Personnel post matching method, device, equipment and medium

Abstract

The invention relates to the technical field of data analysis, and discloses a personnel post matching method and device, equipment and a medium, and the method comprises the steps: obtaining multi-mode resume data of a plurality of target candidates and post requirement data of a plurality of posts, and constructing a post requirement weight matrix according to the post requirement data; converting the multi-modal resume data into candidate feature vectors, converting the post requirement data into post feature vectors, and performing weighted fusion on the candidate feature vectors according to the post requirement weight matrix to obtain candidate weighted vectors; calculating the matching degree between the candidate weighted vector and the post feature vector; carrying out association relation feature extraction on the multi-mode resume data to obtain data association features; and generating a target personnel post matching result according to the data association features and the matching degree. According to the method, the data association features are extracted, the limitation that only a single attribute is concerned in a traditional feature extraction scheme is broken through, and meanwhile the matching accuracy of personnel and posts is improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to methods, devices, equipment and media for matching personnel to positions. Background Technology

[0002] With the continued advancement of economic globalization and the deepening of industrial restructuring, enterprises' demand for talent is becoming increasingly diversified and specialized. Achieving efficient and accurate matching of personnel to positions has become a core common demand of both enterprises and job seekers in the field of human resource management.

[0003] Current personnel-job matching solutions mostly employ manual screening and simple keyword matching. Manual screening requires recruiters to review a large number of resumes one by one and manually judge the degree of match between candidates and positions. However, this method is not only time-consuming and labor-intensive, but also difficult to avoid personal subjective bias, and cannot guarantee the objectivity and consistency of the evaluation results. The simple keyword matching method judges the degree of match by matching keywords in the resume with keywords in the job requirements. However, this matching method cannot deeply understand the candidate's true abilities and the deeper needs of the position, resulting in insufficient accuracy of the matching results.

[0004] Therefore, in the face of the ever-increasing demand for personnel-job matching, the current personnel-job matching methods urgently need to be improved in order to solve the problems of low efficiency and insufficient accuracy of matching results of existing methods. Summary of the Invention

[0005] This invention provides a method, apparatus, equipment, and medium for matching personnel to positions, which mainly addresses the limitations of traditional features that focus on a single attribute while improving the accuracy of matching personnel to positions.

[0006] Firstly, a method for matching personnel to job positions is provided, including: Obtain multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data; The multimodal resume data is converted into candidate feature vectors, and the job requirement data is converted into job feature vectors. The candidate feature vectors are then weighted and fused according to the job requirement weight matrix to obtain a candidate weighted vector. Calculate the matching degree between the candidate weighted vector and the job feature vector; The association features of the multimodal resume data are extracted to obtain data association features; The target personnel job matching results are generated based on the data association features and the matching degree.

[0007] Secondly, a personnel job matching device is provided, comprising: The matrix construction module is used to acquire multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data. The transformation and fusion module is used to transform the multimodal resume data into candidate feature vectors and the job requirement data into job feature vectors, and to perform weighted fusion of the candidate feature vectors according to the job requirement weight matrix to obtain a candidate weighted vector. The matching degree calculation module is used to calculate the matching degree between the candidate weighted vector and the job feature vector; The data association feature extraction module is used to extract association relationship features from the multimodal resume data to obtain data association features; The personnel job result generation module is used to generate target personnel job matching results based on the data association features and the matching degree.

[0008] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned personnel job matching method.

[0009] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned personnel job matching method.

[0010] The aforementioned personnel-job matching method, device, computer equipment, and storage medium achieve the following: By acquiring multimodal resume data and job requirement data, it covers multi-dimensional candidate information and complete job requirements, thereby avoiding matching bias caused by a single data dimension. By constructing a job requirement weight matrix, it highlights the key dimensions of the job, filters irrelevant or secondary features, improves matching targeting, and thus improves the accuracy of matching results. By weighted fusion of candidate feature vectors, it integrates the differences in importance of multi-dimensional candidate features, avoiding excessive influence of a single feature on the results. By calculating the matching degree, the matching basis can be traced subsequently, improving matching credibility. By extracting relational features from multimodal resume data, it breaks through the limitation of traditional features focusing only on a single attribute, making features more in-depth and improving matching accuracy. Combining the matching degree and deep relational features to generate target personnel-job matching results ensures the intuitiveness of the matching results while taking into account complex relationships, avoiding the problem of high scores but poor actual suitability. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of an application environment for a personnel job matching method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a personnel job matching method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a personnel job matching device according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device according to one embodiment of the present invention; Figure 5 This is another structural schematic diagram of a computer device according to one embodiment of the present invention.

