A professional talent capability evaluation method and system based on industrial big data

Abstract

The application discloses a kind of professional talent ability evaluation method and system based on industry big data, belong to the field of ability evaluation;The application first obtains the standardized ability factor data of target talent and the demand vector of industry, constructs the first industry knowledge graph including post skill node, post matching project node and enterprise name node based on historical industry data;Target talent node and demand node are integrated into graph and link relationship is established, forming the second industry knowledge graph, based on the graph, starting from target talent node and demand node respectively, forward and reverse attention propagation is carried out in combination with attention mechanism;By calculating the interaction of attention weight in bidirectional propagation path, generate ability-demand coupling matrix, finally, perform spectral analysis on the matrix, extract the principal eigenvector as the talent industry adaptation degree index and generate evaluation results.The application effectively solves the problem that the ability of talent cannot be accurately and efficiently evaluated in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of capability evaluation, and in particular to a professional talent capability evaluation method and system based on industrial big data. BACKGROUND

[0002] With the acceleration of industrial upgrading and technological change, enterprises' requirements for the capabilities of professional talents are increasingly refined and dynamic. Traditional talent capability evaluation methods mostly rely on keyword matching of resume texts, screening based on fixed rules, or simple quantification of historical qualifications. These methods have significant limitations. First, they usually deal with static and isolated data, which cannot reflect the dynamic relationship between talent skills and rapidly changing industrial demands. Second, traditional evaluation models lack deep insights into the industrial ecosystem and fail to evaluate talents in the complex relationship network composed of skills, projects, and enterprises, resulting in one-sided and lack of forward-looking evaluation results.

[0003] In recent years, although some studies have attempted to use big data and knowledge graph technology to improve talent evaluation, most existing solutions still focus on constructing static person-skill profiles and performing simple semantic matching. These methods essentially fail to simulate the "influence diffusion" path of talent capabilities in the industrial environment and the "traceability" path of industrial demand on talent capabilities. Therefore, their evaluation results cannot quantify the deep and structured adaptation relationship between talents and industrial positions, making it difficult to meet the strategic needs of modern industrial development for precise talent assessment and prediction.

[0004] These deficiencies result in the inability of existing technologies to accurately and efficiently evaluate the capabilities of talents. SUMMARY

[0005] The present application provides a professional talent capability evaluation method and system based on industrial big data to solve the problem of inaccurate and inefficient evaluation of talent capabilities in the prior art.

[0006] In a first aspect, the present application provides a professional talent capability evaluation method based on industrial big data, comprising:

[0007] obtaining standardized capability element data of a target talent and demand vectors of an industry; wherein the standardized capability element data includes historical work projects, a list of skill certificates, and an educational background; the demand vectors are generated by natural language processing and vectorization fusion of job responsibilities and job requirements in enterprise recruitment information within a preset time window;

[0008] Based on the post skills, post matching projects and post enterprise names obtained from historical industrial data, a first industrial knowledge graph is constructed; wherein the nodes of the first industrial knowledge graph include post skill nodes, post matching project nodes and enterprise name nodes, and the edges include co-occurrence relationships or membership relationships between the nodes;

[0009] The standardized ability element data of the target talent is integrated into the first industrial knowledge graph as each target talent node, and based on a preset semantic matching method, the target talent nodes are linked with the post matching project nodes and post skill nodes to which they belong, the demand vector is integrated into the first industrial knowledge graph as each demand node, and the post skill nodes and post matching project nodes matched through vector similarity calculation are linked to obtain a second industrial knowledge graph;

[0010] Based on the second industrial knowledge graph, a preset attention mechanism is combined to perform forward attention propagation from the target talent nodes and reverse attention propagation from the demand nodes; and an ability-demand coupling matrix is generated by calculating the interaction of attention weights in the forward and reverse propagation paths.

[0011] The ability-demand coupling matrix is subjected to spectral analysis, the principal eigenvector is extracted as a talent industrial adaptability index, and a professional talent ability evaluation result is generated according to the talent industrial adaptability index.

[0012] The present application first constructs a unified knowledge graph by vectorizing and integrating target talent data and industrial recruitment information, solves the technical problem of difficult unified processing and correlation analysis of multi-source heterogeneous data, and establishes a standardized data foundation for in-depth analysis. Subsequently, a bidirectional attention propagation mechanism is performed on the constructed second industrial knowledge graph, which enables the system to simultaneously simulate the radiation influence path of talent ability in the industrial environment (forward propagation) and the traceability positioning path of industrial demand on talent ability (reverse propagation), thereby dynamically capturing the complex and nonlinear interaction relationship between talent and demand. Finally, by quantifying the interaction intensity of the bidirectional propagation paths into a coupling matrix and performing spectral analysis, the most core structural features can be extracted as an adaptability index from the complex network relationship, which makes the evaluation result no longer dependent on the surface static matching, but based on the deep and quantitative index dynamically calculated from the entire industrial ecological network, significantly improving the accuracy, dynamics and forward-looking of talent ability evaluation, effectively solving the problem of inability to accurately and efficiently evaluate the ability of talents in the prior art.

[0013] Further, the demand vector is generated by natural language processing and vectorization fusion of post responsibilities and job requirements in enterprise recruitment information within a preset time window, specifically:

[0014] text preprocessing is performed on the post responsibilities and requirements to obtain standardized text; the preprocessing includes removing meaningless symbols, stop word filtering and text segmentation;

[0015] Based on the term frequency-inverse document frequency method, the standardized text is processed to obtain each keyword, and the keywords are classified into skill keywords and project keywords;

[0016] The skill keywords and project keywords are respectively converted into corresponding skill vectors and project vectors through a preset bag-of-words model;

[0017] Based on a preset time window, all skill vectors and project vectors under the same recruitment information are weighted average pooling processed to generate the demand vector.

[0018] Firstly, the post responsibilities and requirements are text preprocessed to remove meaningless symbols, stop word filtering and text segmentation. This step can remove noise information in the text and retain core semantic content, providing high-quality input data for subsequent keyword extraction and vectorization processing. Then, based on the term frequency-inverse document frequency (TF-IDF) method, the standardized text is processed to obtain each keyword, which is classified into skill keywords and project keywords. The TF-IDF method can highlight important words in the text, and the keyword classification further clarifies the core information related to skills and projects in the text, making the subsequent vectorization processing more targeted. Then, the skill keywords and project keywords are respectively converted into corresponding skill vectors and project vectors through a preset bag-of-words model. The bag-of-words model can convert the words in the text into numerical vector representation, which is convenient for computer processing and analysis, thereby converting the text information into mathematical form that can be used for subsequent calculation. Finally, based on a preset time window, all skill vectors and project vectors under the same recruitment information are weighted average pooling processed to generate the demand vector. Weighted average pooling processing can comprehensively consider multiple skill and project information under the same recruitment information to generate a vector that can fully reflect the job demand, so that the demand vector can more accurately represent the actual demand of enterprises for talents.

[0019] Further, based on a preset time window, all skill vectors and project vectors under the same recruitment information are weighted average pooling processed to generate the demand vector, specifically:

[0020] The frequency of the skill vectors and project vectors in all recruitment information within the preset time window is counted;

[0021] According to the frequency, inverse document frequency values of each skill vector and project vector are calculated, and the inverse document frequency values are normalized based on a softmax function, and the normalized results are taken as weights of each vector;

[0022] According to the weights of each vector, weighted average pooling processing is performed on all skill vectors and project vectors under the same recruitment information to obtain the demand vector.

