Patent and expert joint matching method and system for enterprise technology attack demand

Abstract

The invention relates to the field of intelligent technology recommendation, and discloses a patent and expert joint matching method and system oriented to enterprise technology attack demand, and the method comprises the steps: employing an enterprise demand feature model and a patent multi-dimensional feature model, and carrying out the vectorization of enterprise attack demand data and patent achievement data; inputting the obtained demand vector and patent vector into a demand and patent matching model, carrying out weighted similarity calculation on the demand and patent matching model, and grading and sorting according to a calculation result to obtain a patent matching feedback result; according to a patent matching feedback result, expert information is extracted, power law distribution, an expert cooperation network and actual enterprise demands are combined, experts matched with the demands of the enterprise are determined, a more accurate patent and expert matching result is obtained, and patent and expert joint recommendation for enterprise technology attack demands is realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent technology recommendation, and specifically to a patent and expert joint matching method and system for enterprise technology research and development needs. Background Art

[0002] In the context of frequent iterations of emerging technologies in various industries, enterprises are facing fierce technological competition and ever-changing market demands. Technological breakthroughs have become the key to promoting continuous innovation and maintaining competitiveness. In order to effectively respond to technological challenges, enterprises not only need internal R&D capabilities, but also increasingly rely on the introduction of external patented technologies and expert knowledge. However, there are some shortcomings in the existing patent and expert recommendation methods, which to some extent restrict the efficiency and success rate of enterprise technological breakthroughs.

[0003] First, the text data of enterprise technology requirements is often unstructured and contains a large amount of natural language descriptions, which makes it difficult to extract useful information from the requirements text and quantify it accurately. Existing methods often lack effective information extraction and quantification methods when processing such unstructured data, resulting in an inability to accurately grasp the technology requirements of enterprises, which in turn affects the matching quality of patents and experts.

[0004] Secondly, although keyword extraction technology plays an important role in patent analysis, existing technologies often do not fully consider word order and context information, which limits the understanding of the deep semantics of patent texts. Lack of consideration of word order and context may lead to mis-extraction or misunderstanding of keywords, thus affecting the matching accuracy between patents and corporate technology needs.

[0005] In addition, the existing patent demand and patent matching models often rely on simple similarity calculations when processing large-scale patent data, without deeply exploring the technical logic and application scenarios behind the patents. This superficial matching method is difficult to meet the requirements of enterprises for the depth and breadth of patents, and it is also difficult to reveal the inherent connection between patents and enterprise technology needs.

[0006] Finally, in terms of expert recommendation, although a variety of algorithms and technologies have been proposed in order to achieve accurate matching, these methods still have deficiencies in the evaluation and ranking of experts. The existing system may rely too much on quantitative indicators, ignoring the expert's R&D capabilities, industry authority, and fit with the needs of the enterprise, and failing to fully combine the actual needs of the enterprise and the centrality and influence of the expert in his or her field. For example, an expert may be at the core or edge of his or her research field. However, the existing recommendation method may not accurately capture and give the corresponding weight to this position, resulting in underestimation or overestimation in the recommendation ranking. In addition, the existing system is too stereotyped and often cannot customize the recommendation scheme in combination with the actual situation of the enterprise. For example, when the expert corresponding to the patent required by the enterprise is at the head, middle or tail of its power law distribution, the recommendation scheme required by the enterprise is different. These recommended experts may not be the best candidates that best meet the current technical needs of the enterprise, thereby reducing the actual utility of expert recommendation.

[0007] In summary, the existing patent and expert recommendation methods have shortcomings in unstructured data processing, keyword extraction, patent deep matching and expert recommendation. These problems urgently need to be solved through more advanced technologies and methods to improve the efficiency and quality of corporate technological research. Summary of the invention

[0008] In view of the above-mentioned technical deficiencies, the technical problem to be solved by the present invention is to provide a patent and expert joint matching method and system for the technical research needs of enterprises, aiming to solve the problems of insufficient accuracy and relevance of existing patent and expert recommendation methods in unstructured data processing, keyword extraction, patent deep matching and expert recommendation.