[0013] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] The personnel job matching method provided in this embodiment of the invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server can first collect multimodal resumes and job requirement data, construct a job requirement weight matrix, convert the two types of data into feature vectors and calculate the matching degree using weighted averages, extract the correlation features of the resume data, and combine the matching degree and correlation features to generate job matching results, which are then fed back to the client. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0016] Please see Figure 2 As shown, Figure 2A flowchart illustrating a personnel job matching method provided in an embodiment of the present invention includes the following steps: S1. Obtain multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data.

[0017] In this embodiment of the invention, the multimodal resume data includes candidate basic information (such as age, education, and years of work experience), resume information (such as skills and certifications), job description text (such as previous employers and scope of work), and project experience descriptions; the job requirement data includes basic job requirements (such as salary range, education requirements, years of experience, and skills requirements) and job descriptions (job responsibilities), which include project delivery efficiency, collaborative project partners, and company and team background (such as basic company introduction, team size and structure, industry and business scope).

[0018] In this embodiment of the invention, constructing a job requirement weight matrix based on the job requirement data includes: Identify the multidimensional scene features of the job requirement data; Construct a basic weight matrix based on the multi-dimensional scene features; The basic weight matrix is ​​dynamically calibrated to obtain the job requirement weight matrix.

[0019] In this embodiment of the invention, the multi-dimensional scene features include industry category features, job type features, enterprise size features, etc.

[0020] Specifically, natural language processing technology is used to extract keywords from descriptions such as "industry field" and "business scope" in job requirement data to identify industry category characteristics; keywords are extracted from descriptions such as "job responsibility description" in job requirement data to identify job type characteristics; keywords are extracted from descriptions such as "company and team background" in job requirement data; and industry category characteristics, job type characteristics, and company size characteristics are combined to form multi-dimensional scenario characteristics.

[0021] Furthermore, in the process of constructing the basic weight matrix based on multi-dimensional scenario features, the core dimensions directly related to candidate evaluation are first selected from the job type features in the multi-dimensional scenario features. The core dimensions include educational requirements, years of experience, etc. An initial weight value is assigned to each core dimension, and the core dimensions and their corresponding initial weight values ​​constitute the basic weight matrix.

[0022] Furthermore, in the process of dynamically calibrating the basic weight matrix, adjustment factors are first extracted from multi-dimensional scene features. These adjustment factors include industry adjustment factors (extracted from industry category features) and scene adjustment factors (extracted from enterprise size features). The value of each adjustment factor is preset based on historical experience. The basic weight matrix and the adjustment factor values ​​are then multiplied element-wise to obtain the preliminary job demand weight matrix. Subsequently, the preliminary job demand weight matrix is ​​optimized using the gradient descent algorithm to obtain the final job demand weight matrix.

[0023] For example, in the basic weight matrix, the weight of education requirement is 0.3 and the weight of years of experience is 0.7; at the same time, the industry adjustment factor is preset to 1.2 and the scenario adjustment factor is preset to 1.1, thus forming a preliminary dynamic weight matrix: education requirement is 0.39 and years of experience is 0.924.

[0024] S2. The multimodal resume data is converted into candidate feature vectors, and the job requirement data is converted into job feature vectors. The candidate feature vectors are then weighted and fused according to the job requirement weight matrix to obtain a weighted candidate vector.

[0025] In this embodiment of the invention, converting the multimodal resume data into candidate feature vectors includes: The multimodal resume data is standardized to obtain standardized resume data, and the modality type features corresponding to the standardized resume data are extracted. Convert the text data in the standardized resume data into modal high-dimensional semantic vectors; A type feature space is constructed based on the modality type features, and the high-dimensional semantic vector of the modality is semantically mapped based on the type feature space to obtain the candidate feature vector.