[0023] Firstly, the present application calculates the appearance frequency of skill vectors and project vectors in all recruitment information in a preset time window, which can understand the universality of each vector in the market and provide basic data for subsequent weight calculation. Then, the inverse document frequency value of each vector is calculated according to the frequency, and the higher the inverse document frequency value, the stronger the uniqueness of the vector in the specific recruitment information, and the greater the discrimination of the job demand. Then, the inverse document frequency value is normalized based on the softmax function, and the normalized result is taken as the weight, and the softmax function can ensure that the sum of the weights is 1, so that the weights have comparability and conform to the probability distribution characteristics. Finally, according to the weight, weighted average pooling processing is performed on all skill vectors and project vectors under the same recruitment information to obtain the demand vector. The weighted average pooling processing integrates the importance and uniqueness of each vector, so that the generated demand vector can more comprehensively and accurately reflect the core demand of the post. This method based on data-driven and statistical principles can effectively avoid the problem of ignoring key information due to simple average of vectors, improve the quality of demand vectors, and thus improve the accuracy of talent and post matching degree evaluation, which helps enterprises to more efficiently screen professional talents meeting the requirements of the post, and also provides more accurate self-positioning and development direction for talents.

[0024] Further, the first industrial knowledge graph is constructed based on the post skills, post matching projects and post enterprise names obtained from historical industrial data, specifically:

[0025] The standardized skill names, completed project names and enterprise full names are extracted from the historical industrial data through regular expression and dictionary matching;

[0026] The skill names, completed project names and enterprise full names are normalized to obtain each post skill node, post matching project node and enterprise name node;

[0027] Based on each node, according to the co-occurrence relationship of skills and projects often appearing in the same recruitment demand, edges between post skill nodes and post matching project nodes are constructed, and according to the membership relationship of projects being led or completed by enterprises, edges between post matching project nodes and enterprise name nodes are established;

[0028] Based on each edge and each node, a first industry knowledge graph is constructed.

[0029] The application extracts standardized skill names, completed project names and enterprise full names from historical industry data through regular expression and dictionary matching. This step can ensure the accuracy and consistency of the extracted data, and provide high-quality basic data for subsequent knowledge graph construction. Then, the extracted skill names, project names and enterprise full names are normalized to obtain various post skill nodes, post matching project nodes and enterprise name nodes. Normalization can eliminate data duplication and redundancy, making node information more standardized and unified. Then, based on each node, the edges between post skill nodes and post matching project nodes are constructed according to the co-occurrence relationship of skills and projects in the same recruitment demand. This co-occurrence relationship reflects the actual association between skills and projects, and can effectively reveal the application of skills in projects. At the same time, according to the membership relationship of projects led or completed by enterprises, the edges between post matching project nodes and enterprise name nodes are established. The membership relationship clarifies the attribution relationship between projects and enterprises, further enriching the semantic information of the knowledge graph. Finally, based on each edge and each node, a first industry knowledge graph is constructed. This knowledge graph not only contains rich node information, but also reveals the internal relationship between nodes through the relationship of edges, forming a structured industry knowledge system.

[0030] Further, the preset semantic matching method links the target talent nodes with the corresponding post matching project nodes and post skill nodes, specifically:

[0031] The historical work project description of the target talent and the description text of the post matching project node are respectively converted into sentence embedding vectors;

[0032] The first cosine similarity between the sentence embedding vector of the historical work project description and the description vector of each post matching project node is calculated;

[0033] The skill certificate name of the target talent and the name of the post skill node in the graph are respectively converted into word embedding vectors;

[0034] The second cosine similarity between the word embedding vector of the skill certificate name and the name vector of each post skill node is calculated;

[0035] If the first cosine similarity and the second cosine similarity exceed the preset first threshold, a relationship edge is established between the target talent node and the corresponding graph node.

[0036] The historical work project description of the target talent and the description text of the post matching project node are respectively converted into sentence embedding vectors in the application. This step can convert text information into numerical vector representation, which is convenient for subsequent similarity calculation. Then, the first cosine similarity between the sentence embedding vector of the historical work project description and the description vector of each post matching project node is calculated. Cosine similarity can quantify the semantic similarity between texts, thereby evaluating the relevance of the target talent's historical work project and the post matching project. Then, the skill certificate name of the target talent and the name of the post skill node in the graph are respectively converted into word embedding vectors. Similarly, word embedding vectors can convert skill certificate names and post skill names into numerical vector representations, which is convenient for calculating their semantic similarity. Then, the second cosine similarity between the word embedding vector of the skill certificate name and the vector of each post skill node name is calculated. This step further evaluates the matching degree of the target talent's skills and post skills. Finally, if the first cosine similarity and the second cosine similarity exceed the first preset threshold, a relationship edge is established between the target talent node and the corresponding graph node. By setting the threshold, the post matching project and post skill highly related to the target talent can be screened out, ensuring the accuracy and reliability of the link. This semantic matching-based method considers the semantic relevance of the target talent's historical work project and skill certificate and the post, which can more comprehensively evaluate the matching degree of talent and post, thereby improving the accuracy and scientificity of talent evaluation, and providing strong support for precise selection and career development of talents.

[0037] Further, the post skill nodes and post matching project nodes matched by vector similarity calculation are linked, specifically:

[0038] From the first industry knowledge graph, obtain the pre-stored vector representation of all post skill nodes and post matching project nodes to form a candidate vector set;

[0039] Calculate the cosine similarity between the demand vector and each vector in the candidate vector set to obtain a similarity score;

[0040] According to the similarity score, all post skill nodes and post matching project nodes are ranked in descending order;

[0041] Filter out the nodes ranked in the first preset number of positions and with a similarity score exceeding a second preset threshold as key demand nodes, and establish a relationship edge between the demand node and each filtered key demand node.

[0042] The application obtains the vector representation pre-stored in all post skill nodes and post matching project nodes in the first industry knowledge graph to form a candidate vector set. This step provides comprehensive candidate objects for subsequent similarity calculation, ensuring the comprehensiveness of matching. Then, the cosine similarity of the demand vector and each vector in the candidate vector set is calculated to obtain a similarity score. As an effective similarity measurement method, the cosine similarity can quantify the semantic correlation between the demand vector and the candidate vector, thereby providing a scientific basis for screening matching nodes. Then, all post skill nodes and post matching project nodes are ranked in descending order according to the similarity score. This step can quickly locate the most relevant nodes to the demand vector, improving the efficiency and accuracy of matching. Finally, the nodes ranked in the first preset number and having a similarity score exceeding a preset second threshold are selected as key demand nodes, and a relationship edge is established between the demand node and each selected key demand node. By setting the threshold and ranking limit, it can be ensured that the selected nodes not only meet the similarity requirement but also have a certain control in quantity, avoiding too many or too few matching results, improving the practicality and reliability of matching. This vector similarity-based matching method can effectively improve the matching accuracy of the demand vector and the nodes in the industry knowledge graph, providing more accurate post demand information for talent capability evaluation, thereby improving the scientificity and effectiveness of talent and post matching degree evaluation, helping enterprises more efficiently screen professional talents meeting the post requirements, and also providing more accurate self-positioning and development direction reference for talents.

