation relationship between the industry, academia and research institutions based on the link prediction model of the domain experts and the industry, academia and research institutions, and use the predicted score of the cooperation relationship as the matching similarity between the domain expert data and the data of the industry, academia and research institutions to be cooperated with, and then confirm the domain experts selected by the industry, academia and research institutions to be cooperated with based on the matching similarity.

[0083] Beneficial Effects: This invention provides a method for identifying experts and industry-academia-research institutions in key core technology research. It constructs a domain expert knowledge network based on a scientific literature thesaurus to obtain data on domain experts in key core technology research. This knowledge network comprehensively considers scientific literature to provide accurate ranking and identification of domain experts, and offers multi-dimensional coverage of expert profiles, providing a more comprehensive description of the technical characteristics involved. Furthermore, the constructed network for identifying industry-academia-research collaboration institutions, by comprehensively considering technical correlation analysis, institutional competition analysis, core-edge analysis between institutions in the collaboration network, and institution type, can assist in identifying potential industry-academia-research collaboration partners in core technology areas. Finally, by obtaining the matching similarity between domain expert data and the data of potential collaboration institutions, the selected domain experts from the potential collaboration institutions are confirmed based on the matching similarity. This solution addresses the problem that the lack of personal information about experts in the technical field leads to inaccurate profiles of these experts, or that the profiles cannot cover all aspects and comprehensively describe the technical characteristics of the experts. As a result, it enables accurate identification of experts when establishing industry-academia-research projects, greatly improving the efficiency of finding experts in specific technical fields. Attached Figure Description

[0084] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0085] Figure 1 This is a flowchart illustrating the identification method for experts and industry-academia-research institutions in the field of tackling key core technologies, as described in this invention.

[0086] Figure 2 This is a schematic diagram of the expert knowledge hypernetwork in the key core technology field of this embodiment;

[0087] Figure 3 This is a flowchart illustrating the process of acquiring domain expert data for tackling key core technologies in this embodiment.

[0088] Figure 4 This is a schematic diagram of the identification process of the industry-academia-research cooperation institution identification network in this embodiment;

[0089] Figure 5 This is a flowchart illustrating the link prediction process between domain experts and industry-academia-research institutions in this embodiment. Detailed Implementation

[0090] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0091] This embodiment provides a method for identifying experts and industry-academia-research institutions in the field of tackling key core technologies, such as... Figure 1 As shown, it includes the following steps:

[0092] S1: Obtain a thesaurus of scientific literature in key core technology fields;

[0093] Specifically, regarding key core technology areas, although the state has issued overall plans, different industries and provinces / cities have subdivided them into different technology fields. For example, "Made in China 2025" involves ten key areas, while Guangzhou and Foshan have refined 66 key core technologies across eight major areas. Therefore, this invention first constructs a research dictionary for key core technology areas. Specifically, this embodiment obtains key core technology keyword data from the State Council website, government websites, provincial and municipal science and technology bureau websites, the National Science and Technology Report System, the "China High-Tech Industry Statistical Yearbook," the "China Science and Technology Statistical Yearbook," regional statistical yearbooks and science and technology statistical yearbooks, the official website of the National Bureau of Statistics, and official webpages of enterprises related to key core technology areas through targeted crawling or web scraping. The obtained keyword data is then stored in text format according to categories, serving as the original research dictionary data.

[0094] To ensure the accuracy of the research dictionary in key core technology fields and improve the efficiency and accuracy of subsequent data processing, it is necessary to denoise the keyword data in the original research dictionary. In this invention, firstly, new keyword data is found and added to the research dictionary by retrieving relevant literature, materials, and professional terms in the research field, thus expanding the keyword data. Secondly, existing keyword data is reorganized and categorized to better meet the needs of the research field, achieving keyword data reconstruction. Furthermore, constructing a research dictionary (research literature thesaurus) for key core technologies effectively improves the effectiveness and accuracy of recognition.

[0095] S2: Based on the aforementioned scientific research literature thesaurus, construct a domain expert knowledge network oriented towards key core technologies to obtain domain expert data for tackling key core technologies, specifically including the following steps;

[0096] S21: Filter the data in the acquired scientific literature thesaurus, and obtain key information of the literature based on the literature in the filtered scientific research thesaurus;

[0097] The screening includes removing non-journal articles and documents lacking key information;

[0098] The non-journal paper data includes conference notices, calls for papers, and topic selection guidelines;

[0099] The key information includes at least the author of the document, the title of the document, and keywords;

[0100] S22: Construct a domain expert knowledge subnetwork based on the key information in the aforementioned documents;

[0101] The domain expert knowledge subnetwork includes the author collaboration subnetwork V. A Document Sub-network V L Sub-network V in the core technology field D and patent sub-network V P ;

[0102] Specifically, such as Figure 2 As shown, based on the key technology core domain scientific research dictionary (scientific literature thesaurus) obtained by S1, and on the basis of the classic knowledge hypernetwork model (knowledge element, knowledge carrier, and knowledge subject), a "domain expert" knowledge hypernetwork model for tackling key core technologies is constructed by combining the main domain expert identification methods (co-authorship network, co-citation network, co-word network, topic clustering, etc.). Specifically, this model includes:

[0103] (1) Author subnetwork: In the field of key core technologies, the authors of the literature are the knowledge subjects in the knowledge network. Therefore, the authors are used as nodes and the co-authorship relationship between authors is used as the edge to connect the author nodes, thereby establishing the author subnetwork.

[0104] (2) Document sub-network: Becoming a domain expert indicates that the scholar has research results with a certain influence in the field. As a knowledge carrier, documents are the crystallization of the wisdom of the knowledge subject. Therefore, a document sub-network is constructed with documents as nodes and the citation relationship between documents as edges.

