A data analysis-based job matching method, equipment, and medium

Abstract

This invention discloses a data analysis-based job matching method, device, and medium, relating to the field of intelligent job matching technology. The method includes: extracting features from structured industrial behavior data using a behavioral pattern mining algorithm to obtain a multi-dimensional feature set; constructing a multi-dimensional industrial capability profile vector through normalization and weighted fusion with preset capability dimension weights; constructing a semantic association mapping graph using the multi-dimensional industrial capability profile vector and a job semantic knowledge graph; mapping and inferring the semantic association mapping graph using a mismatch inference algorithm to identify potentially undermatched job areas and generate user job matching scores; and ranking the user job matching scores and combining job priority with user preferences to generate a personalized job recommendation list. By constructing the semantic association mapping graph, a deep connection between user capabilities and job semantics is achieved, effectively improving the accuracy and robustness of multimodal information matching.

Description

Technical Field

[0001] This invention relates to the field of intelligent job matching technology, and in particular to a job matching method, device and medium based on data analysis. Background Technology

[0002] Against the backdrop of rapid development in digital human resource management, job matching technology has gradually evolved from the traditional static resume-based matching model to a precise job-person matching system that integrates multi-source heterogeneous data and intelligent algorithms. Currently, mainstream job matching methods generally rely on static attribute information such as resume data, educational background, and years of work experience provided by users, as well as job descriptions and qualifications. Basic job-person matching models are constructed through keyword extraction, vectorized encoding, and similarity calculation. With the continuous enhancement of natural language processing technology and knowledge graph modeling capabilities, some studies have attempted to structure the semantic relationships of job positions, thereby improving the accuracy of job information representation. Simultaneously, various user profile generation mechanisms based on deep learning have emerged, enhancing the modeling ability of job seekers' preferences and abilities.

[0003] However, existing data-driven job matching methods have not yet systematically modeled and effectively utilized data on job seekers' operational behaviors, especially in skilled positions such as industrial manufacturing and smart factories. Job seekers' behavioral characteristics are highly temporal and structured, and this type of behavioral information has direct value in reflecting abilities during job suitability assessments. Current technologies largely focus on static descriptive attributes or subjective self-assessments, making it difficult to reflect the actual performance dimensions of skilled industrial workers in their work processes. This results in room for improvement in job matching results regarding skill matching accuracy and job adaptability assessment. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a data analysis-based job matching method to address the problem of lack of user operation behavior data modeling and competency profile quantification in industrial skilled positions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a data analysis-based employment matching method, which includes collecting user industrial operation behavior data, performing time series processing, event labeling and field standardization to form structured industrial behavior data;

[0008] By using behavioral pattern mining algorithms to extract features from structured industrial behavioral data, a multi-dimensional feature set is obtained. After normalization and weighted fusion with preset capability dimension weights, a multi-dimensional industrial capability profile vector is constructed.

[0009] We collect job descriptions, job responsibility texts, and job task process data from enterprises, and use natural language processing methods to construct a job semantic knowledge graph.

[0010] A semantic association mapping graph is constructed by using multidimensional industrial capability profile vectors and job semantic knowledge graphs;

[0011] A mismatch reasoning algorithm is used to map and reverse the semantic association mapping graph to identify potentially undermatched job areas and generate user job matching scores.

[0012] The system sorts users' job matching scores and combines job priority with user preferences to generate a personalized job recommendation list.

[0013] As a preferred embodiment of the data analysis-based employment matching method described in this invention, the specific steps for collecting user industrial operation behavior data, performing time-series processing, event tagging, and field standardization to form structured industrial behavior data are as follows.

[0014] The dynamic time warping algorithm is used to align the user's industrial operation behavior data to a time series, generate aligned behavior sequences, and perform hierarchical time series pattern mining and multi-granularity behavior event recognition to obtain operation pattern data.

[0015] A non-linear mapping method is used to normalize and unify the types of operation mode data, generating standardized fields;

[0016] By aligning and integrating the alignment behavior sequences, operation mode data, and standardized fields along the time axis, structured industrial behavior data is formed.

[0017] As a preferred embodiment of the data analysis-based employment matching method of the present invention, the step of using a behavioral pattern mining algorithm to extract features from structured industrial behavioral data to obtain a multi-dimensional feature set is as follows:

[0018] Structured industrial behavior data is divided into multiple dynamic sliding time windows. Time-domain statistical analysis and frequency-domain transformation are performed on each dynamic sliding time window to extract time-domain statistical features and frequency-domain features. Weighted nonlinear fusion processing is then performed to generate a comprehensive behavior feature vector.

[0019] The comprehensive behavioral feature vectors are concatenated in chronological order to form a multi-dimensional behavioral feature set.

[0020] As a preferred embodiment of the data analysis-based employment matching method of the present invention, the step of constructing a multi-dimensional industrial capability profile vector by normalization processing and weighted fusion based on preset capability dimension weights is as follows:

[0021] The multi-dimensional behavioral feature set is uniformly normalized to obtain a uniform scale feature set, and the corresponding fusion weights are calculated through a nonlinear function.

[0022] By using fusion weights to perform nonlinear interactive fusion of a unified scale feature set, a capability expression vector is generated, and vector concatenation is performed to construct a multidimensional industrial capability profile vector.