[0026] In this embodiment of the invention, during the standardization process of multimodal resume data, the multimodal resume data is first deduplicated to remove redundant information. The text data in the multimodal resume data after removing redundancy and the text data in the job requirement data are then unified into UTF-8 encoding and PDF / Word format. Subsequently, the multimodal resume data with unified encoding format is processed for missing values ​​and outliers are detected. Finally, the multimodal resume data that has undergone missing value processing and outlier detection is used as standardized resume data.

[0027] Furthermore, the standardized resume data is segmented into words, and then part-of-speech tagging is performed on the segmented standardized resume data using natural language processing. Subsequently, based on the tagged parts of speech, the key entities of both are identified from the segmented standardized resume data using a BERT pre-trained model. Finally, based on the identified key entities, the textual data in the standardized resume data is converted into modal high-dimensional semantic vectors using word embedding technology.

[0028] Furthermore, the modal type features include text modal type features, numerical modal type features, etc., and the job modal type features include explicit requirement features, implicit requirement features, etc.

[0029] Specifically, core semantic features are extracted from textual data (such as project experience descriptions) in standardized resume data using natural language processing technology. These core semantic features are used as text modality features. Core numerical features such as age and years of work experience are extracted from standardized resume data as numerical modality features. The text modality features and numerical modality features are then combined to form modality features.

[0030] Furthermore, in the process of constructing the type feature space based on modal type features, each modal type feature is first assigned a unique dimension identifier by dictionary encoding. The total dimension of the type feature space can be determined based on the dimension identifier, so that each modal type feature corresponds to an independent dimension in the type feature space. Then, each modal type feature is assigned a weight according to a preset weighting rule, and the weighted features and corresponding dimensions are used to construct the type feature space.

[0031] For example, modal type features include age (numerical modal type feature), core semantics - project experience description (textual modal type feature), and years of work experience (numerical modal type feature); assign dimension label "Dimension 1" to "Age", dimension label "Dimension 2" to "Core semantics - Project Management", and dimension label "Dimension 3" to "Years of Work Experience", determining the total dimension of the type feature space to be 3; according to the preset weighting rules, the weights of each dimension are distributed as follows: Dimension 1 (Age) is 0.3, Dimension 2 (Core semantics - Project experience description) is 0.4, and Dimension 3 (Years of Work Experience) is 0.3, forming a 3-dimensional space model.

[0032] Furthermore, based on the number of dimensions in the type feature space, the number of dimensions of the modal high-dimensional semantic vector is aligned with the number of dimensions in the type feature space according to the principle of supplementing less and deleting more, to ensure that the number of dimensions of the high-dimensional semantic vector is consistent with the total dimension of the type feature space. Then, each dimension of the aligned high-dimensional semantic vector is multiplied by the weight of the corresponding dimension in the type feature space, and the result is used as the value of each dimension of the aligned high-dimensional semantic vector. Finally, the high-dimensional semantic vector with adjusted dimension values ​​is used as the candidate feature vector according to the dimensional order of the type feature space.

[0033] In this embodiment of the invention, the steps of converting the job requirement data into a job feature vector are the same as those of converting the multimodal resume data into a candidate feature vector, and will not be described in detail here.

[0034] In this embodiment of the invention, the step of weighting and fusing the candidate feature vectors according to the job requirement weight matrix to obtain a candidate weighted vector includes: Establish a dimensional mapping relationship between the job requirement weight matrix and the candidate feature vector; The candidate feature vectors are weighted according to the dimensional mapping relationship to obtain weighted candidate feature vectors. The weighted candidate feature vectors are normalized to obtain the candidate weighted vectors.

[0035] Furthermore, in the process of establishing the mapping relationship between the job requirement weight matrix and the corresponding dimensions of the candidate feature vector, the weight parameters of each core dimension in the job requirement weight matrix are bound one by one to the dimension values ​​of the same core dimension in the candidate feature vector.

[0036] Furthermore, in the process of weighting the candidate feature vectors according to the mapping relationship, the dimension values ​​of the candidate feature vectors that are the same as the core dimension are multiplied by the weight parameters of the corresponding dimension in the job requirement weight matrix according to the binding relationship. After calculating the dimension values ​​of all the same dimensions in sequence, the weighted candidate feature vector is obtained.

[0037] Furthermore, the weighted candidate feature vector is normalized using a normalization algorithm (such as Min-Max). The normalized values ​​of the dimensions in the weighted candidate feature vector are integrated to form the candidate weighted vector, eliminating the dimensional differences of the original quantized values ​​of different dimensions and making the values ​​of each dimension of the candidate weighted vector fall within the same numerical range, which facilitates the subsequent matching degree calculation.