[0105] (3) Domain Subnetwork: The research domain is the basis for classifying experts, representing their knowledge and serving as their identification label. The nodes of the domain subnetwork in this invention must contain elements of the key core technology research dictionary constructed in step 1. Using the research domains of scholars in the key core technology research dictionary as nodes, the relationships between domains connect the domain nodes, thereby establishing the domain subnetwork.

[0106] (4) Patent sub-network: Due to their innovative characteristics, patents have become the most important evaluation basis in key core technology fields. Therefore, this invention uses cooperative patents (including invention patents, utility model patents, and design patents) as nodes and patent cooperation relationships as edges to construct an undirected weighted industry-university-research patent cooperation sub-network.

[0107] Therefore, for the identification method of experts in key core technology fields, this invention establishes a knowledge hypernetwork model consisting of four layers: author subnetwork, document subnetwork, key core technology field subnetwork, and patent subnetwork. Data sources include CNKI and Wanfang as document data sources, keyword search from the key core technology field thesaurus constructed in this invention, and exclusion of non-journal papers such as conference notices, call for papers, and topic guidelines, as well as papers with missing information. The collected data is preprocessed to extract authors, document titles, keywords, etc. An author subnetwork is constructed based on author collaboration relationships; a document subnetwork is constructed based on citation relationships between documents; and the author subnetwork and document subnetwork are linked based on the publication relationship between authors and documents. The topic-feature word distribution is obtained based on LDA topic models, and each research topic is manually identified to determine domain labels, establishing links between the document subnetwork and the domain subnetwork. The patent subnetwork uses patents jointly applied for and authorized by industry, academia, and research institutions from 2011 to 2023 as samples, with sample patent data sourced from the China Patent Information Center of the State Intellectual Property Office. Based on the triple helix theory, this paper categorizes industry-academia-research patent collaborations, using universities, research institutions, and enterprises as the three major innovation entities in the triple helix model. Based on the patent collaboration types derived from mutual cooperation among these three innovation entities, patent collaborations involving different types of entities in joint applicants (patent holders) are summarized into four collaboration models: "industry-research," "academia-research," "industry-academia," and "industry-academia-research." This allows for more precise extraction of effective collaboration networks from the innovation network, eliminating ineffective innovation collaborations. The search criteria are set as follows: four search criteria are selected: applicant, application date, abstract (key core technology dictionary), and applicant address. Each search field is set, and patents in the key core technology dictionary are searched sequentially and then aggregated to obtain collaborative patent applications. The specific establishment of the knowledge hypernetwork model for experts in key core technology fields is as follows:

[0108] The specific definitions are as follows: (1) Author subnetwork V A :

[0109] The author collaboration subnetwork V A Define author nodes and obtain the author cooperation subnetwork V based on the co-occurrence relationships of authors as undirected edges. A ;

[0110] The expression for the author collaboration subnetwork is:

[0111] V A =(A, E) A-A )

[0112] E A-A ={(A i A j (i,j=1,2,..n)

[0113] In the formula: A = {A1, A2, A3, ... A} n} represents a finite set of co-author nodes for a document; (A i A j ) represents V A The set of undirected edges in the middle, and (A i A j ) = 1 indicates that A i With A j There is a cooperative relationship; (A) i A j ) = 0 means A i With A j There is no cooperative relationship;

[0114] (2) The document subnetwork V L Define document nodes and use the citation relationships in different documents as undirected edges to obtain the document subnetwork V. L ;

[0115] The document subnetwork V L The expression is

[0116] V L = (L, E L→L )

[0117] E L→L ={(L i L j (i,j=1,2,..n)

[0118] In the formula: L = {L1, L2, L3, ... L} n} represents a finite set of documents, which is a document subnetwork V for "domain experts" focusing on tackling key core technologies. L The set of midpoints; (L i ,L j ) represents V L The set of edges in the network and if (L i ,L j ) represents V L The set of undirected edges in the middle, and (L i ,L j ) = 1 indicates L i With L j There is a reference relationship; (L)i ,L j ) = 0 indicates L i With L j There is no reference relationship;

[0119] (3) Sub-network V in key core technology fields D :

[0120] Definition: Nodes represent the main research areas of key core technologies. Undirected edges are constructed based on the relationships between these areas to create a research area network:

[0121] V D =(D, E) D-D )

[0122] E D-D ={(D i D j )}(i,j=1,2,..n),

[0123] In the formula: D={D1,D2,D3,…D n} represents a subset of the key technology core domain research dictionary in S1, which is a domain subnetwork V for key core technologies oriented towards "domain experts" in key core technology breakthroughs. D The set of nodes in the middle; (D i D j ) is V D The set of undirected edges in a network; and (D i D j ) = 1 indicates D i With D j Existence relation; (D) i D j ) = 0 indicates that D i With D j There is no relationship;

[0124] (4) Patent Subnetwork V P :

[0125] Definition: Using the patent application subject as the node and the patent cooperation relationship between authors (inventors) of different institutions as undirected edges, an undirected weighted sub-network of industry-academia-research patent cooperation is constructed.