[0043] Further, based on the second industry knowledge graph, forward attention propagation is performed starting from the target talent node, and reverse attention propagation is performed starting from the demand node, specifically:

[0044] Assigning a feature vector to each node in the second industry knowledge graph;

[0045] Taking each target talent node as a first center node, finding all neighbor nodes directly connected to the first center node through edges, the neighbor nodes including post skill nodes, post matching project nodes, and enterprise name nodes;

[0046] Based on the attention mechanism, calculating the association degree of the first center node and the feature vector of each neighbor node to calculate the first weight of each neighbor node for each first center node; performing weighted summation on the feature vector of the neighbor node according to the first weight to obtain a first aggregated feature, and updating the feature vector of the first center node in combination with the first aggregated feature;

[0047] Taking the demand node as a second center node, an upstream node directly pointed to the second center node by an edge is found, and the upstream node includes a post skill node, a post matching project node, and an enterprise name node;

[0048] Based on the attention mechanism, each second weight of each upstream node of the second center node is calculated by calculating the correlation degree of the second center node and the feature vector of each upstream node; and the feature vector of the upstream node is weighted and summed according to the second weight, to obtain a second aggregated feature, and the feature vector of the second center node is updated in combination with the second aggregated feature;

[0049] Until the forward and reverse propagation of the attention mechanism complete the calculation of a preset number of layers.

[0050] The application allocates feature vectors to target talent nodes and demand nodes in the second industrial knowledge graph, which provides an initial feature representation for subsequent attention propagation. Then, taking the target talent node as a first center node, all directly connected neighbor nodes are found, including post skill nodes, post matching project nodes, and enterprise name nodes, which can determine the industrial information directly related to the target talent. Then, based on the attention mechanism, a first weight of the first center node for each neighbor node is calculated, and the feature vectors of the neighbor nodes are weighted and summed according to the weights to obtain a first aggregated feature, and the feature vector of the first center node is updated. This process enables the target talent node to aggregate important information of its related neighbor nodes, thereby more comprehensively reflecting the industrial adaptability of the target talent. Similarly, taking the demand node as a second center node, all directly pointed upstream nodes are found, and based on the attention mechanism, a second weight is calculated to obtain a second aggregated feature, and the feature vector of the demand node is updated. This process enables the demand node to aggregate important information of its related upstream nodes, thereby more accurately reflecting the industrial background of the post demand. Through forward and reverse attention propagation, until the calculation of a preset number of layers is completed, the multi-level correlation between the target talent and the post demand can be gradually and deeply mined, so that the finally updated feature vector can more accurately reflect the matching degree of talent and post. This propagation method based on the attention mechanism not only highlights the information of important nodes, but also captures complex industrial relationships through multi-layer propagation, thereby improving the scientificity and accuracy of talent evaluation, and providing strong support for precise selection and career development of talents.

[0051] Further, the ability-demand coupling matrix is generated by calculating the interaction of the attention weights in the forward and reverse propagation paths, specifically:

[0052] When the post skill nodes and the post matching project nodes are extracted as neighbor nodes from the final calculation result of the forward attention propagation, each first weight calculated by the target talent node as a first center node is taken as a capability influence factor, and the maximum weight in the first weights is taken as the capability influence factor;

[0053] When the post skill nodes and the post matching project nodes are extracted as upstream nodes from the final calculation result of the reverse attention propagation, each second weight calculated by the demand node as each second center node is taken as a demand urgency factor, and the maximum weight in the second weights is taken as the demand urgency factor;

[0054] The capability influence factor and the demand urgency factor of each post skill node and post matching project node are multiplied to obtain an interaction intensity value of the post skill node and the post matching project node.

[0055] The interaction intensity value of each post skill node and post matching project node is filled into a preset two-dimensional matrix at an intersection position corresponding to a row of the post skill node and a column of the post matching project node, to generate a capability-demand coupling matrix; one dimension in the two-dimensional matrix represents all post skill nodes, and the other dimension represents all post matching project nodes.

[0056] The application extracts the maximum first weight as a capability influence factor from the forward attention propagation result of the target talent node based on the second industry knowledge graph. This step highlights the strongest association of the target talent in each post skill and project, and reflects the potential contribution of the core competence of the talent to the post. Then, the maximum second weight is extracted as a demand urgency factor from the reverse attention propagation result of the demand node. This step highlights the strongest association of the post demand in each post skill and project, and reflects the urgent demand of the post for specific skills and projects. Then, the capability influence factor and the demand urgency factor are multiplied to obtain an interaction intensity value. This step comprehensively considers the matching degree of talent competence and post demand in a quantitative way. The higher the interaction intensity value is, the higher the matching degree of the talent and the post in the skill or the project is. Finally, the interaction intensity value is filled into a preset two-dimensional matrix to generate a capability-demand coupling matrix. The matrix directly shows the matching intensity distribution between all post skill nodes and post matching project nodes, and provides a comprehensive and accurate data basis for subsequent talent evaluation and post matching.

[0057] Further, the capability-demand coupling matrix is subjected to spectral analysis, a principal feature vector is extracted as a talent industry adaptation index, and a professional talent competence evaluation result is generated according to the talent industry adaptation index. Specifically,

[0058] The capability-demand coupling matrix is subjected to centering processing.

[0059] calculate a covariance matrix of the centering processed ability-demand coupling matrix, and perform eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and a corresponding eigenvector;

[0060] select an eigenvalue with the maximum modulus from each eigenvalue, and take the eigenvector corresponding to the eigenvalue with the maximum modulus as a principal eigenvector;

[0061] calculate an L2 norm value of the principal eigenvector, and map the L2 norm value to a predefined score interval, and take the mapped result as the talent industry adaptability index.

[0062] The present application performs centering processing on the ability-demand coupling matrix. This step eliminates the mean deviation of the data, so that the mean value of each row and each column of the matrix is zero, providing a standardized data basis for subsequent covariance matrix calculation. Then, the covariance matrix of the centering processed matrix is calculated, and eigenvalue decomposition is performed to obtain each eigenvalue and a corresponding eigenvector. Eigenvalue decomposition can reveal the most important features and directions in the matrix, where the size of the eigenvalue reflects the importance of the corresponding eigenvector in the data. Then, an eigenvalue with the maximum modulus is selected from each eigenvalue, and the eigenvector corresponding to the eigenvalue with the maximum modulus is taken as a principal eigenvector. The principal eigenvector represents the most important change direction in the data, and can reflect the core information of talent and industry adaptability to the greatest extent. Finally, the L2 norm of the principal eigenvector is calculated, and is mapped to a predefined score interval to obtain the talent industry adaptability index. The L2 norm measures the length of the eigenvector, and by mapping to the score interval, complex vector information can be converted into intuitive score values, facilitating understanding and application. This method based on spectral analysis can extract the most representative information from the complex coupling matrix through rigorous mathematical processing and feature extraction, thereby generating a scientific and objective talent industry adaptability index, providing a quantitative and comparable standard for professional talent ability evaluation, and helping enterprises to more accurately screen and cultivate talents that meet the needs of the industry.

[0063] In a second aspect, the present application provides a professional talent ability evaluation system based on industrial big data, comprising:

[0064] an acquisition module configured to acquire standardized ability element data of a target talent and demand vectors of an industry; wherein the standardized ability element data includes historical work projects, a list of skill certificates, and an educational background; and the demand vectors are generated by performing natural language processing and vectorization fusion on job responsibilities and job requirements in enterprise recruitment information within a preset time window;

[0065] The construction module is configured to construct a first industrial knowledge graph based on obtaining post skills, post matching projects, and post enterprise names from historical industrial data. The nodes of the first industrial knowledge graph include post skill nodes, post matching project nodes, and enterprise name nodes, and the edges include co-occurrence relationships or membership relationships between the nodes.