[0038] S3. Calculate the matching degree between the candidate weighted vector and the job feature vector.

[0039] In this embodiment of the invention, calculating the matching degree between the candidate weighted vector and the job feature vector includes: Deep semantic encoding is performed on the candidate weighted vector and the job feature vector respectively to obtain the candidate encoding vector and the job encoding vector; Calculate the similarity between the candidate coding vector and the job coding vector; The similarity is normalized, and the normalized similarity is converted into a matching score.

[0040] In this embodiment of the invention, during the deep semantic encoding of the candidate weighted vector, the candidate weighted vector is linearly transformed by a fully connected layer in the dual-tower neural network architecture to obtain a linearly transformed candidate weighted vector. Then, a nonlinear activation function layer in the dual-tower neural network architecture performs nonlinear mapping on the linearly transformed candidate weighted vector to obtain a nonlinearly activated candidate weighted vector. Next, a normalization layer in the dual-tower neural network architecture normalizes the nonlinearly activated candidate weighted vector to obtain a normalized candidate weighted vector. Finally, a fully connected layer in the dual-tower neural network architecture performs dimensionality compression on the normalized candidate weighted vector to obtain a candidate encoding vector. The steps for deep semantic encoding the job feature vector to obtain a job encoding vector are the same as those for deep semantic encoding the candidate weighted vector to obtain a candidate encoding vector, and will not be elaborated here.

[0041] Furthermore, the similarity between the candidate's coding vector and the job's coding vector is calculated using a similarity algorithm such as cosine similarity. During the normalization of the similarity, a standardization algorithm (such as Min-Max normalization) is used to map the similarity to the [0,1] interval to obtain the normalized similarity. The normalized similarity is then mapped according to a preset rule to obtain the matching score. For example, if the normalized similarity is 0.85, a linear mapping yields 85 points (matching score), ensuring that the score is intuitive and quantifiable for comparison.

[0042] S4. Extract association features from the multimodal resume data to obtain data association features.

[0043] In this embodiment of the invention, the step of extracting association features from the multimodal resume data to obtain data association features includes: Extract basic candidate employment information from the multimodal resume data, and extract job requirement information from the job requirement data; Construct a set of multiple node types based on the candidate's basic employment information and the job requirements information; Determine the skill-related edge set and the company-related edge set based on the multi-type node set; Construct a heterogeneous graph based on the set of multiple types of nodes, the set of skill-related edges, and the set of company-related edges; The heterogeneous graph is divided into hierarchical architectures, and a topological relationship graph is built based on the divided architectures. Each node in the topological graph is transformed into an initial feature vector, resulting in a set of initial feature vectors for each node. For each node in the initial feature vector set, perform high-order correlation feature fusion to obtain the correlation fusion vector of each node; The associated fusion vectors are concatenated to form data association features.

[0044] In this embodiment of the invention, during the process of constructing a heterogeneous graph based on multimodal resume data, firstly, attribute information such as "education level and years of work experience" is extracted from the multimodal resume data to form candidate nodes; attribute information such as "education requirements and job responsibilities" is extracted from the job requirement data to form job nodes; attribute information such as "skills mastered" is extracted from the multimodal resume data, and attribute information such as "skill requirements" is extracted from the job requirement data to form skill nodes (including skill nodes mastered by candidates and skill nodes required by the job); attribute information such as "previous employers" is extracted from the multimodal resume data, and attribute information such as "basic company introduction" is extracted from the job requirement data to form company nodes (including company nodes of candidates' previous employers and company nodes of the company to which the job belongs).

[0045] Furthermore, edge relationships are defined between candidate nodes and the skill nodes possessed by the candidate, forming candidate-skill edges; edge relationships are defined between job nodes and the skill nodes required for the job, forming job-skill edges; edge relationships are defined between candidate nodes and the company nodes of the candidate's previous employers, forming candidate-company edges; edge relationships are defined between job nodes and the company nodes of the company to which the job belongs, forming job-company edges; the above nodes and edge relationships are combined in a "node-edge-node" manner to form a heterogeneous graph.