[0126] V P =(P, E) P-P )

[0127] E P-P ={(P i P j )}(i,j=1,2,..n),

[0128] In the formula: P = {P1, P2, P3, ... P} n} represents a finite set of patents, which is a patent subnetwork V for "domain experts" focusing on tackling key core technologies. P The set of midpoints; (P) i ,P j ) is V P The set of undirected edges in a network;

[0129] And (P) i ,P j ) = 1 indicates p i With P j Existence relationship; (P) i ,P j ) = 0 indicates that p i With P j There is no relationship;

[0130] S23: Obtain a domain expert knowledge network oriented towards key core technologies based on the domain expert knowledge sub-network;

[0131] The expression for the domain expert knowledge network is:

[0132] KVN={V A V L V D V P SE A-L-D-P}

[0133]

[0134] In the formula: KVN represents a domain expert knowledge network; SE A-L-D-P This indicates that the subnetwork V represents the author collaboration relationship. A Document Sub-network V L Sub-network V in the core technology field D and patent sub-network V P Defined hyperedge; A i L represents the i-th author node; j D represents the j-th document node; k P represents the k-th core technology node; l Indicates the l-th existing patent node; α(A i ,L j ) = 1 indicates that the author node A i and document node L j There is a correlation between them; α(A) i D k ) = 1 indicates that the author node A i and Node D in the core technology field k There is a correlation between them; α(A) i ,P l ) = 1 indicates that the author node Ai and patent node P l There is a correlation between them; α(L) j D k ) = 1 indicates that the document node L j and Node D in the core technology field k There is a correlation between them; α(D) k ,P l ) = 1 indicates that node D is in the core technology field. k and patent node P l There is a relationship between them;

[0135] S24: Based on the domain expert knowledge network, construct a domain expert recommendation strategy to obtain domain expert data for tackling key core technologies;

[0136] This embodiment identifies experts in a specific field within the context of industry-academia-research collaborative innovation. This requires comprehensively considering the expert's academic influence, the quantity and quality of published literature, patent innovation level, and semantically identifying the expert's research expertise. Drawing on the ideas and principles of the PageRank algorithm, this invention proposes a hyperedge ranking algorithm based on a knowledge expert hypernetwork model in key core technology fields. This algorithm comprehensively considers factors such as the influence of literature contained in different hyperedges, domain similarity, and patent citations. (The calculation of author a...) i Calculate the average hyperedge score S(a) for each author based on the scores of all hyperedges involved in their formation. i The scholars ranked are considered experts in the field.

[0137] Specifically, such as Figure 3 As shown, the domain expert recommendation strategy is specifically as follows:

[0138] S241: Define the author collaboration subnetwork V A The main attributes of the nodes that participate in forming the hyperedges are used to obtain the number of hyperedges in the domain expert knowledge network;

[0139] The node's main attributes include node degree SD and hyperedge connectivity L(SE). i );

[0140] Among them, (1) Node super-degree SD: The node super-degree in the hypernetwork is similar to the node connectivity in the social network analysis method. The node super-degree SD of a certain node is defined as the number of hyperedges that the node participates in forming.

[0141] (2) Hyperedge connectivity L(SE) i In a hypernet, each hyperedge SEi contains multiple nodes. If two hyperedges contain the same nodes, then the two hyperedges are connected through the common nodes. The hyperedge connectivity L(SEi) is the degree of connectivity between the two hyperedges. iThis represents the number of other superedges that share nodes with this superedge.

[0142] S242: Obtain the document subnetwork V based on the number of hyperedges of the domain expert knowledge network. L Influence of Chinese literature nodes; in the literature subnetwork V L In this context, the higher the influence of the document nodes contained in a certain hyperedge, the easier it is for that hyperedge to be "discovered," that is, the greater the probability of it being linked by other hyperedges, and the greater the probability of it becoming a recommendation expert.

[0143] The formula for calculating the influence of the document node is as follows:

[0144]

[0145] In the formula: I(l) i ) represents the influence degree of the document node; F(l i ) indicates document node l i The total number of hyperedges composed of cited literature in the domain expert knowledge hypernetwork model; A(l i ) represents the relationship between the author subnetwork and the document node l i The total number of authors whose cited references together form the hyperedge; N represents the total number of hyperedges in the hypernetwork model; N a This represents the total number of author nodes in the author subnetwork;

[0146] S243: Obtain the core technology sub-network V based on the number of hyperedges of the domain expert knowledge network. D The similarity of nodes within a domain is considered. In a domain subnetwork, the greater the similarity between the domain containing a hyperedge and other domains, the higher the probability of becoming a recommendation expert. In the domain subnetwork, LDA topic model is used for domain label extraction. The topic-feature word probability distribution is obtained from the LDA topic model, and domain labels are extracted from it. Then, the similarity between domain label nodes is calculated based on the feature word distribution probability under each domain. The KL distance, commonly used in statistical natural language processing, is used to calculate the similarity of domain labels. A larger KL distance indicates lower similarity between domain labels; therefore, a domain label semantic similarity metric, Sim, is defined. didj Represents the domain node d i and d j To assess the similarity between KL distance formulas, and considering the asymmetry of the KL distance formula, JS distance is used to address the asymmetry issue of KL distance.

[0147] The formula for calculating the similarity of the domain nodes is as follows:

[0148]

[0149] Where: Sim didj V represents the sub-network in the core technology field. DMiddle domain node d i With domain node d j Similarity; Let P and Q represent the KL distance of the feature word probability distribution, and let P and Q represent the sub-network V in the core technology field, respectively. D All terms in the domain are defined by the domain node d. i With domain node d j The events that occur in the topic distribution; P(i) represents the domain node d i Feature word probability distribution; Q(j) represents the domain node d j The feature word probability distribution;

[0150] S244: Obtain the patent subnetwork V based on the number of hyperedges of the domain expert knowledge network. P The influence of patent nodes in the patent subnetwork; and the more edges a certain hyperedge has with other hyperedges in the patent subnetwork, the greater the probability of becoming a recommendation expert;

[0151] The formula for calculating the influence of the patent node is as follows:

[0152]

[0153] In the formula: K(p) i ) represents the patent node p i Influence degree; R(p) i ) represents the patent node p i The total number of hyperedges composed of cited patents in the domain expert knowledge hypernetwork model; A(p i ) represents the author's subnetwork and the patent node p i The cited patents together form the total number of author nodes in the hyperedge; N represents the total number of hyperedges in the domain expert knowledge hypernetwork model; N a This represents the total number of author nodes in the author subnetwork;

[0154] S245: Based on the influence degree of the document nodes, the similarity of the domain nodes, and the influence degree of the patent nodes, construct a hyperedge scoring model for the domain expert knowledge network of key core technologies to obtain the hyperedge scoring value of the domain experts of key core technologies.