[0066] The linking module is configured to integrate standardized ability element data of the target talents as various target talent nodes into the first industrial knowledge graph, and link the various target talent nodes with the post matching project nodes and the post skill nodes to which the various target talent nodes belong based on a preset semantic matching method. The linking module is also configured to integrate the demand vectors as various demand nodes into the first industrial knowledge graph, and link the demand nodes with the post skill nodes and the post matching project nodes matched through vector similarity calculation to obtain a second industrial knowledge graph.

[0067] The calculation module is configured to, based on the second industrial knowledge graph, combine a preset attention mechanism, and perform forward attention propagation starting from the target talent nodes and reverse attention propagation starting from the demand nodes. The calculation module is also configured to generate an ability-demand coupling matrix by calculating the interaction of attention weights in the forward and reverse propagation paths.

[0068] The result output module is configured to perform spectral analysis on the ability-demand coupling matrix, extract a principal feature vector as a talent industrial adaptability index, and generate a professional talent ability evaluation result according to the talent industrial adaptability index.

[0069] This application presents a professional talent competency evaluation system based on industrial big data. Through modular design and collaborative work, it achieves accurate evaluation of professional talent capabilities, providing a scientific basis for matching talent with industry. First, the acquisition module collects standardized competency element data of target talent and industry demand vectors. The standardized competency element data covers historical work projects, skill certificate lists, and educational backgrounds, while the demand vectors are generated through natural language processing and vectorization fusion, providing comprehensive and quantitative information on talent and positions for subsequent evaluation. Next, the construction module builds a primary industry knowledge graph based on historical industry data, using job skills, job-matching projects, and company names as nodes. Edges are established between nodes through co-occurrence or membership relationships, forming a structured industry knowledge system and providing a basic framework for talent-job correlation analysis. Then, the linking module integrates target talent nodes and demand nodes into the knowledge graph and links them through semantic matching and vector similarity calculation, generating a secondary industry knowledge graph. This step achieves accurate mapping of talent and positions within the knowledge graph, providing a connection foundation for subsequent in-depth analysis. The computation module utilizes an attention mechanism for forward and backward attention propagation, and generates a capability-demand coupling matrix by calculating the interaction of attention weights. This process deeply explores the potential correlation between talent and positions, quantifying their matching degree. Finally, the results output module performs spectral analysis on the coupling matrix, extracts the principal eigenvector as the talent-industry fit index, and generates professional talent competency evaluation results. This step transforms complex matrix information into intuitive evaluation indicators, providing clear guidance for precise talent selection and career development. Through the orderly collaboration of its modules, the entire system automates the entire process from data collection to result output, improving the efficiency and accuracy of talent evaluation. This helps companies more efficiently screen professional talent that meets job requirements, while also providing talent with a more accurate reference for self-positioning and development direction. Attached Figure Description

[0070] Figure 1 : A schematic diagram of an embodiment of the professional talent competency evaluation method based on industrial big data provided in this application;

[0071] Figure 2 This is a schematic diagram of an embodiment of the professional talent competency evaluation system based on industrial big data provided in this application. Detailed Implementation

[0072] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0073] Embodiment one: please refer to Figure 1 In order to solve the problem that the prior art cannot accurately and efficiently evaluate the ability of talents, the present application provides a professional talent ability evaluation method based on industrial big data, which comprises steps S01-S05:

[0074] S01: obtaining standardized ability element data of a target talent and demand vectors of an industry; wherein the standardized ability element data comprises historical work projects, a skill certificate list and an educational background; the demand vectors are generated by natural language processing and vectorization fusion of post responsibilities and service requirements in enterprise recruitment information within a preset time window.

[0075] As a preferred embodiment of the present embodiment, the standardized ability element data of the target talent and the demand vectors of the industry are specifically:

[0076] When obtaining the standardized ability element data of the target talent, firstly, the personal ability related information of the target talent is collected and standardized. The historical work project information specifically includes the full name of the project participated or led by the talent, the field to which the project belongs, the specific responsibilities undertaken in the project, the core technology and tools involved in the project, and the project results (such as completion efficiency, economic benefits or industry influence, etc.), and these information are recorded through a unified data format (such as JSON structure), to ensure the comparability of the project information of different talents. The skill certificate list covers various professional skill certifications held by the talent, including certificate name (such as “senior software engineer certification” “PMP project management certification” etc.), issuing agency, certification time and certificate validity period, and the authenticity of the certificate is verified by comparing with the information database of authoritative certification agencies, and the expression of the certificate name is unified (such as “computer software qualification and level test - senior” is unified as “software senior engineer certificate”). The educational background information contains the highest education of the talent, the graduation college, the major, the graduation time and the professional course grades during the school period (if necessary), which are also recorded in a standardized field, for example, “bachelor” and “bachelor's degree” are unified as “bachelor”, to ensure the consistency of the information format;

[0077] For the demand vector of an industry, its generation process is based on the enterprise recruitment information within a preset time window. The preset time window can be flexibly set according to the speed of change of the demand for talents in the industry. For example, for industries such as the Internet and artificial intelligence that develop rapidly, the time window can be set to nearly 6 months, while for traditional manufacturing industries and other industries with relatively stable demand, the time window can be set to nearly 1 year to ensure that the captured demand is timely and representative. The enterprise recruitment information is collected from effective information on public recruitment platforms (such as recruitment websites, enterprise official recruitment pages, etc.), covering the job responsibilities (such as “responsible for the construction and maintenance of the big data platform” “lead the development of the customer relationship management system” etc.) and job requirements (such as “need 5 years of Java development experience” “preferably with AWS certification” etc.) published by various enterprises;

[0078] In generating the demand vector, first, the collected job responsibilities and job requirements text are preprocessed: meaningless symbols (such as special punctuation, garbled characters, etc.) in the text are removed through regular expressions, stop words (including “of” “is” “responsible” etc. words with no actual semantic meaning) are filtered using a stop word dictionary, and a word segmentation tool (such as a deep learning-based word segmentation model or the jieba word segmentation tool) is used to split the text into independent word units to obtain standardized text. Subsequently, the standardized text is processed based on the term frequency-inverse document frequency (TF-IDF) method, the frequency of each word in a single recruitment information (term frequency) and its distribution in all recruitment information (inverse document frequency) are calculated, the words with higher TF-IDF values are selected as keywords, and the keywords are classified into skill keywords (such as “Python programming” “machine learning algorithm”) and project keywords (such as “e-commerce platform construction” “data center migration”) according to their semantic attributes;

[0079] Next, the skill keywords and project keywords are respectively vectorized by a preset Bag-of-Words model: the Bag-of-Words model is pre-constructed to include a dictionary of common skills and project words in the industry, and each keyword is converted into a corresponding vector according to its position in the dictionary (for example, the position corresponding to the vector of "Python programming" is 1, and the positions of the rest are 0), to obtain the skill vector and the project vector. Then, based on the above preset time window, the total frequency of each skill vector and project vector appearing in all recruitment information in the time window is counted, the inverse document frequency value (IDF value, formula: log (total number of recruitment information / number of recruitment information containing the keyword corresponding to the vector)) of each vector is calculated according to the frequency, and the IDF values of all vectors are normalized by the softmax function (so that the sum of the weights of all vectors is 1), to obtain the weight of each vector (the higher the frequency and the more concentrated the distribution in the recruitment information, the greater the weight). Finally, for all skill vectors and project vectors under the same recruitment information, weighted average pooling processing is performed according to their respective weights (i.e., each vector element is multiplied by the corresponding weight and then summed), to generate the demand vector corresponding to the recruitment information; the demand vectors of all recruitment information in the same industry are weighted averaged or directly spliced to obtain the demand vector of the industry;

[0080] Through the above processing, the normalized target talent ability data and the quantified industry demand data are obtained, which lays a reliable data foundation for subsequent knowledge graph construction and deep analysis.