[0046] Furthermore, in the process of constructing a topological relationship graph based on the node relationship types in the heterogeneous graph, candidate-skill edges and job-skill edges are first classified into skill associations, and candidate-company edges and job-company edges are classified into company associations. The skill associations and company associations are used as the upper-level architecture of the topological relationship graph. Then, the set of node pairs corresponding to candidate-skill edges and job-skill edges in the skill associations and the set of node pairs corresponding to candidate-company edges and job-company edges in the company associations are used as the lower-level architecture of the topological relationship graph. The upper-level architecture and the lower-level architecture are combined to construct the topological relationship graph.

[0047] Furthermore, in the process of extracting high-order relationship features of nodes from the multimodal resume data based on the topological relationship graph, each node in the topological relationship graph is transformed into an initial feature vector. The initial feature vector is then fused with the first-order (e.g., candidate-skill, candidate-company), second-order (e.g., candidate-skill-position), and higher-order association features of the nodes through a graph neural network model (e.g., GraphSAGE / GAT model). After multiple rounds of fusion, the embedding vector corresponding to each node is output, which is the data association feature.

[0048] For example, the candidate node (e.g., Bachelor's degree = 0.8, 3 years of experience = 0.7) is transformed into an initial feature vector [0.8 (education level), 0.7 (work experience)]. The candidate node is then fused with its first-order association feature, which is a Java programming skills node with a feature vector of 0.9, and the company A node with a feature vector of 0.8. The fused vector is [0.8, 0.7, 0.9, 0.8].

[0049] In this embodiment of the invention, a heterogeneous graph is constructed to visualize the relationship between candidates, skills, positions, and companies using nodes / edges, breaking down data dispersion barriers. By building a topological relationship graph, the complexity of extracting high-order features is reduced. By extracting high-order relationship features of nodes from multimodal resume data, node attributes and high-order relationships are integrated to uncover hidden relationships and improve the accuracy of subsequent matching.

[0050] S5. Generate the target personnel job matching result based on the data association features and the matching degree.

[0051] In this embodiment of the invention, generating the target personnel job matching result based on the data association features and the matching degree includes: The candidate coding vector and the job coding vector corresponding to the matching degree are concatenated into a fusion vector; Based on the data association features, the fused vector is projected into the same vector space to obtain the projected feature vector; Attention fusion is performed on the data association features and the projected feature vector to obtain comprehensive features; Obtain the multi-target matching dimensions and assign weights to each target dimension in the multi-target matching dimensions to obtain the target dimension weights for each target dimension; The comprehensive features are weighted and fused with the target dimension weights to obtain a multi-target comprehensive matching value; The initial personnel job matching scheme is determined based on the multi-target comprehensive matching value and the preset target matching threshold. The multi-objective comprehensive matching value is updated based on the preset historical candidate target dimension weights to obtain the updated multi-objective comprehensive matching value; The initial personnel job matching scheme is adjusted based on the updated multi-objective comprehensive matching value to obtain the target personnel job matching result.

[0052] In this embodiment of the invention, during the feature fusion process of the candidate encoding vector and job encoding vector corresponding to the matching degree with the data association features, the candidate encoding vector and job encoding vector are first concatenated into a fusion vector. Then, the fusion vector is projected into the same vector space as the data association features through a fully connected neural network layer to output the matching feature vector. Finally, the matching feature vector and the data association features are fused by an attention network to obtain the comprehensive matching features.

[0053] Furthermore, in the process of multi-objective decision analysis of comprehensive matching features, firstly, multi-objective dimensions are set, including skill matching degree objective and work experience matching degree objective, and weights are assigned to each objective dimension. The weight of each objective dimension is multiplied by the feature vector corresponding to the objective dimension in the comprehensive matching features to obtain the objective dimension score. All objective dimension scores are summarized to obtain the multi-objective comprehensive score. The multi-objective comprehensive score and the scores of each objective dimension are combined into objective decision features. The multi-objective comprehensive scores in the objective decision features are sorted in descending order, and candidates with multi-objective comprehensive scores ≥ a preset threshold (such as 0.7) are selected to generate a preliminary personnel job matching scheme.

[0054] Furthermore, the multimodal resume data of candidates in the preliminary personnel-job matching scheme are compared with the basic job requirements in the job requirement data to check whether the candidates meet the basic job requirements. Candidates who do not meet the requirements are eliminated, and the remaining candidates are obtained.