[0009] In order to solve the above technical problems, the present invention adopts the following technical solutions: The present invention provides a patent and expert joint matching method for enterprise technical research needs, including: S1. Import enterprise demand into the enterprise demand feature model to obtain the demand vector; S2. Determine the patent field according to the needs of the enterprise, obtain patent data in related fields, and import the patent data in related fields into the patent multi-dimensional feature model to obtain patent vectors; S3. Input the demand vector and patent vector into the demand and patent matching model to perform weighted similarity calculation, rank the calculation results, and feedback the patent matching results in order of high and low; S4. According to the patent matching results, based on the power law distribution, expert cooperation network and enterprise needs, integrate expert information and feedback the joint matching results of patents and experts.

[0010] Furthermore, the step S1 includes the following steps: S11. Preprocess the enterprise demand text, extract keywords from the preprocessed data, and manually expand the enterprise demand text to construct the enterprise demand feature text; S12. Import the enterprise demand feature text into the Doc2vec model to obtain a demand vector D.

[0011] Furthermore, the patent multidimensional feature model includes an LDA layer, a Doc2vec layer and a multidimensional feature construction layer.

[0012] Furthermore, the step S2 includes the following steps: S21, extract patent title information, patent abstract information and patent technical efficacy information from patent data in related fields and import them into the LDA layer to generate patent subject information; S22, integrating the patent title information, patent abstract information, patent technical efficacy information and patent subject information into patent achievement information and importing it into the Doc2vec layer to obtain the patent title vector X, the patent abstract vector Y, the patent technical efficacy vector Z and the patent subject vector W; S23. Import the patent title vector X, the patent abstract vector Y, the patent technical efficacy vector Z and the patent theme vector W into the Doc2vec layer multidimensional feature construction layer, and obtain the patent vector E according to the multidimensional feature model.

[0013] Furthermore, the step S3 includes the following steps: S31, importing the demand vector D and the patent vector E into the demand and patent matching model, performing weighted similarity calculation, and obtaining the similarity between the demand vector and the patent vector; S32. Sort the similarities between the demand vector D and the patent vector E in order of high and low, and take out the top N patents with the highest scores to feedback the patent matching results.

[0014] Furthermore, the step S31 includes the following steps: S311, calculate the cosine similarity of the demand vector D and the patent vector E to obtain the , , and The similarity index is: is the similarity between the demand vector D and the patent title vector X, is the similarity between the demand vector D and the patent abstract vector Y, is the similarity between the demand vector D and the patent technology efficacy vector Z, is the similarity between the demand vector D and the patent subject vector W; S312: Input similarity index to weighted similarity calculation layer to calculate similarity between demand vector D and patent vector E as follows: ;

[0015] Among them, a, b, c and d are the similarity weights between the enterprise demand feature vector and the patent title vector, patent abstract vector, patent technical efficacy vector and patent theme vector, and they take different values ​​in different data sets based on the calculation results.

[0016] Further, the step S4 includes the following steps: S4. According to the patent matching results, based on the power law distribution, expert cooperation network and enterprise needs, integrate expert information and feedback the joint matching results of patents and experts.

[0017] S41, based on the top N patents with the highest scores taken out in S3, feedback patent matching results, identify the inventor IDs corresponding to the top N patents, perform disambiguation processing on inventors with the same name to obtain a list of experts with independent IDs; S42. According to the expert list, the information of the patents invented by the experts is sorted out, and by counting the number of patents invented by the experts, the experts are divided into three layers: head, middle and tail using power law distribution; According to the power law distribution formula: ;

[0018] in, is the shape parameter, is the scale parameter, is a random variable; The shape parameters are determined by using maximum likelihood estimation , calculate the KS statistic of the fitted distribution and the empirical distribution as the goodness of fit statistic, and take the minimum KS statistic to determine the lower bound , and calculate the goodness of fit value , draw the expert-patent quantity distribution graph and add the fitted power-law distribution curve to obtain the power-law distribution graph; According to the shape parameters determined by , Nether and the goodness-of-fit value The fitting results and the features of the drawn power-law distribution graph are used to divide the experts into head experts, middle experts and tail experts, and the power-law division results are sorted into the expert list as part of the expert's label; S43. Based on the expert list, an expert cooperation network is constructed based on the relationship between experts cooperating in publishing patents, and network indicators are calculated to classify the positions of experts in the network as core or marginal positions as another part of the expert label; S44. Combining the power law partition result and the calculated network index to form an expert label, and determining a suitable expert selection scheme.