[0155] The expression for the hyperedge scoring model is:

[0156]

[0157]

[0158] In the formula: SER(SE) i ) represents the hyperedge score of the domain expert knowledge network; N represents the total number of hyperedges in the domain expert knowledge network; I li Indicates the influence degree of document nodes; Simdidj Represents the domain node d i With domain node d i Similarity between them; K pi Indicates the influence of the patent node; β represents the set weight parameter; L(SE) i ) represents the superedge SE i Hyperedge connectivity; SER(SE) ai ) represents the author node a i The score of the hyperedges formed; SD ai Author node a i The node superdegree value; S(a) represents the superedge score value of author node a;

[0159] S246: Based on the hyperedge score of the domain experts in key core technologies, sort the author nodes in the domain expert knowledge network in descending order to obtain the domain expert recommendation order list;

[0160] Based on the aforementioned domain expert recommendation order list, domain expert data for tackling key core technologies is obtained.

[0161] S3: Construct an identification network for industry-academia-research collaboration institutions based on key core technology areas to select data on potential collaboration partners, such as... Figure 4 As shown, the specific steps include:

[0162] S31: Identify the industry-academia-research characteristic variables in key core technology areas;

[0163] The industry-academia-research characteristic variables include technology-theme-related characteristic variables, institutional competitiveness characteristic variables, and geographical relationship characteristic variables;

[0164] S32: Based on social network analysis technology, the industry-academia-research characteristic variables are used as a set of actors in the social network to obtain a three-mode network of institution-keyword-region;

[0165] Furthermore, a two-mode network is obtained based on the three-mode network of the institution-keyword-region, in order to expand the multi-mode matrix of industry-academia-research collaboration in key core technology fields;

[0166] The dual-mode network includes at least a dual-mode network of institutional-key core technology themes;

[0167] Specifically, industry-academia-research collaboration in key core technology areas is a complex social behavior. Motivation for collaboration arises only when the needs and goals of two or more institutions align to some extent. Therefore, this invention, in identifying industry-academia-research collaborative institutions, utilizes extended multi-modal data analysis. It selects technological thematic relevance, institutional competitiveness, and geographical relationships as analytical variables, and conducts extended multi-modal data analysis to identify potential collaborative institutions in the field. Specifically, this includes...

[0168] The selection of multi-modal variables for "industry-academia-research collaboration in key core technology fields": If the technological themes of institutions are highly related and complementary, their competitiveness is low, and their geographical locations are close, then the institutions will have the greatest potential for cooperation. Therefore, the variables selected are: technological theme correlation, institutional competitiveness, and geographical relationship as feature variables; specifically, the technological theme correlation feature variables include technological similarity and technological complementarity. Institutions with complementarity in the same technological field will generate greater benefits from cooperation. Since the IPC classification number is the only internationally recognized method for classifying patent documents, it is marked by patent examiners based on the content of the patent text, and has objectivity and authority. Therefore, a combination of patent IPC classifications is used to calculate technological similarity. By calculating the Jaccard similarity coefficient, cosine similarity, and Euclidean distance of the patent IPC fields involved by the enterprise and each inventor, and comprehensively considering set similarity, relative distance, and absolute distance, these are used as features for predicting cooperation. These three metrics are widely used to measure similarity. Jaccard similarity considers set similarity and judges the consistency of common features; cosine similarity reflects the relative distance between two entities, emphasizing the vector direction relationship; and Euclidean distance expresses the absolute distance between two entities, influenced by the frequency of occurrence of elements in the set. Text analysis was conducted on scientific and technological literature produced by institutions in the "key core technology field" (industry-academia-research collaboration). Thematic terms related to industry-academia-research collaboration and technology were extracted from unstructured text, and the extracted institutions underwent data cleansing and deep cleaning. Text analysis techniques were then applied to the cleaned dataset to generate a three-mode network of "region-institution-technology theme." Next, theme clustering and technology association analysis were performed on the three-mode network. Complex network community identification technology was used to cluster the two-mode network of institution-technology theme, thereby obtaining the similarity and technology association of technical themes among institutions in the "key core technology field" (industry-academia-research collaboration).

[0169] Institutional competitiveness characteristics: The magnitude of competitiveness is an important consideration for institutional cooperation. When selecting scientific and technological cooperation partners, institutions often consider the comprehensive strength and competitive advantages of the other institutions. This embodiment analyzes the competitive position among institutions through the core-periphery structure of the cooperation network.

[0170] Geographical relationship characteristic variable: This embodiment can identify potential cooperative relationships within a province to a certain extent by identifying the number of identical technical topics between institutions within the same province or between adjacent provinces.