[0055] Furthermore, historical personnel job matching data is collected, and the actual weight values ​​of the target dimensions of historical candidate data in the historical personnel job matching data are analyzed through machine learning models. The weights of each target dimension of the retained candidates are updated based on the actual weight values. The target dimension scores are recalculated based on the updated weights, and the target dimension scores are summarized to obtain the updated multi-target comprehensive score. The updated multi-target comprehensive scores are then sorted to generate the target personnel job matching results.

[0056] As can be seen, in the above solution, for the business of constructing target personnel job matching results, multimodal resume data of several target candidates and job requirement data of several positions are obtained, and a job requirement weight matrix is ​​constructed based on the job requirement data; the multimodal resume data is converted into candidate feature vectors, and the job requirement data is converted into job feature vectors, and the candidate feature vectors are weighted and fused according to the job requirement weight matrix to obtain a candidate weighted vector; the matching degree between the candidate weighted vector and the job feature vector is calculated; correlation features are extracted from the multimodal resume data to obtain data correlation features; and the target personnel job matching results are generated based on the data correlation features and the matching degree. By extracting data correlation features, the solution overcomes the limitation of traditional feature extraction schemes that only focus on a single attribute, while also improving the accuracy of personnel and job matching.

[0057] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0058] In one embodiment, a personnel job matching device is provided, which corresponds one-to-one with a personnel job matching method described in the above embodiments. For example... Figure 3 As shown, this personnel job matching device includes a matrix construction module 101, a transformation and fusion module 102, a matching degree calculation module 103, a data association feature extraction module 104, and a personnel job result generation module 105. Detailed descriptions of each functional module are as follows: The matrix construction module 101 is used to acquire multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data. The transformation and fusion module 102 is used to transform the multimodal resume data into candidate feature vectors and the job requirement data into job feature vectors, and to perform weighted fusion of the candidate feature vectors according to the job requirement weight matrix to obtain a candidate weighted vector. The matching degree calculation module 103 is used to calculate the matching degree between the candidate weighted vector and the job feature vector; The data association feature extraction module 104 is used to extract association relationship features from the multimodal resume data to obtain data association features; The personnel job result generation module 105 is used to generate target personnel job matching results based on the data association features and the matching degree.

[0059] In one embodiment, the matrix construction module 101, when constructing a job requirement weight matrix based on the job requirement data, is used to: Extract job scenario description data from the job requirement data, and perform scenario feature extraction on the job scenario description data to obtain job scenario features; Filter the candidate job evaluation dimensions from the job scenario features, and assign weights to the candidate job evaluation dimensions to obtain a basic weight matrix; The basic weight matrix is ​​adjusted according to the job scenario characteristics to obtain the job requirement weight matrix.

[0060] In one embodiment, the conversion and fusion module 102, when converting the multimodal resume data into candidate feature vectors, is used to: The multimodal resume data is standardized to obtain standardized resume data, and the modality type features corresponding to the standardized resume data are extracted. Convert the text data in the standardized resume data into modal high-dimensional semantic vectors; A type feature space is constructed based on the modality type features, and the high-dimensional semantic vector of the modality is semantically mapped based on the type feature space to obtain the candidate feature vector.

[0061] In one embodiment, when the conversion and fusion module 102 performs weighted fusion of the candidate feature vectors according to the job requirement weight matrix to obtain a candidate weighted vector, it is used to: Establish a dimensional mapping relationship between the job requirement weight matrix and the candidate feature vector; The candidate feature vectors are weighted according to the dimensional mapping relationship to obtain weighted candidate feature vectors. The weighted candidate feature vectors are normalized to obtain the candidate weighted vectors.

[0062] In one embodiment, the matching degree calculation module 103, when calculating the matching degree between the candidate weighted vector and the job feature vector, is used to: Deep semantic encoding is performed on the candidate weighted vector and the job feature vector respectively to obtain the candidate encoding vector and the job encoding vector; Calculate the similarity between the candidate coding vector and the job coding vector; The similarity is normalized, and the normalized similarity is converted into a matching score.