[0081] S02: Based on the job skills, job matching projects and job enterprise names obtained from historical industry data, a first industry knowledge graph is constructed; wherein the nodes of the first industry knowledge graph include job skill nodes, job matching project nodes and enterprise name nodes, and the edges include co-occurrence relationships or membership relationships between nodes.

[0082] As a preferred embodiment of the present embodiment, the construction of the first industry knowledge graph based on the job skills, job matching projects and job enterprise names obtained from historical industry data is specifically:

[0083] When constructing the first industry knowledge graph, first, the core elements are extracted from the multi-source historical industry data, which includes but is not limited to the public recruitment information database of the industry in the past 5 years, the enterprise project cooperation record, the technical white paper published by the industry association, the major project information disclosed in the annual report of listed companies and the completed project details publicized on the official website of the enterprise, etc.

[0084] In the element extraction phase, a combination of regular expressions and dictionary matching is used for accurate extraction: For job skills, a standardized dictionary covering general and specialized skills in various industries is pre-built (e.g., "distributed system development" "data cleaning" "blockchain deployment" in the information technology field), and the skill names in the text are identified through dictionary matching, and emerging skill terms are captured using regular expressions (e.g., "[A-Za-z]+[development / technology / engineering]"); For job matching projects, based on the characteristics of project names (such as containing "project" "engineering" "system" keywords), regular expressions are used to extract project names, and combined with industry project libraries for verification; For the name of the enterprise, the full name is matched with the enterprise business registration information library, and regular expressions are used to filter and extract the legal full name of the enterprise, ensuring that abbreviations or irregular names are excluded;

[0085] After extraction, the above elements are normalized to form standardized nodes: For skill names, synonyms or different names are unified (e.g., "Python programming" "Python development" are unified as "Python technology application"), and redundant modifiers are removed (e.g., "advanced Java development" is simplified to "Java development" as the basic skill node); For project names, different expressions of the same project (e.g., "XX e-commerce platform upgrade project" and "XX e-commerce system iteration project") are merged into the same node by comparing the core content (e.g., the technology field involved, the implementation subject); For the name of the enterprise, the full name of the enterprise business registration is used, and its branch information is associated (e.g., "XX Technology Co., Ltd. Beijing Branch" belongs to "XX Technology Co., Ltd." main node), and finally three types of nodes are formed: job skill nodes (e.g., "big data analysis" "mechanical design"), job matching project nodes (e.g., "smart city security system construction project" "new energy vehicle battery research and development project"), and enterprise name nodes (e.g., "XX Information Technology Co., Ltd." "YY Equipment Manufacturing Group");

[0086] In the construction of inter-node relationships, edges are established based on the internal association between elements: for the post skill node and the post matching project node, an edge is constructed by counting the frequency of their co-occurrence (i.e., co-occurrence relationship) in historical recruitment information and project demand documents, for example, the "machine learning algorithm" skill and the "user behavior prediction system development project" co-occur in 100 documents, then a co-occurrence edge is established between the two nodes, and the weight of the edge can be set based on the co-occurrence frequency (the higher the frequency, the greater the weight); for the post matching project node and the enterprise name node, the affiliation relationship is determined according to the implementer of the project, for example, the record information of "XX cloud computing data center construction project" shows that it is implemented by "ZZ cloud service Co., Ltd.", then an affiliation edge is established between the project node and the enterprise node, and the attribute of the edge is marked as "main implementation";

[0087] Through the above node construction and relationship linking, a first industry knowledge graph is formed, which contains three types of nodes: post skill, post matching project and enterprise name, and two types of edges: co-occurrence relationship and affiliation relationship. This graph can clearly present the association network between skills, projects and enterprises within the industry.

[0088] S03: Integrate the standardized ability element data of the target talent into the first industry knowledge graph as each target talent node, and link the target talent node with the post matching project node and the post skill node based on a pre-set semantic matching method. Integrate the demand vector into the first industry knowledge graph as each demand node, and link the matched post skill node and post matching project node through vector similarity calculation to obtain a second industry knowledge graph.

[0089] As a preferred embodiment of the present embodiment, the integration of the standardized ability element data of the target talent into the first industry knowledge graph as each target talent node, and the linking of the target talent node with the post matching project node and the post skill node based on a pre-set semantic matching method, specifically includes:

[0090] In the construction of the second industry knowledge graph, the standardized ability element data of the target talent is first converted into a target talent node and integrated into the first industry knowledge graph. Each target talent node contains the core identification information (such as unique identity code) of the talent and the structured data of the standardized ability elements (including detailed description of historical work projects, verified skill certificate list, standardized education background, etc.), which are stored as node attributes;

[0091] To realize the linking of the target talent node and the existing nodes in the first industry knowledge graph, a preset semantic matching method is adopted: for the association of the historical work project and the post matching project node, the description of the target talent's historical work project (such as "leading the algorithm optimization of the financial risk control system, using the XGBoost model to improve the prediction accuracy by 15%") and the description text of the post matching project node in the first industry knowledge graph (such as "financial field risk control model optimization project, involving machine learning algorithm tuning") are respectively converted into dimension-unified sentence embedding vectors through a pre-trained sentence embedding model (such as the BERT model); the first cosine similarity (value range 0-1) of the two vectors is calculated, and if the similarity exceeds a preset first threshold (such as 0.7), it is determined that there is a strong association between the two. For the association of the skill certificate and the post skill node, the skill certificate name of the target talent (such as "AWS Certified Solutions Architect") and the name of the post skill node in the graph (such as "AWS Cloud Architecture Design") are converted into word embedding vectors through a word embedding model (such as Word2Vec), and the second cosine similarity of the two is calculated. When the similarity exceeds the above-mentioned first threshold, it is determined that there is an association. For the cases that meet the association conditions, a weighted relationship edge is established between the target talent node and the corresponding post matching project node and post skill node, and the weight of the edge is the corresponding cosine similarity value, to quantify the association strength.

[0092] Secondly, the demand vector is integrated into the first industry knowledge graph as a demand node, and the matched post skill node and post matching project node are linked through vector similarity calculation to obtain a second industry knowledge graph, specifically:

[0093] At the same time, the demand vector of the industry is converted into a demand node and integrated into the first industry knowledge graph, and each demand node contains the feature information of the vector and the corresponding time window identifier. To realize the linking of the demand node and the existing nodes in the graph, the matching is completed through vector similarity calculation: the pre-stored vector representation of all post skill nodes and post matching project nodes in the first industry knowledge graph is extracted (wherein the post skill node vector is generated based on its name and description text through a word embedding model, and the post matching project node vector is generated based on its project description through a sentence embedding model), to form a candidate vector set; the cosine similarity of the demand vector and each vector in the candidate vector set is calculated to obtain the similarity scores of each node; all nodes are sorted in descending order of similarity scores, and the nodes with a score exceeding a preset second threshold (such as 0.65) and ranking in the top 20 are selected as key demand nodes; a weighted relationship edge is established between the demand node and each key demand node, and the weight of the edge is the corresponding similarity score;

[0094] By the above manner, the target talent node, the demand node, and the post skill node, the post matching project node, and the enterprise name node in the first industry knowledge graph are associated and linked, to form a second industry knowledge graph containing multiple types of nodes of talents, demands, skills, projects, and enterprises and multiple dimensional relationship edges. The graph can intuitively present the associated path of talent ability and industry demand in the industry network.