[0171] Expanding the Multi-Modal Matrix of Industry-University-Research Collaboration in Key Core Technology Fields: To better identify potential industry-university-research collaboration institutions in key core technology fields, an expanded multi-modal matrix is ​​formed by supplementing the traditional three-modal network of "institution-keyword-region" (in social network analysis, a module refers to the set of actors, which is the number of types of social actor sets in the network. A network of relationships between one type of actor set and another type of actor set is called a two-modal network, and a network of relationships between three types of actor sets is called a three-modal network). This is based on the characteristics of industry-university-research collaboration networks (the characteristics of industry-university-research collaboration networks for key core technology breakthroughs refer to the collaborative relationships established between enterprises, universities, and research institutions for the purpose of breakthroughs in core technologies, achieving mutual promotion and common development through cooperation and sharing of resources, technology, and knowledge). Specifically, the "institution-institution co-occurrence network" can show whether there has been previous cooperation between institutions and the intensity of such cooperation, helping to identify the possibility of potential cooperation. The "Key Core Technology Theme Keywords - Key Core Technology Theme Keywords Co-occurrence Network" can identify the relationships between technology themes and assist in finding institutions with technological connections. "Regional Co-occurrence" comes from institutional data co-occurrence statistics, which are not very meaningful here; therefore, the "Region-Region Co-occurrence" module is set to module zero. The extended multimodal co-occurrence network is only a data processing method; in actual analysis, the meaning and data characteristics of specific modules vary considerably, and module settings should be flexible.

[0172] S33: Using Natural Language Processing (NLP) technology, extract key core technology keywords corresponding to each industry, academia, and research institution based on the unstructured text of scientific and technological literature published by various industry, academia, and research institutions in the key core technology field.

[0173] Based on the LDA topic model, the key core technology keywords are clustered to obtain several industry-academia-research institution clusters related to the technology topics;

[0174] Specifically, this embodiment examines patent document output and collaborative applications to identify potential industry-academia-research collaboration partners in key core technology fields. By analyzing the technical strengths, competitive positions among institutions, and geographical locations of these partners, potential industry-academia-research collaboration partners are discovered.

[0175] This embodiment performs text analysis on scientific and technological literature produced by institutions in the "key core technology field of industry-academia-research collaboration". Natural Language Processing (NLP) is used to extract industry-academia-research structure and key core technology keywords from unstructured text information, and the extracted institutions' data is cleaned and purified. Text analysis techniques are then applied again to the cleaned dataset to generate a "region-institution-technology theme" three-mode network. Next, theme clustering and technology association analysis are performed on the three-mode network (this invention uses Latent Dirichlet Allocation (LDA), a generative probabilistic model, to discover potential themes from the data and achieve theme clustering analysis; association rule mining technology is used to reveal the association relationships between different items by discovering association rules in the dataset). Complex network community identification technology is used to cluster the institution-technology theme network, thereby obtaining the similarity and technology association of technology themes among institutions in the "key core technology field of industry-academia-research collaboration".

[0176] S34: Determine the similarity index between cluster members in each industry-academia-research institution cluster, and establish the core edge network structure between institutions based on the similarity index;

[0177] The similarity metrics include one or more combinations of Jaccard similarity coefficient, cosine similarity, and Euclidean distance;

[0178] The rules for establishing the core edge network structure are as follows:

[0179] The cluster members in the cluster of industry-academia-research institutions are used as network nodes, and the similarity index between each cluster member is used as the edge weight between network nodes.

[0180] Furthermore, each cluster of the aforementioned industry-academia-research institutions corresponds to a core-edge pair;

[0181] The core-edge pair includes core cluster members and edge cluster members, and the core cluster members are cluster members in the industry-academia-research institution cluster whose similarity index is greater than a preset similarity threshold.

[0182] Specifically, the competitiveness of institutions is a crucial factor in institutional cooperation. When selecting partners for scientific and technological collaboration, institutions often consider the comprehensive strength and competitive advantages of the other party. This embodiment analyzes the competitive positions of institutions through the core-periphery structure in a cooperation network. A core-periphery structure is a special structure composed of several interconnected elements, characterized by a tightly linked center and a sparsely distributed periphery. The purpose of core-periphery structure analysis is to study which nodes in a social network occupy core positions and which occupy peripheral positions. In this embodiment, it is used to identify the competitiveness between institutions; companies in the middle structure possess stronger competitiveness.

[0183] S35: Based on Ucinet software as a network analysis tool, obtain the identification network of industry-university-research cooperation institutions according to the core edge network structure between institutions, and obtain the core degree of each network node in the identification network of industry-university-research cooperation institutions.

[0184] The network nodes are sorted in descending order according to their coreness to obtain a recommended sequence list of network nodes. Based on the recommended sequence list of network nodes, data from industry-academia-research institutions to be cooperated with are selected.

[0185] Specifically, Ucinet software was selected as the network analysis tool to construct a collaborative network for industry-academia-research collaboration in key core technology areas. Ucinet was used to calculate the coreity of each node (institution). The coreity reflects the position and influence of each node within the entire network. The higher the coreity of a node, the higher its strategic position within the network, indicating a more significant and active role in information dissemination, resource sharing, and knowledge diffusion within the network, and a greater likelihood of collaboration. Finally, recommendations for institutional collaboration were derived based on the Ucinet coreity calculation results.

[0186] The formula for calculating the coreness of the node (mechanism) is as follows:

[0187]

[0188] In the formula: C i C represents the centrality of node i; max The centrality of node i represents the maximum value of the centrality, reflecting the tendency of the entire network to concentrate on one or a few points. The larger the degree centrality, the stronger the central tendency of the network.

[0189] S4: Obtain the matching similarity between domain expert data and the data of the industry-academia-research institutions to be collaborated with, and confirm the domain experts required by the industry-academia-research institutions to be collaborated with based on the matching similarity, such as... Figure 5 As shown, the specific steps include:

[0190] S41: The acquired domain expert data and the data of the industry-academia-research institutions to be cooperated with are used as feature nodes. The feature nodes obtain the directed network of each node according to the preset link path.