[0019] Furthermore, the step S43 includes the following steps: S431, respectively calculate the expert's degree centrality, closeness centrality and betweenness centrality indices; S432, combining the calculation results of the above indicators and the distribution characteristics of expert nodes in the cooperation network to determine whether the expert is in a core or peripheral position in the network; The core or edge position is determined as follows: the degree centrality, closeness centrality and betweenness centrality indicators of the experts are sorted in ascending order, with R being the largest number. The position of the experts in the network is delineated for the threshold, and the experts ranked before the threshold are marked as being in the core position, and the rest are in the edge position.

[0020] Furthermore, the degree centrality, closeness centrality and betweenness centrality indicators of experts are calculated as follows: Expert’s degree centrality: ;

[0021] in, is the number of direct links between node i and other nodes in the network. The larger the number, the greater the degree centrality of the node. ij Indicates that there is an edge between node i and node j, and N is the number of nodes in the network; Closeness Centrality: ;

[0022] in, is the reciprocal of the average distance from node i to all other nodes in the network, represents the shortest distance between node i and node j; Betweenness Centrality: ;

[0023] in, is the number of shortest paths passing through node i, represents the total number of shortest paths from node s to node t, Indicates the nodes passed through in these shortest paths The number of paths.

[0024] A patent and expert joint matching system for enterprise technology research needs, including: The acquisition module imports enterprise demand into the enterprise demand feature model to obtain the demand vector; The input module determines the patent field according to the enterprise's needs, obtains patent data in related fields, and imports the patent data in related fields into the patent multi-dimensional feature model to obtain patent vectors; The calculation module inputs the demand vector and patent vector into the demand and patent matching model to perform weighted similarity calculation, scores and sorts the calculation results, and feeds back the patent matching results in order of high and low; The output module integrates expert information and feeds back the joint matching results of patents and experts based on the power law distribution, expert cooperation network and enterprise needs according to the patent matching results. The beneficial effects of the present invention are: 1. The patent and expert matching results obtained by this invention are highly accurate, realizing the joint recommendation of patents and experts for the technological breakthrough needs of enterprises.

[0025] 2. The present invention comprehensively considers the expert's R&D capabilities, industry status and compatibility with enterprise needs, and quickly and comprehensively provides a customized expert recommendation plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 A flowchart of a patent and expert joint matching method for enterprise technology research needs provided by an embodiment of the present invention.

[0028] Figure 2 Schematic diagram of the process of obtaining the demand vector.

[0029] Figure 3 Schematic diagram of the process of obtaining patent vectors.

[0030] Figure 4 A flowchart for obtaining patent matching results; Figure 5 Schematic diagram of the process of obtaining patent and expert joint matching results. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Example 1

[0033] Reference Figure 1~Figure 5This embodiment provides a patent and expert joint matching method for enterprise technology research needs, such as Figure 1 As shown, including: S1. Import enterprise demand into the enterprise demand feature model to obtain the demand vector. It should be noted that: like Figure 2 As shown, S11, importing the enterprise demand text into the enterprise demand feature model, performing word segmentation and stop word filtering on the demand text in turn, extracting keywords, searching for synonyms, antonyms, hypernyms and hyponyms for the extracted keywords using professional terminology dictionaries and natural language processing tools, adding details and specific descriptions to the demand text and expanding related professional and technical concepts, realizing the artificial expansion of the demand text to generate the final demand text, and obtaining the enterprise demand text "selecting an electrolyte with high ionic conductivity, through ***, thereby improving ***, reducing ***, and improving battery ***." S12. Import the enterprise demand feature text into the Doc2vec model to obtain the demand vector D.