[0095] S04: Based on the second industry knowledge graph, forward attention propagation is performed starting from the target talent node, and reverse attention propagation is performed starting from the demand node in combination with a preset attention mechanism; and an ability-demand coupling matrix is generated by calculating the interaction of attention weights in the forward and reverse propagation paths.

[0096] As a preferred embodiment of the present embodiment, based on the second industry knowledge graph, forward attention propagation is performed starting from the target talent node, and reverse attention propagation is performed starting from the demand node in combination with a preset attention mechanism, specifically:

[0097] When attention propagation is performed based on the second industry knowledge graph, first, initial feature vectors are assigned to all nodes in the graph: the initial feature vector of the target talent node is spliced from the standardized ability element data (such as the feature vector of the historical work project, the coding vector of the skill certificate, etc.) of the target talent node, and the dimension is set to 256; the initial feature vector of the demand node directly uses the corresponding demand vector (the dimension is also 256); and the initial feature vectors of the post skill node, the post matching project node, and the enterprise name node are generated based on the pre-stored description text thereof in the first industry knowledge graph through a pre-training language model (such as RoBERTa), to ensure uniformity of the dimension.

[0098] The forward attention propagation starts from a target talent node: taking each target talent node as a first center node, all neighbor nodes directly connected to it are found through traversing the graph edge relationship, including the talent-associated post skill nodes (such as “Python technology application”), the post matching project nodes (such as “financial risk control system optimization project”) participated in, and the enterprise name nodes (such as “XX financial technology company”) to which the project belongs. The first weight of the first center node to each neighbor node is calculated based on the attention mechanism (which can be Graph Attention Network (GAT)): specifically, the feature vectors of the center node and the neighbor nodes are mapped through a shared linear transformation layer, then processed by a LeakyReLU activation function, and then the weight values of all neighbor nodes are normalized by using a softmax function (formula: α_ij = softmax_j (LeakyReLU (w^T [h_i || h_j])), where h_i is the center node feature, h_j is the neighbor node feature, and w is a learnable parameter), to obtain the normalized first weight of each neighbor node. Then, the feature vectors of all neighbor nodes are weighted and summed according to the first weight to obtain the first aggregated feature, and the aggregated feature is combined with the initial feature vector of the center node through a residual connection (Residual Connection) to update the feature vector of the target talent node. This process is one layer of propagation, and the process is repeated until a preset number of layers (such as 3 layers) to gradually fuse the information of the associated nodes into the feature of the target talent node.

[0099] The reverse attention propagation starts from a demand node and proceeds in the opposite direction: taking the demand node as a second center node, all upstream nodes directly pointing to the node are found through the graph edge relationship, i.e., the post skill nodes (such as “big data analysis”) associated with the demand node, the post matching project nodes (such as “user behavior analysis platform construction project”), and the enterprise name nodes (such as “YY data service company”) corresponding to the project. The same attention mechanism as the forward propagation is used to calculate the second weight of the second center node to each upstream node: the feature vectors of the center node and the upstream nodes are mapped through a linear transformation layer, and then normalized by LeakyReLU activation and softmax to obtain the normalized second weight of each upstream node. The feature vectors of the upstream nodes are weighted and summed based on the second weight to obtain the second aggregated feature, which is also combined with the initial feature vector of the demand node through a residual connection to update the feature vector of the demand node, and the process is repeated until the same preset number of layers as the forward propagation, so that the feature of the demand node is fused with the information of its upstream nodes.

[0100] Further, the ability-demand coupling matrix is generated by calculating the interaction of attention weights in the forward and backward propagation paths, specifically:

[0101] In generating the ability-demand coupling matrix, first, from the forward attention propagation results of completing the preset number of layers of propagation, all post skill nodes and post matching project nodes are extracted as neighbor nodes, and the first weight calculated by the target talent node is taken as the maximum first weight of each node in the propagation process as the ability influence factor (representing the influence strength of talent ability on the node) corresponding to the node; from the reverse attention propagation results, the above nodes are extracted as upstream nodes, and the second weight calculated by the demand node is taken as the maximum second weight of each node as the demand urgency factor (representing the dependence strength of industrial demand on the node);

[0102] Subsequently, the interaction strength value of each post skill node and post matching project node is calculated, that is, the product of the ability influence factor and the demand urgency factor of the node (for example, the ability influence factor of the "Python technology application" node is 0.8, and the demand urgency factor is 0.7, so the interaction strength value is 0.56). Finally, a preset two-dimensional matrix is constructed, where the rows of the matrix correspond to all post skill nodes (sorted by standardized names), and the columns correspond to all post matching project nodes (sorted by project field classification). The interaction strength value of each node is filled into the intersection position of the corresponding row and column of the matrix to form the ability-demand coupling matrix, and the numerical value in the matrix directly reflects the correlation strength of talent ability and industrial demand in each skill and project dimension.

[0103] S05: Perform spectral analysis on the ability-demand coupling matrix, extract the principal eigenvector as the talent industry adaptability index, and generate a professional talent ability evaluation result according to the talent industry adaptability index.

[0104] As a preferred embodiment of the present embodiment, the spectral analysis of the ability-demand coupling matrix extracts the principal eigenvector as the talent industry adaptability index, and generates a professional talent ability evaluation result according to the talent industry adaptability index, specifically:

[0105] In the spectral analysis of the ability-demand coupling matrix, first, the matrix centering process is carried out to eliminate the interference of the numerical value deviation of each element in the matrix on the subsequent analysis results. Specifically, first, the mean of all elements in the ability-demand coupling matrix is calculated, and then each element in the matrix is subtracted from the mean to obtain the centralized matrix; for example, if the mean of the elements in the matrix is 0.4, and the original value of a certain element is 0.6, then the value of the element after centering is 0.2. Through this step, the matrix data is distributed around the origin, and the correlation between dimensions is more accurately reflected.

[0106] After the centering is completed, the covariance matrix of the centering matrix is calculated. The calculation of the covariance matrix takes the rows (representing the post skill node dimension) or columns (representing the post matching project node dimension) of the centering matrix as different variables, and the matrix is constructed by calculating the covariance between any two variables; the specific formula is: for the centering matrix X (dimension m x n, m is the number of post skill nodes, n is the number of post matching project nodes), the element C_ij in the covariance matrix C (dimension m x m or n x n) is [Σ(X_ik - μ_i)(X_jk - μ_j)] / (k-1) (where μ_i and μ_j are the mean of the i-th row and the j-th row, respectively, and k is the number of columns), which can clearly present the linear correlation degree between different post skill dimensions or different post matching project dimensions, and lay a foundation for subsequent feature extraction.