[0191] The preset link paths include domain expert-domain expert link paths, industry-academia-research institution-domain expert link paths, and domain expert-industry-academia-research institution link paths;

[0192] The directed network of nodes includes a domain expert-domain expert directed network, an industry-academia-research institution-domain expert directed network, and a domain expert-industry-academia-research institution directed network;

[0193] Specifically, 1) Based on the expert knowledge hypernetwork model (domain expert knowledge network) in the key core technology field in S2, the feature vector of “domain expert” is constructed, including: the ranking of domain experts in the key core technology breakthrough data and the total number of documents, the number of document citations, the total number of patents, and the number of patent citations of the corresponding domain experts are used as the features of “domain expert” nodes to construct a “domain expert-domain expert” cooperation network.

[0194] 2) In addition to using the ranking of recommended industry-university-research collaborative institutions in the "key core technology fields" obtained in S3 as one dimension of the feature vector, the characteristics of "industry-university-research" collaborative institutions also include business status, enterprise size, and patent achievements as feature vectors (the data source for business status and enterprise size is mainly from the annual reports published by enterprises) to construct node features. At the same time, based on the data source of the patent sub-network, the "industry-university-research institution-domain expert" collaborative network is obtained from the patent collaborative achievements of "industry-university-research institutions" and "domain experts".

[0195] 3) The single-layer collaboration network of "industry-academia-research institutions - domain experts" in 2) includes nodes of enterprises and scholars who have actually collaborated, but does not include scholars who have never collaborated with enterprises. However, after joining the "domain expert" collaboration network in 1), all scholars who have collaborated with each other are included in the network. The main objective of this invention is to add scholars who have not collaborated with enterprises to the list of potential collaborators. Therefore, the overall sample of "industry-academia-research institutions - domain experts" collaboration is a fully connected layer of enterprises and scholars, meaning that every enterprise is connected to all scholars, thus obtaining the possibility of collaboration.

[0196] 4) In the construction of the R-GCN model, since R-GCN is applicable to directed networks, the construction of the "domain expert-domain expert" cooperation network for "key core technologies" in step 1) and the "industry-university-research institution-domain expert" cooperation network for "key core technology industry-university-research" in step 2) are further divided into 3 layers, namely "domain expert cooperation network", "industry-university-research institution-domain expert" directed cooperation network, and "domain expert-industry-university-research institution" directed cooperation network;

[0197] S42: Define the target node vector and the neighbor node vectors of the target node according to the directed network of each node;

[0198] The target node vector is used to represent the cooperative relationship between any two nodes in a directed network;

[0199] The neighbor node vector is used to represent the feature information of domain experts or industry-academia-research institutions in the directed network of each node.

[0200] Furthermore, the characteristic information of the domain expert includes at least the total number of documents, the number of citations of the documents, the total number of patents, and the number of citations of the patents.

[0201] The characteristic information of the industry-academia-research institutions includes at least their operating status, enterprise size, and patent achievements;

[0202] Based on a graph convolutional neural network, a heterogeneous graph network is obtained according to the target node vector and the neighbor node vectors, and the heterogeneous graph network is used as the input layer of the R-GCN model;

[0203] Specifically, the GCN convolution module of the R-GCN model is used to generate single-layer node vector representations. After completion, the three-layer network (domain expert-domain expert directed network, industry-academia-research institution directed network, and domain expert-industry-academia-research institution directed network) is stacked to obtain a multi-layer heterogeneous graph network to generate node vector representations. The input layer of the R-GCN model is formed according to the pattern of nodes and edges that are linked in the obtained heterogeneous graph network.

[0204] S43: Based on the element-wise dot product method, the node distance is obtained according to the target node vector and the neighbor node vectors, and the node distance is used as the predicted score of the cooperation relationship between domain experts and industry-academia-research institutions output by the R-GCN model;

[0205] Furthermore, the target node vector and neighbor node vectors input to the input layer are aggregated using an activation function to obtain the output features of the center node of the R-GCN model;

[0206] The expression for the output feature of the central node of the R-GCN model is as follows:

[0207]

[0208] In the formula: r∈{1,...R}=R represents the type of cooperation relationship; i∈{1,...N}=N represents the type of node; The result represents the l-th level node representation of node i; Represents the l-th level nodes of all neighboring nodes of node j; c represents the set of neighboring nodes of node i whose relation is r; i，r Represents a regularization constant; Represents a linear transformation function;

[0209] S44: Randomly initialize the network weight parameters of the R-GCN model, take the target node vector and neighbor node vector in the obtained heterogeneous graph network as input data, take the predicted score of the cooperation relationship between domain experts and industry-academia-research institutions as output data, train the R-GCN model to obtain the link prediction model between domain experts and industry-academia-research institutions.

[0210] In this embodiment, since it is a binary classification problem of whether or not there is a link between industry, academia and research institutions and domain experts, the number of neurons in the output layer of the R-GCN model is set to 2.

[0211] S45: Select HingeLoss as the loss function of the R-GCN model, obtain the loss value of the output features of the center node of the R-GCN model and the output features of the center node of the R-GCN model obtained in the previous model training, and compare the loss value with the preset loss threshold.

[0212] The preset loss threshold is specifically 0.0002;

[0213] If the loss value is greater than or equal to the preset loss threshold, the network weight parameters of the R-GCN model are adaptively updated, and S44 is executed repeatedly.

[0214] If the loss value is less than the preset loss threshold, then the trained R-GCN model is the link prediction model between domain experts and industry-academia-research institutions.

[0215] S46: Obtain the predicted score of the cooperation relationship between the industry, academia and research institutions based on the link prediction model of the domain experts and the industry, academia and research institutions, and use the predicted score of the cooperation relationship as the matching similarity between the domain expert data and the data of the industry, academia and research institutions to be cooperated with, and then confirm the domain experts selected by the industry, academia and research institutions to be cooperated with based on the matching similarity.