[0034] S2. This embodiment obtains 3526 pieces of patent data in related fields from the patent database according to the enterprise demand text. The patent data in related fields include the title, abstract, technical effect statement, inventor, first inventor and current right holder information of each patent, and imports the patent data in related fields into the patent multidimensional feature model to obtain patent vectors; wherein, the patent multidimensional feature model includes LDA layer, Doc2vec layer and multidimensional feature construction layer. It should be noted that: like Figure 3 As shown, S21, extract patent title information, patent abstract information and patent technical efficacy information of patent data in related fields and import them into LDA layer, obtain subject word text through LDA layer, and the field patent data is divided into 5 themes and each theme feature word generates patent subject information; S22, integrating the patent title information, patent abstract information, patent technical efficacy information and patent subject information into patent achievement information and importing it into the Doc2vec layer to obtain the patent title vector X, the patent abstract vector Y, the patent technical efficacy vector Z and the patent subject vector W; S23. Import the patent title vector X, the patent abstract vector Y, the patent technical efficacy vector Z and the patent theme vector W into the Doc2vec layer multidimensional feature construction layer, and obtain the patent vector E according to the multidimensional feature model.

[0035] S3. Input the demand vector and patent vector into the demand and patent matching model to calculate the weighted similarity, rank the calculation results, and feedback the patent matching results in order of high and low. It should be noted that: like Figure 4As shown, S31, the demand vector D and the patent vector E are introduced into the demand and patent matching model, and weighted similarity calculation is performed to obtain the similarity between the demand vector and the patent vector.

[0036] S311, calculate the cosine similarity of the demand vector D and the patent vector E to obtain the , , and The similarity index is: is the similarity between the demand vector D and the patent title vector X, is the similarity between the demand vector D and the patent abstract vector Y, is the similarity between the demand vector D and the patent technology efficacy vector Z, is the similarity between the demand vector D and the patent subject vector W; S312: Input similarity index to weighted similarity calculation layer to calculate similarity between demand vector D and patent vector E as follows: ;

[0037] Among them, a, b, c and d are the similarity weights between the enterprise demand feature vector and the patent title vector, patent abstract vector, patent technical efficacy vector and patent theme vector, and they take different values ​​in different data sets according to the calculation effect. According to the data characteristics, in this example, the values ​​of a, b, c and d are 0.1, 0.4, 0.3 and 0.2 respectively.

[0038] S32. Sort the similarity between the demand vector D and the patent vector E by high or low, and take out the top N patents with the highest scores to feedback the patent matching results, where N is 100. The calculation results of the similarity between the demand vector D and the patent vector E are shown in Table 1: Table 1: Some similarity results.

[0039]

[0040] S4. Based on the patent matching results, power law distribution, expert cooperation network and enterprise needs, integrate expert information and feedback the patent and expert joint matching results. It should be noted that: like Figure 5 As shown, S41, based on the top 100 patents with the highest scores taken out in S3, feedback patent matching results are obtained, the inventor IDs corresponding to the top N patents are identified, and the inventors with the same name are disambiguated to obtain a list of experts with independent IDs; S42. According to the expert list, the information of the patents invented by the experts is sorted out, and by counting the number of patents invented by the experts, the experts are divided into three layers: head, middle and tail using power law distribution; According to the power law distribution formula: ; in, is the shape parameter, is the scale parameter, is a random variable; The shape parameters are determined by using maximum likelihood estimation , calculate the KS statistic of the fitted distribution and the empirical distribution as the goodness of fit statistic, and take the minimum KS statistic to determine the lower bound , and calculate the goodness of fit value , draw the expert-patent quantity distribution graph and add the fitted power-law distribution curve to obtain the power-law distribution graph; According to the shape parameters determined by , Nether and the goodness-of-fit value The fitting results and the features of the drawn power-law distribution graph are used to divide the experts into head experts, middle experts and tail experts, and the power-law division results are sorted into the expert list as part of the expert's label; S43. Based on the expert list, an expert cooperation network is constructed based on the relationship between experts cooperating in publishing patents, and network indicators are calculated. The positions of experts in the network are divided into core or marginal positions as another part of the expert label.