[0107] Then the eigenvalue decomposition is performed on the covariance matrix, and the covariance matrix is decomposed into an eigenvalue matrix and an eigenvector matrix through matrix operation, wherein the eigenvalue matrix is a diagonal matrix, and the elements on the diagonal are the eigenvalues. Each column in the eigenvector matrix corresponds to an eigenvector, and each eigenvector corresponds to an eigenvalue. The size of the eigenvalue reflects how much information the corresponding eigenvector carries in the covariance matrix. The larger the eigenvalue, the more core information the corresponding eigenvector contains. Therefore, the largest eigenvalue is selected from all eigenvalues, and the eigenvector corresponding to the largest eigenvalue is determined as the principal eigenvector. This vector can concentrate the core association characteristics of the ability-demand coupling matrix, and cover the main association information between the post skill and the post matching project dimension.

[0108] Next, the L2 norm of the principal eigenvector is calculated. The L2 norm is calculated by taking the square root of the sum of the squares of all elements in the principal eigenvector. The result can quantify the overall "strength" of the principal eigenvector, and indirectly reflect the matching degree of talent ability and industrial demand in the core dimension. Then the L2 norm value is mapped to a predefined score interval (such as 0-100 points). The mapping process uses linear mapping, for example, if the value range of the L2 norm is 0- (m is the dimension of the eigenvector), when the L2 norm of a principal eigenvector is 0.9 times of , the mapped score in the 0-100 point interval is 90 points. The mapped score is the talent industry adaptation index;

[0109] In generating the professional talent capability evaluation results, the talent industry adaptation index and the actual demand scenario of the industry are combined to set the evaluation level and interpretation standard. For example, if the adaptation index is in the 80-100 interval, it can be classified as a high adaptation level, representing that the target talent's ability is highly matched with the industry demand, and the post skills and project experience they master can fully meet the current and short-term development needs of the industry, and they can be introduced or trained as key talents in the industry; if the index is in the 60-79 interval, it can be classified as a medium adaptation level, indicating that the talent's ability basically meets the industry demand, but there is a small gap in individual post skills or project experience, which can be improved by targeted training (such as supplementing certain skill certification or participating in specific project practice); if the index is less than 60, it can be classified as a low adaptation level, indicating that there is a significant gap between the talent's ability and the industry demand, and the ability structure needs to be systematically optimized in combination with the industry demand direction. At the same time, specific ability gap analysis needs to be attached in the evaluation results, such as pointing out that "the interaction intensity value of a certain post skill dimension is low, and the learning of this skill needs to be strengthened", to provide clear guidance for talent development and industry talent allocation.

[0110] In summary, the present application first constructs a unified knowledge graph by vectorizing and fusing the target talent data and industry recruitment information, solving the technical problem of difficult unified processing and correlation analysis of multi-source heterogeneous data, and establishing a standardized data foundation for in-depth analysis. Subsequently, a bidirectional attention propagation mechanism is executed on the constructed second industry knowledge graph, which enables the system to simultaneously simulate the radiation influence path of talent ability in the industry environment (forward propagation) and the traceability positioning path of industry demand on talent ability (reverse propagation), thereby dynamically capturing the complex and nonlinear interaction relationship between talent and demand. Finally, by quantifying the interaction intensity of the bidirectional propagation path as a coupling matrix and performing spectral analysis, the most core structural features can be extracted as the adaptation index from the complex network relationship, which makes the evaluation results no longer dependent on the surface static matching, but based on the deep and quantitative indicators dynamically calculated from the entire industry ecological network, significantly improving the precision, dynamics and forward-looking of the talent capability evaluation, effectively solving the problem of inaccurate and inefficient evaluation of talent's ability in the prior art.

[0111] Embodiment Two: Please refer to Figure 2 The professional talent capability evaluation system based on industry big data provided in the embodiments of the present application.

[0112] In this embodiment, the professional talent capability evaluation system based on industry big data includes an acquisition module 10, a construction module 20, a linking module 30, a calculation module 40 and a result output module 50.

[0113] The acquisition module 10 is configured to acquire standardized ability element data of target talents and demand vectors of industries, wherein the standardized ability element data comprises historical work projects, a list of skill certificates and an educational background, and the demand vectors are generated by performing natural language processing and vectorization fusion on job responsibilities and service requirements in enterprise recruitment information in a preset time window;

[0114] The construction module 20 is configured to construct a first industrial knowledge graph based on acquired job skills, job matching projects and enterprise names from historical industrial data, wherein nodes of the first industrial knowledge graph comprise job skill nodes, job matching project nodes and enterprise name nodes, and edges comprise co-occurrence relationships or membership relationships between the nodes;

[0115] The linking module 30 is configured to integrate the standardized ability element data of the target talents as target talent nodes into the first industrial knowledge graph, and link the target talent nodes with the job matching project nodes and the job skill nodes to which the target talent nodes belong based on a preset semantic matching method; and integrate the demand vectors as demand nodes into the first industrial knowledge graph, and link the demand nodes with the job skill nodes and the job matching project nodes matched by vector similarity calculation, to obtain a second industrial knowledge graph.

[0116] The calculation module 40 is configured to perform forward attention propagation starting from the target talent nodes and perform reverse attention propagation starting from the demand nodes based on the second industrial knowledge graph and in combination with a preset attention mechanism, and generate an ability-demand coupling matrix by calculating the interaction of attention weights in the forward and reverse propagation paths.

[0117] The result output module 50 is configured to perform spectral analysis on the ability-demand coupling matrix, extract a principal feature vector as a talent-industry adaptability index, and generate a professional talent ability evaluation result according to the talent-industry adaptability index.

[0118] For the convenience and brevity of description, the system embodiments of the present application include all the implementation manners of the above-described professional talent ability evaluation method embodiments based on industrial big data, which will not be described herein.