[0063] In one embodiment, the data association feature extraction module 104, when extracting association features from the multimodal resume data to obtain data association features, is used to: Extract basic candidate employment information from the multimodal resume data, and extract job requirement information from the job requirement data; Construct a set of multiple node types based on the candidate's basic employment information and the job requirements information; Determine the skill-related edge set and the company-related edge set based on the multi-type node set; Construct a heterogeneous graph based on the set of multiple types of nodes, the set of skill-related edges, and the set of company-related edges; The heterogeneous graph is divided into hierarchical architectures, and a topological relationship graph is built based on the divided architectures. Each node in the topological graph is transformed into an initial feature vector, resulting in a set of initial feature vectors for each node. For each node in the initial feature vector set, perform high-order correlation feature fusion to obtain the correlation fusion vector of each node; The associated fusion vectors are concatenated to form data association features.

[0064] In one embodiment, when the personnel job result generation module 105 generates a target personnel job matching result based on the data association features and the matching degree, it is used to: The candidate coding vector and the job coding vector corresponding to the matching degree are concatenated into a fusion vector; Based on the data association features, the fused vector is projected into the same vector space to obtain the projected feature vector; Attention fusion is performed on the data association features and the projected feature vector to obtain comprehensive features; Obtain the multi-target matching dimensions and assign weights to each target dimension in the multi-target matching dimensions to obtain the target dimension weights for each target dimension; The comprehensive features are weighted and fused with the target dimension weights to obtain a multi-target comprehensive matching value; The initial personnel job matching scheme is determined based on the multi-target comprehensive matching value and the preset target matching threshold. The multi-objective comprehensive matching value is updated based on the preset historical candidate target dimension weights to obtain the updated multi-objective comprehensive matching value; The initial personnel job matching scheme is adjusted based on the updated multi-objective comprehensive matching value to obtain the target personnel job matching result.

[0065] This invention provides a personnel-job matching device. For the business of constructing target personnel-job matching results, it acquires multimodal resume data of several target candidates and job requirement data of several positions, and constructs a job requirement weight matrix based on the job requirement data. The multimodal resume data is converted into candidate feature vectors, and the job requirement data is converted into job feature vectors. The candidate feature vectors are then weighted and fused according to the job requirement weight matrix to obtain a weighted candidate vector. The matching degree between the weighted candidate vector and the job feature vector is calculated. Association features are extracted from the multimodal resume data to obtain data association features. Target personnel-job matching results are generated based on the data association features and the matching degree. By extracting data association features, this invention overcomes the limitations of traditional feature extraction schemes that only focus on a single attribute, while also improving the accuracy of personnel-job matching.

[0066] For specific limitations regarding a personnel job matching device, please refer to the limitations of a personnel job matching method described above, which will not be repeated here. Each module in the aforementioned personnel job matching device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0067] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a personnel job matching method on the server side.

[0068] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a personnel job matching method on the client side.

[0069] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data; The multimodal resume data is converted into candidate feature vectors, and the job requirement data is converted into job feature vectors. The candidate feature vectors are then weighted and fused according to the job requirement weight matrix to obtain a candidate weighted vector. Calculate the matching degree between the candidate weighted vector and the job feature vector; The association features of the multimodal resume data are extracted to obtain data association features; The target personnel job matching results are generated based on the data association features and the matching degree.

[0070] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data; The multimodal resume data is converted into candidate feature vectors, and the job requirement data is converted into job feature vectors. The candidate feature vectors are then weighted and fused according to the job requirement weight matrix to obtain a candidate weighted vector. Calculate the matching degree between the candidate weighted vector and the job feature vector; The association features of the multimodal resume data are extracted to obtain data association features; The target personnel job matching results are generated based on the data association features and the matching degree.

[0071] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0073] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0074] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. If any software tools or components other than those of our company appear in the embodiments, they are merely illustrative examples and do not represent actual use. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for matching personnel to job positions, characterized in that, include: Obtain multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data; The multimodal resume data is converted into candidate feature vectors, and the job requirement data is converted into job feature vectors. The candidate feature vectors are then weighted and fused according to the job requirement weight matrix to obtain a candidate weighted vector. Calculate the matching degree between the candidate weighted vector and the job feature vector; The association features of the multimodal resume data are extracted to obtain data association features; The target personnel job matching results are generated based on the data association features and the matching degree.