[0041] S431, respectively calculate the expert's degree centrality, closeness centrality and betweenness centrality indices; S432, combining the calculation results of the above indicators and the distribution characteristics of expert nodes in the cooperation network to determine whether the expert is in a core or peripheral position in the network; The core or edge position is determined as follows: the degree centrality, closeness centrality and betweenness centrality indicators of the experts are sorted in ascending order, with R being the largest number. The position of the experts in the network is delineated for the threshold, and the experts ranked before the threshold are marked as being in the core position, and the rest are in the edge position.

[0042] Preferably, the degree centrality, closeness centrality and betweenness centrality indices of the experts are calculated as follows: Expert’s degree centrality: ; in, is the number of direct links between node i and other nodes in the network. The larger the number, the greater the degree centrality of the node. ij Indicates that there is an edge between node i and node j. N is the number of nodes in the network. The higher the degree of a node, the more important it is in the network and it may be in a core position. Closeness Centrality: ; in, is the reciprocal of the average distance from node i to all other nodes in the network, It represents the shortest distance between node i and node j. The higher the closeness centrality of a node, the closer the connection between the node and other nodes in the network is, and it may be in a core position. Betweenness Centrality: ; in, is the number of shortest paths passing through node i, represents the total number of shortest paths from node s to node t, Indicates the nodes passed through in these shortest paths The number of paths that a node passes through; betweenness centrality measures the degree to which a node acts as an intermediary or bridge in the network, that is, the frequency with which a node passes information between other nodes. The higher the betweenness centrality of a node, the more important the node plays a connecting role in the network.

[0043] For example, the degree centrality of Li**, Zhang**, Wu* and others far exceeds that of others, indicating that they are involved in more patent research and development in this field and have a high frequency of cooperation with other inventors. Zhang**, Wu*, Liu**, Cao** and others are ranked at the front in terms of closeness centrality, and they cooperate closely with most inventors and have certain research and development capabilities or industry authority; Li**, Liu**, Yu* and others are ranked at the front in terms of betweenness centrality, and they promote exchanges among multiple individuals and organizations in cooperative innovation, which is conducive to the circulation of technical resources in the field.

[0044] S44. Combining the power law partition result and the calculated network index to form an expert label, and determining a suitable expert selection scheme.

[0045] Specifically, the patent results matched in step S32 and the expert results matched in step S432 are sorted, and the sorted results are provided to the enterprise as shown in Table 2. The enterprise evaluates the fit of the results, and adjusts and optimizes the matching process based on the enterprise's evaluation feedback, and finally obtains a solution that is adapted to the enterprise in actual application.

[0046] Table 2: Example of patent-expert joint matching results.

[0047] Serial number title Similarity First Inventor Expert Tags 779 For lithium metal*** 71.523 List** Middle - close to the central core 1252 A high-load*** application 73.769 open** Middle - close to the central core 1490 High-capacity *** preparation method and application 70.838 king** Middle-Intermediate Centrality Core 1921 A *** application 74.705 Cui** Head-degree centrality core 2377 A kind of *** preparation method 75.41 List** Head-Mesocentric Core This embodiment also provides a patent and expert joint matching system for enterprise technology research needs, including: The acquisition module imports enterprise demand into the enterprise demand feature model to obtain the demand vector; The input module determines the patent field according to the enterprise's needs, obtains patent data in related fields, and imports the patent data in related fields into the patent multi-dimensional feature model to obtain patent vectors; The calculation module inputs the demand vector and patent vector into the demand and patent matching model to perform weighted similarity calculation, scores and sorts the calculation results, and feeds back the patent matching results in order of high and low; The output module integrates expert information and feeds back the joint matching results of patents and experts based on the power-law distribution, expert cooperation network and enterprise needs according to the patent matching results.