[0119] The above-described specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are only specific embodiments of the present application and are not used to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1.A professional talent capability evaluation method based on industrial big data, characterized in that, The method comprises the following steps: obtaining standardized ability element data of target talents and demand vectors of an industry; wherein the standardized ability element data comprises historical work projects, a list of skill certificates and an educational background; the demand vectors are generated by natural language processing and vectorization fusion of job responsibilities and job requirements in enterprise recruitment information within a preset time window; constructing a first industry knowledge graph based on acquired job skills, job matching projects and enterprise names from historical industry data; wherein nodes of the first industry knowledge graph comprise job skill nodes, job matching project nodes and enterprise name nodes, and edges comprise co-occurrence relationships or membership relationships between the nodes; integrating the standardized ability element data of the target talents into the first industry knowledge graph as target talent nodes, and linking the target talent nodes with the job matching project nodes and the job skill nodes to which the target talent nodes belong based on a preset semantic matching method; integrating the demand vectors into the first industry knowledge graph as demand nodes, and linking the demand nodes with the job skill nodes and the job matching project nodes matched through vector similarity calculation, to obtain a second industry knowledge graph; based on the second industry knowledge graph, combining a preset attention mechanism, and performing forward attention propagation from the target talent nodes and reverse attention propagation from the demand nodes; and generating an ability-demand coupling matrix by calculating the interaction of attention weights in the forward and reverse propagation paths; performing spectral analysis on the ability-demand coupling matrix, extracting a principal eigenvector as a talent industry adaptability index, and generating a professional talent ability evaluation result according to the talent industry adaptability index. 2.The industrial big data-based professional talent capability evaluation method according to claim 1, characterized in that, The demand vectors are generated by natural language processing and vectorization fusion of job responsibilities and job requirements in enterprise recruitment information within a preset time window, specifically as follows: performing text preprocessing on the job responsibilities and job requirements to obtain standardized text; the preprocessing comprises removing meaningless symbols, filtering stop words and text segmentation; based on a term frequency-inverse document frequency method, processing the standardized text to obtain various keywords, and classifying the various keywords into skill keywords and project keywords; converting the skill keywords and the project keywords into corresponding skill vectors and project vectors through a preset bag-of-words model respectively; based on a preset time window, performing weighted average pooling processing on all skill vectors and project vectors under the same recruitment information to generate the demand vectors. 3.The industrial big data-based professional talent capability evaluation method according to claim 2, characterized in that, The demand vectors are generated by performing weighted average pooling processing on all skill vectors and project vectors under the same recruitment information based on a preset time window, specifically as follows: counting the frequencies of the skill vectors and the project vectors in all recruitment information within the preset time window; calculating inverse document frequency values of the various skill vectors and the project vectors according to the frequencies, and performing normalization processing on the inverse document frequency values based on a softmax function, taking the normalization processing results as weights of the various vectors; and According to the weight of each vector, all skill vectors and project vectors under the same recruitment information are weighted average pooling processing to obtain the demand vector. 4.The industrial big data-based professional talent capability evaluation method according to claim 1, characterized in that, The first industrial knowledge graph is constructed based on the job skills, job matching projects and job enterprise names obtained from historical industrial data, specifically: The standardized skill names, completed project names and enterprise full names are extracted from the historical industrial data through regular expression and dictionary matching; The skill names, completed project names and enterprise full names are normalized to obtain various job skill nodes, job matching project nodes and enterprise name nodes; Based on each node, according to the co-occurrence relationship of skills and projects often appearing in the same recruitment demand, edges between job skill nodes and job matching project nodes are constructed, and according to the membership relationship of projects being led or completed by enterprises, edges between job matching project nodes and enterprise name nodes are established; Based on each edge and each node, the first industrial knowledge graph is constructed. 5.The industrial big data-based professional talent capability evaluation method according to claim 1, characterized in that, The various target talent nodes are linked with the corresponding job matching project nodes and job skill nodes based on the preset semantic matching method, specifically: The historical work project description of the target talent and the description text of the job matching project node are respectively converted into sentence embedding vectors; The first cosine similarity between the sentence embedding vector of the historical work project description and each job matching project node description vector is calculated; The skill certificate name of the target talent and the name of the job skill node in the graph are respectively converted into word embedding vectors; The second cosine similarity between the word embedding vector of the skill certificate name and each job skill node name vector is calculated; If the first cosine similarity and the second cosine similarity exceed a preset first threshold, a relationship edge is established between the target talent node and the corresponding graph node. 6.The industrial big data-based professional talent capability evaluation method according to claim 1, characterized in that, The job skill nodes and job matching project nodes matched through vector similarity calculation are linked, specifically: From the first industrial knowledge graph, the pre-stored vector representation of all job skill nodes and job matching project nodes is obtained to form a candidate vector set; The cosine similarity between the demand vector and each vector in the candidate vector set is calculated to obtain a similarity score; According to the similarity score, all job skill nodes and job matching project nodes are ranked in descending order; The nodes ranked in the first preset number of positions and having a similarity score exceeding a preset second threshold are selected as key demand nodes, and relationship edges are established between the demand node and each selected key demand node. 7.The industrial big data-based professional talent capability evaluation method according to claim 1, characterized in that, Based on the second industrial knowledge graph, forward attention propagation is performed from the target talent node, and reverse attention propagation is performed from the demand node based on a preset attention mechanism, specifically: Each node in the second industrial knowledge graph is assigned a feature vector; Each target talent node is taken as a first center node, all neighbor nodes directly connected to the first center node through edges are found, and the neighbor nodes include job skill nodes, job matching project nodes and enterprise name nodes; The first weights of each first center node to each neighbor node are calculated based on an attention mechanism by calculating the correlation degree of the first center node and the feature vector of each neighbor node. The demand node is taken as a second center node, and an upstream node directly pointed to the second center node by an edge is found, including a post skill node, a post matching project node, and an enterprise name node. The second weights of the second center node to each upstream node are calculated based on an attention mechanism by calculating the correlation degree of the second center node and the feature vector of each upstream node. The forward and reverse propagation of the attention mechanism are completed until a preset number of layers is calculated. 8.The industrial big data-based professional talent capability evaluation method according to claim 1, characterized in that, The interaction of the attention weights in the forward and reverse propagation paths is calculated to generate a capability-demand coupling matrix, specifically as follows: Based on the second industrial knowledge graph, the post skill node and the post matching project node are extracted as neighbor nodes from the final calculation result of the forward attention propagation, and each first weight calculated by the target talent node as a first center node is taken as a capability influence factor. The post skill node and the post matching project node are extracted as upstream nodes from the final calculation result of the reverse attention propagation, and each second weight calculated by the demand node as each second center node is taken as a demand urgency factor. The capability influence factor and the demand urgency factor of each post skill node and post matching project node are multiplied to obtain an interaction intensity value of the post skill node and the post matching project node. The interaction intensity value of each post skill node and post matching project node is filled into a preset two-dimensional matrix at the intersection position corresponding to the row of the post skill node and the column of the post matching project node to generate a capability-demand coupling matrix; one dimension of the two-dimensional matrix represents all post skill nodes, and the other dimension represents all post matching project nodes. 9.The industrial big data-based professional talent capability evaluation method of claim 1, wherein, The capability-demand coupling matrix is subjected to spectral analysis, and a main eigenvector is extracted as a talent industry adaptation degree index, and a professional talent capability evaluation result is generated according to the talent industry adaptation degree index, specifically as follows: The capability-demand coupling matrix is subjected to centering processing. The covariance matrix of the centering-processed capability-demand coupling matrix is calculated, and the covariance matrix is subjected to eigenvalue decomposition to obtain each eigenvalue and a corresponding eigenvector. The eigenvalue with the maximum modulus is selected from each eigenvalue, and the eigenvector corresponding to the eigenvalue with the maximum modulus is taken as a main eigenvector. An L2 norm value of the principal eigenvector is calculated, and the L2 norm value is mapped to a predefined score interval, and the mapped result is taken as the talent industry adaptability index. 10.A professional talent capability evaluation system based on industrial big data, characterized in that, Comprise: An acquisition module is configured to acquire standardized ability element data of a target talent and a demand vector of an industry; the standardized ability element data includes historical work projects, a list of skill certificates, and an educational background; the demand vector is generated by natural language processing and vectorization fusion on job responsibilities and job requirements in enterprise recruitment information within a preset time window; A construction module is configured to construct a first industry knowledge graph based on acquired job skills, job matching projects, and job enterprise names from historical industry data; nodes of the first industry knowledge graph include job skill nodes, job matching project nodes, and enterprise name nodes, and edges include co-occurrence relationships or membership relationships between the nodes; A linking module is configured to integrate the standardized ability element data of the target talent as each target talent node into the first industry knowledge graph, and link the each target talent node with the job matching project node and the job skill node based on a preset semantic matching method; the demand vector is integrated as each demand node into the first industry knowledge graph, and the job skill node and the job matching project node matched through vector similarity calculation are linked to obtain a second industry knowledge graph; A calculation module is configured to start forward attention propagation from the target talent node and start reverse attention propagation from the demand node based on the second industry knowledge graph and in combination with a preset attention mechanism; an ability-demand coupling matrix is generated by calculating the interaction of attention weights in the forward and reverse propagation paths; A result output module is configured to perform spectral analysis on the ability-demand coupling matrix, extract a principal eigenvector as a talent industry adaptability index, and generate a professional talent ability evaluation result according to the talent industry adaptability index.

Images