2. The personnel job matching method as described in claim 1, characterized in that, The step of converting the multimodal resume data into candidate feature vectors includes: The multimodal resume data is standardized to obtain standardized resume data, and the modality type features corresponding to the standardized resume data are extracted. Convert the text data in the standardized resume data into modal high-dimensional semantic vectors; A type feature space is constructed based on the modality type features, and the high-dimensional semantic vector of the modality is semantically mapped based on the type feature space to obtain the candidate feature vector.

3. The personnel job matching method as described in claim 1, characterized in that, The step of constructing a job requirement weight matrix based on the job requirement data includes: Extract job scenario description data from the job requirement data, and perform scenario feature extraction on the job scenario description data to obtain job scenario features; Filter the candidate job evaluation dimensions from the job scenario features, and assign weights to the candidate job evaluation dimensions to obtain a basic weight matrix; The basic weight matrix is ​​adjusted according to the job scenario characteristics to obtain the job requirement weight matrix.

4. The personnel job matching method as described in claim 1, characterized in that, The step of weighting and fusing the candidate feature vectors according to the job requirement weight matrix to obtain a candidate weighted vector includes: Establish a dimensional mapping relationship between the job requirement weight matrix and the candidate feature vector; The candidate feature vectors are weighted according to the dimensional mapping relationship to obtain weighted candidate feature vectors. The weighted candidate feature vectors are normalized to obtain the candidate weighted vectors.

5. The personnel job matching method as described in claim 1, characterized in that, The calculation of the matching degree between the candidate weighted vector and the job feature vector includes: Deep semantic encoding is performed on the candidate weighted vector and the job feature vector respectively to obtain the candidate encoding vector and the job encoding vector; Calculate the similarity between the candidate coding vector and the job coding vector; The similarity is normalized, and the normalized similarity is converted into a matching score.

6. The personnel job matching method as described in claim 1, characterized in that, The step of extracting association features from the multimodal resume data to obtain data association features includes: Extract basic candidate employment information from the multimodal resume data, and extract job requirement information from the job requirement data; Construct a set of multiple node types based on the candidate's basic employment information and the job requirements information; Determine the skill-related edge set and the company-related edge set based on the multi-type node set; Construct a heterogeneous graph based on the set of multiple types of nodes, the set of skill-related edges, and the set of company-related edges; The heterogeneous graph is divided into hierarchical architectures, and a topological relationship graph is built based on the divided architectures. Each node in the topological graph is transformed into an initial feature vector, resulting in a set of initial feature vectors for each node. For each node in the initial feature vector set, perform high-order correlation feature fusion to obtain the correlation fusion vector of each node; The associated fusion vectors are concatenated to form data association features.

7. The personnel job matching method as described in claim 1, characterized in that, The step of generating target personnel job matching results based on the data association features and the matching degree includes: The candidate coding vector and the job coding vector corresponding to the matching degree are concatenated into a fusion vector; Based on the data association features, the fused vector is projected into the same vector space to obtain the projected feature vector; Attention fusion is performed on the data association features and the projected feature vector to obtain comprehensive features; Obtain the multi-target matching dimensions and assign weights to each target dimension in the multi-target matching dimensions to obtain the target dimension weights for each target dimension; The comprehensive features are weighted and fused with the target dimension weights to obtain a multi-target comprehensive matching value; The initial personnel job matching scheme is determined based on the multi-target comprehensive matching value and the preset target matching threshold. The multi-objective comprehensive matching value is updated based on the preset historical candidate target dimension weights to obtain the updated multi-objective comprehensive matching value; The initial personnel job matching scheme is adjusted based on the updated multi-objective comprehensive matching value to obtain the target personnel job matching result.

8. A personnel job matching device, characterized in that, include: The matrix construction module is used to acquire multimodal resume data of several target candidates and job requirement data of several positions, and construct a job requirement weight matrix based on the job requirement data. The transformation and fusion module is used to transform the multimodal resume data into candidate feature vectors and the job requirement data into job feature vectors, and to perform weighted fusion of the candidate feature vectors according to the job requirement weight matrix to obtain a candidate weighted vector. The matching degree calculation module is used to calculate the matching degree between the candidate weighted vector and the job feature vector; The data association feature extraction module is used to extract association relationship features from the multimodal resume data to obtain data association features; The personnel job result generation module is used to generate target personnel job matching results based on the data association features and the matching degree.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the personnel job matching method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the personnel job matching method as described in any one of claims 1 to 7.

Images