[0048] Example 2

[0049] This embodiment is the second embodiment of the present invention. This embodiment is different from the first embodiment in that it provides a verification test of a patent and expert joint matching method for enterprise technical research needs, and verifies and explains the technical effects used in this method.

[0050] This embodiment selects a confusion matrix and compares this method with the traditional recommendation algorithm by calculating the accuracy, precision, recall and F1 index. The traditional method is a commonly used simple similarity calculation method using patent titles and abstracts. The evaluation results are shown in Table 3.

[0051] Table 3: Comparison of evaluation indicators.

[0052] Serial number Accuracy Accuracy Recall F1 value Traditional methods 94% 34% 98% 50% This method 98% 60% 95% 73% According to the table above, this method performs better in terms of patent matching, with accuracy, precision, and F1 value reaching 98%, 60%, 95%, and 73% respectively. The F1 value is much higher than the traditional method. In addition, the expert matching result test is obtained based on the actual application evaluation of the enterprise, and this method has been recognized by the enterprise in actual application.

[0053] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A patent and expert joint matching method for enterprise technology research needs, characterized by: include: S1. Import enterprise demand into the enterprise demand feature model to obtain the demand vector; S2. Determine the patent field according to the needs of the enterprise, obtain patent data in related fields, and import the patent data in related fields into the patent multi-dimensional feature model to obtain patent vectors; S3. Input the demand vector and patent vector into the demand and patent matching model to perform weighted similarity calculation, rank the calculation results, and feedback the patent matching results in order of high and low; S4. According to the patent matching results, based on the power law distribution, expert cooperation network and enterprise needs, integrate expert information and feedback the joint matching results of patents and experts.

2. The patent and expert joint matching method for enterprise technology research needs as claimed in claim 1, characterized in that: The step S1 comprises the following steps: S11. Preprocess the enterprise demand text, extract keywords from the preprocessed data, and manually expand the enterprise demand text to construct the enterprise demand feature text; S12. Import the enterprise demand feature text into the Doc2vec model to obtain a demand vector D.

3. The patent and expert joint matching method for enterprise technology research needs as claimed in claim 1, characterized in that: The patented multidimensional feature model includes an LDA layer, a Doc2vec layer and a multidimensional feature construction layer.

4. The patent and expert joint matching method for enterprise technology research needs as described in any one of claims 1 or 3, characterized in that: The step S2 comprises the following steps: S21, extract patent title information, patent abstract information and patent technical efficacy information from patent data in related fields and import them into the LDA layer to generate patent subject information; S22, integrating the patent title information, patent abstract information, patent technical efficacy information and patent subject information into patent achievement information and importing it into the Doc2vec layer to obtain the patent title vector X, the patent abstract vector Y, the patent technical efficacy vector Z and the patent subject vector W; S23. Import the patent title vector X, the patent abstract vector Y, the patent technical efficacy vector Z and the patent theme vector W into the Doc2vec layer multidimensional feature construction layer, and obtain the patent vector E according to the multidimensional feature model.

5. The patent and expert joint matching method for enterprise technology research needs according to claim 1, characterized in that: The step S3 comprises the following steps: S31, importing the demand vector D and the patent vector E into the demand and patent matching model, performing weighted similarity calculation, and obtaining the similarity between the demand vector and the patent vector; S32. Sort the similarities between the demand vector D and the patent vector E in order of high and low, and take out the top N patents with the highest scores to feedback the patent matching results.

6. The patent and expert joint matching method for enterprise technology research needs as claimed in claim 5, characterized in that: The step S31 comprises the following steps: S311, calculate the cosine similarity of the demand vector D and the patent vector E to obtain the , , and The similarity index is is the similarity between the demand vector D and the patent title vector X, is the similarity between the demand vector D and the patent abstract vector Y, is the similarity between the demand vector D and the patent technology efficacy vector Z, is the similarity between the demand vector D and the patent subject vector W; S312: Input similarity index to weighted similarity calculation layer to calculate similarity between demand vector D and patent vector E as follows: ； Among them, a, b, c and d are the similarity weights between the enterprise demand feature vector and the patent title vector, patent abstract vector, patent technical efficacy vector and patent theme vector, and they take different values ​​in different data sets based on the calculation results.