[0216] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0217] 1. To construct a domain expert knowledge network oriented towards key core technologies, this embodiment establishes a knowledge hypernetwork model consisting of four layers: author subnetwork, literature subnetwork, key core technology subnetwork, and patent subnetwork. Combining hypernetwork measurement methods and bibliometrics, it applies the LDA topic model and draws on the PageRank ranking approach. The domain expert identification method based on the knowledge hypernetwork can comprehensively consider multiple factors to provide accurate ranking and identification of "domain experts".

[0218] 2. To construct an identification network for industry-academia-research collaboration institutions based on key core technology areas, by extending multimodal data analysis and comprehensively considering the technology correlation analysis, institutional competition analysis, core-edge analysis between institutions in the collaboration network, and institution type, it can help identify potential industry-academia-research collaboration partners in the field.

[0219] 3. The R-GCN algorithm is used to train the model to obtain a link prediction model between domain experts and industry-academia-research institutions, and the distance between the key core technologies "domain experts" and "industry-academia-research cooperation structure" is calculated. As a supplement to the identification method, it can provide cooperation prediction for industry-academia-research cooperation.

[0220] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying experts in key core technology research areas and institutions of learning, research and production, characterized in that, The method comprises the following steps: S1: obtaining a scientific literature corpus of key core technology fields; S2: constructing a field expert knowledge network for key core technology based on the scientific literature corpus, to obtain field expert data for key core technology research; S3: constructing an identification network of industry-university-research cooperation institutions based on key core technology fields, to select industry-university-research institutions to be cooperated; S4: obtaining the matching similarity of field expert data and industry-university-research institutions to be cooperated, and confirming the required field experts of the industry-university-research institutions to be cooperated according to the matching similarity.

2. The method according to claim 1, wherein the method is characterized by, The S2 specifically comprises the following steps: S21: screening the data of the obtained scientific literature corpus, and obtaining literature key information according to the screened scientific literature corpus; The screening comprises eliminating non-journal paper data and literature lacking key information; The non-journal paper data comprises conference notices, call for papers and topic guidelines; The key information at least comprises literature authors, literature titles and key words; S22: constructing a field expert knowledge sub-network according to the literature key information; The field expert knowledge sub-networks include an author cooperation relationship sub-network V A , a document sub-network V L , a core technology field sub-network V D , and a patent sub-network V P ; S23: obtaining a field expert knowledge network for key core technology according to the field expert knowledge sub-network; The expression of the field expert knowledge network is KVN = {V A , V L , V S , V P , SE A-L-D-P} In the formula, KVN represents the field expert knowledge network; SE A-L-D-P represents the super edge defined according to the author cooperation relationship subnetwork V A , the literature subnetwork V L , the core technology field subnetwork V D , and the patent subnetwork V P ; A i represents the i th author node; L j represents the j th literature node; D k represents the k th core technology field node; P l represents the l th existing patent node; α (A i , L j ) = 1 indicates that there is an association relationship between the author node A i and the literature node L j ; α (A i , D k ) = 1 indicates that there is an association relationship between the author node A i and the core technology field node D k ; α (A i , P l ) = 1 indicates that there is an association relationship between the author node A i and the patent node P l ; α (L j , D k ) = 1 indicates that there is an association relationship between the literature node L j and the core technology field node D k ; α (D k , P l ) = 1 indicates that there is an association relationship between the core technology field node D k and the patent node P l . S24: constructing a field expert recommendation strategy according to the field expert knowledge network to obtain field expert data for key core technology research. 3.The method of identifying experts and research institutions in key core technology fields according to claim 2, characterized in that, The field expert recommendation strategy constructed in S24 is specifically S241: define the author cooperation sub-network V A The node subject attribute involved in forming the hyperedge is used to obtain the number of hyperedges of the domain expert knowledge network. The node body attribute includes node hyperdegree SD and hyperedge connectivity L(SE i ); S242: obtaining a document sub-network V according to the number of hyper-edges of the domain expert knowledge network L Chinese document node influence degree; The calculation formula of the literature node influence degree is wherein: I(l i ) denotes the document node influence degree; F(l i ) denotes the document node l i The total number of hyperedges composed of cited documents in the field expert knowledge hypernetwork model; A(l i ) denotes the total number of authors in the author subnetwork that form hyperedges with the cited documents of the document node l i ; N denotes the total number of hyperedges in the hypernetwork model; N a denotes the total number of author nodes in the author subnetwork; S243: obtaining a core technology field sub-network V according to the number of hyper-edges of the field expert knowledge network D the similarity of the intermediate field nodes; The calculation formula of the field node similarity is wherein: Sim didj represents the core technology field sub-network V D represents the field node d i represents the similarity of the field node d j ; represents the KL distance of the feature word probability distribution, and P and Q represent all the word items in the core technology field sub-network V D represents the field node d i represents the field node d j topic distribution occurrence event; P(i) denotes the probability distribution of the characteristic words of the domain node d i Q(j) denotes the probability distribution of the characteristic words of the domain node d j ​ S244: obtaining a patent sub-network V according to the number of hyper-edges of the field expert knowledge network P influence degree of the patent node; The calculation formula of the patent node influence degree is wherein: K(p i ) represents the influence degree of patent node p i ; R(p i ) represents the total number of hyper-edges in which patent node p i participates in the field expert knowledge hyper-network model; A(p i ) represents the total number of author nodes that form hyper-edges with the cited patents of patent node p i in the author sub-network; N represents the total number of hyper-edges in the field expert knowledge hyper-network model; and N a represents the total number of author nodes in the author sub-network. S245: constructing a hyperedge scoring model of the field expert knowledge network for key core technology according to the literature node influence degree, the field node similarity and the patent node influence degree, to obtain the hyperedge scoring value of the field expert for key core technology; The expression of the hyperedge scoring model is In the formula: SER(SE) i ) represents the hyperedge score of the domain expert knowledge network; N represents the total number of hyperedges in the domain expert knowledge network; I li Indicates the influence degree of document nodes; Sim didj Represents the domain node d i With domain node d i Similarity between them; K pi Indicates the influence of the patent node; β represents the set weight parameter; L(SE) i ) represents the superedge SE i Hyperedge connectivity; SER(SE) ai ) represents the author node a i The score of the hyperedges formed; SD ai Author node a i The node superdegree value; S(a) represents the superedge score value of author node a; S246: arranging the author nodes in the field expert knowledge network in descending order according to the hyperedge scoring value of the field expert for key core technology, to obtain a field expert recommendation order table; Obtaining field expert data for key core technology research based on the field expert recommendation order table.