7. The patent and expert joint matching method for enterprise technology research needs according to claim 1, characterized in that: The step S4 comprises the following steps: S4. Based on the patent matching results, power law distribution, expert cooperation network and enterprise needs, integrate expert information and feedback the patent and expert joint matching results; S41, based on the top N patents with the highest scores taken out in S3, feedback patent matching results, identify the inventor IDs corresponding to the top N patents, perform disambiguation processing on inventors with the same name to obtain a list of experts with independent IDs; S42. According to the expert list, the information of the patents invented by the experts is sorted out, and by counting the number of patents invented by the experts, the experts are divided into three layers: head, middle and tail using power law distribution; According to the power law distribution formula: ； in, is the shape parameter, is the scale parameter, is a random variable; The shape parameters are determined by using maximum likelihood estimation , calculate the KS statistic of the fitted distribution and the empirical distribution as the goodness of fit statistic, and take the minimum KS statistic to determine the lower bound , and calculate the goodness of fit value , draw the expert-patent quantity distribution graph and add the fitted power-law distribution curve to obtain the power-law distribution graph; According to the shape parameters determined by , Nether and the goodness-of-fit value The fitting results and the features of the drawn power-law distribution graph are used to divide the experts into head experts, middle experts and tail experts, and the power-law division results are sorted into the expert list as part of the expert's label; S43. Based on the expert list, an expert cooperation network is constructed based on the relationship between experts cooperating in publishing patents, and network indicators are calculated to classify the positions of experts in the network as core or marginal positions as another part of the expert label; S44. Combining the power law partition result and the calculated network index to form an expert label, and determining a suitable expert selection scheme.

8. The patent and expert joint matching method for enterprise technology research needs according to claim 7, characterized in that: The step S43 comprises the following steps: S431, respectively calculate the expert's degree centrality, closeness centrality and betweenness centrality indices; S432, combining the calculation results of the above indicators and the distribution characteristics of expert nodes in the cooperation network to determine whether the expert is in a core or peripheral position in the network; The core or edge position is determined as follows: the degree centrality, closeness centrality and betweenness centrality indicators of the experts are sorted in ascending order, with R being the largest number. The position of the experts in the network is delineated for the threshold, and the experts ranked before the threshold are marked as being in the core position, and the rest are in the edge position.

9. The patent and expert joint matching method for enterprise technology research needs as claimed in claim 8, characterized in that: The degree centrality, closeness centrality, and betweenness centrality indicators of experts are calculated as follows: Expert’s degree centrality: ； in, is the number of direct links between node i and other nodes in the network. The larger the number, the greater the degree centrality of the node. ij Indicates that there is an edge between node i and node j, and N is the number of nodes in the network; Closeness Centrality: ； in, is the reciprocal of the average distance from node i to all other nodes in the network, represents the shortest distance between node i and node j; Betweenness Centrality: ； in, is the number of shortest paths passing through node i, represents the total number of shortest paths from node s to node t, Indicates the nodes passed through in these shortest paths The number of paths.

10. A system based on the patent and expert joint matching method for enterprise technology breakthrough needs as claimed in claim 1, characterized in that: The acquisition module imports enterprise demand into the enterprise demand feature model to obtain the demand vector; The input module determines the patent field according to the enterprise's needs, obtains patent data in related fields, and imports the patent data in related fields into the patent multi-dimensional feature model to obtain patent vectors; The calculation module inputs the demand vector and patent vector into the demand and patent matching model to perform weighted similarity calculation, scores and sorts the calculation results, and feeds back the patent matching results in order of high and low; The output module integrates expert information and feeds back the joint matching results of patents and experts based on the power-law distribution, expert cooperation network and enterprise needs according to the patent matching results.

Images