4. The method according to claim 1, wherein the method is characterized by, The S3 specifically comprises the following steps: S31: confirming the industry-university-research characteristics variables of key core technology fields; The industry-university-research characteristics variables comprise technology theme correlation characteristics variables, institution competitiveness characteristics variables and geopolitical relationship characteristics variables; S32: taking the industry-university-research characteristics variables as the actor set in a social network based on social network analysis technology, to obtain a three-mode network of institution-theme word-region; And obtaining a two-mode network according to the three-mode network of institution-theme word-region, to expand the multi-mode matrix of industry-university-research of key core technology fields; The two-mode network at least comprises an institution-key core technology theme word two-mode network; S33: using natural language processing technology NLP, extracting the corresponding key core technology theme words of each industry-university-research institution according to the unstructured text of the scientific and technological literature published by each industry-university-research institution in the key core technology field; and the key core technology keywords are clustered based on the LDA topic model to obtain a plurality of clusters of the technology subject associated with the clusters of the research institutions; S34: Similarity indexes between the clustered members in each cluster of the research institutions are determined, and a core-edge network structure between the research institutions is established based on the similarity indexes; The similarity indexes include one or more combinations of a Jaccard similarity coefficient, a cosine similarity, and an Euclidean distance; The establishment rule of the core-edge network structure is The clustered members in the clusters of the research institutions are taken as network nodes, and the similarity indexes between the clustered members are taken as edge weights between the network nodes; One of the clusters of the research institutions corresponds to one core-edge pair; The core-edge pair includes a core clustered member and an edge clustered member, and the core clustered member is a clustered member with a similarity index greater than a preset similarity threshold in the cluster of the research institutions; S35: A research institution identification network is obtained from each core-edge network structure between the research institutions by using Ucinet software, and a core degree of each network node in the research institution identification network is obtained. The network nodes are arranged in descending order according to the core degrees to obtain a network node recommendation list, and the research institutions to be cooperated are selected based on the network node recommendation list.

5. The method according to claim 4, wherein the method is characterized by, The S4 specifically includes the following steps: S41: The obtained field expert data and the research institution data to be cooperated are taken as feature nodes, and each node directed network is obtained from a preset link path; The preset link path includes a field expert-field expert link path, a research institution-field expert link path, and a field expert-research institution link path; The node directed network includes a field expert-field expert directed network, a research institution-field expert directed network, and a field expert-research institution directed network; S42: A target node vector and a neighbor node vector of a target node are defined according to each node directed network; The target node vector is used to represent the cooperation relationship between nodes in any two node directed networks; The neighbor node vector is used to represent the feature information of the field experts or the feature information of the research institutions in each node directed network; The feature information of the field experts at least includes the total number of literature of the field experts, the number of cited literature, the total number of patents, and the number of cited patents; The feature information of the research institutions at least includes the business situation, the enterprise size, and the patent achievements; An heterogeneous graph network is obtained from the target node vector and the neighbor node vector based on a graph convolutional neural network, and the heterogeneous graph network is taken as an input layer of an R-GCN model; S43: A node distance is obtained from the target node vector and the neighbor node vector based on an element-by-element point multiplication method, and the node distance is taken as a predicted score of the cooperation relationship between the field experts and the research institutions output by the R-GCN model; The target node vector and the neighbor node vector input by the input layer are aggregated by an activation function to obtain a center node output feature of the R-GCN model; An expression of the center node output feature of the R-GCN model is where: r e {1,...,R} = R represents the type of collaboration relationship; i e {1,...,N} = N represents the type of node; represents the lth layer node representation of node i; represents the lth layer node representation of all neighbor nodes of node j; represents the set of neighbor nodes of node i with relationship r; c i，r represents a regularization constant; represents a linear transformation function; S44: Randomly initialize the network weight parameters of the R-GCN model, take the target node vector and the neighbor node vector in the obtained heterogeneous graph network as input data, take the predicted score of the cooperation relationship of the field expert and the industry-university-research institution as output data, train the R-GCN model to obtain a link prediction model of the field expert and the industry-university-research institution; S45: Select HingeLoss as the loss function of the R-GCN model, obtain the Loss value of the center node output feature of the R-GCN model and the center node output feature of the R-GCN model obtained by the last model training, and compare the Loss value with a preset loss threshold value; If the Loss value is greater than or equal to the preset loss threshold value, the network weight parameters of the R-GCN model are adaptively updated, and S44 is repeatedly executed; If the Loss value is less than the preset loss threshold value, the trained R-GCN model is the link prediction model of the field expert and the industry-university-research institution; S46: Obtain the predicted score of the cooperation relationship of the industry-university-research institution according to the link prediction model of the field expert and the industry-university-research institution, take the predicted score of the cooperation relationship as the matching similarity of the field expert data and the industry-university-research institution data to be cooperated, and then confirm the field expert selected by the industry-university-research institution to be cooperated according to the matching similarity.

Images