[0023] As a preferred embodiment of the data analysis-based job matching method of the present invention, the steps of collecting enterprise job descriptions, job responsibility texts, and job task flow data, and constructing a job semantic knowledge graph using natural language processing methods are as follows.

[0024] The system performs word segmentation and part-of-speech tagging on enterprise job descriptions, job responsibility texts, and job task process data to generate a tagged word sequence.

[0025] The labeled word sequence is input into a context-aware language model to generate multi-dimensional vectors, and then formed into a semantic feature space through deep semantic encoding.

[0026] By using multivariate nonlinear interaction functions, we can calculate the higher-order interaction tensor between entity pairs in the semantic feature space and obtain the weighted entity nodes and relation edges.

[0027] The weighted entity nodes and relation edges are imported into the weighted entity relation network to construct a weighted directed graph structure, and the semantic relation features between entity nodes are extracted to construct weighted directed edges.

[0028] Weighted directed edges are semantically fused and integrated with graph entities, and organized into a job semantic knowledge graph according to job attributes and relationships.

[0029] As a preferred embodiment of the data analysis-based job matching method of the present invention, the step of constructing a semantic association mapping graph using multidimensional industrial capability profile vectors and job semantic knowledge graphs involves the following specific steps.

[0030] The multidimensional industrial capability profile vector is nonlinearly normalized to obtain the capability feature vector.

[0031] Multidimensional semantic vectors are generated for entity nodes in the job semantic knowledge graph, and high-dimensional semantic expression vectors are constructed by combining job attributes.

[0032] The capability feature vector and the high-dimensional semantic representation vector are fused using temporal and static features to output a unified node representation vector;

[0033] Calculate the similarity and association weights between entity nodes in the unified node representation vector to obtain the entity node similarity matrix. Then, filter the entity node pairs in the entity node similarity matrix that exceed the dynamic adjustment threshold, construct weighted directed edges to connect the nodes, and generate a semantic association mapping graph.

[0034] As a preferred embodiment of the data analysis-based job matching method of the present invention, the step of using a mismatch reasoning algorithm to map and reverse-reason on the semantic association mapping graph, identifying potentially undermatched job areas, and generating a user job matching score is as follows.

[0035] The multidimensional deviations between user capability vectors and entity nodes in the semantic association mapping graph are mapped to form a job capability mismatch graph structure.

[0036] By using a semantic path feedback mechanism to perform reverse reasoning on the job competency mismatch graph structure, the propagation path and intensity distribution of job competency mismatch are obtained.

[0037] The propagation path and intensity distribution of job competency mismatch are associated with high-dimensional semantic expression vectors. After weighted summation, a non-linear activation function is applied to generate a user job matching score.

[0038] As a preferred embodiment of the data analysis-based job matching method of the present invention, the specific steps for generating a personalized job recommendation list by sorting user job matching scores and combining job priority with user preferences are as follows.

[0039] Collect job priority information and user preferences, and construct a set of ranking factors together with user job matching scores;

[0040] The ranking factor set is input into the job ranking decision function to calculate the ranking score of each job. Jobs with ranking scores lower than the dynamic quantile threshold are filtered out by setting a dynamic quantile threshold for ranking scores. The ranking scores of the remaining jobs are normalized to generate a job recommendation priority ranking list.

[0041] The similarity between the target positions in the priority ranking list and the ability feature vectors is calculated, and then weighted and fused with the ranking weights to generate the final position score.

[0042] The final job scores are sorted in descending order, and the job with the highest final score is selected to generate a personalized job recommendation list.

[0043] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein the computer program, when executed by the processor, implements any step of the data analysis-based employment matching method described in the first aspect of the present invention.

[0044] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the employment matching method based on data analysis as described in the first aspect of the present invention.

[0045] The beneficial effects of this invention are as follows: By constructing a semantic association mapping graph, a deep connection between user capabilities and job semantics is achieved, effectively improving the accuracy and robustness of multimodal information matching; further, a mismatch inference algorithm is introduced to identify fine-grained difference paths between capabilities and job requirements, and an accurate job matching score is generated through a semantic path feedback mechanism, enhancing the personalization and interpretability of recommendations. Overall, an intelligent employment matching system with semantic understanding and reasoning evaluation capabilities is formed, significantly improving the matching effect and system practicality. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0047] Figure 1 This is a flowchart of a data analysis-based job matching method.

[0048] Figure 2 A flowchart for generating structured industrial behavior and capability profile vectors.

[0049] Figure 3 A flowchart for constructing a semantic knowledge graph for job-related texts.

[0050] Figure 4 A flowchart for semantic association mismatch reasoning and matching score generation. Detailed Implementation

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0054] Reference Figures 1-4 This is one embodiment of the present invention, which provides a data analysis-based employment matching method, including the following steps:

[0055] S1. Collect user industrial operation behavior data, perform time series processing, event tagging, and field standardization to form structured industrial behavior data.

[0056] S1.1. The dynamic time warping algorithm is used to align the user's industrial operation behavior data to a time series, generate aligned behavior sequences, and perform hierarchical time series pattern mining and multi-granularity behavior event recognition to obtain operation pattern data.

[0057] Furthermore, a dynamic time warping algorithm is used to align the time series of user industrial operation behavior data, ensuring accurate matching of behavior data of different time periods and frequencies on the time axis. Based on the aligned behavior sequences, hierarchical time series pattern mining is performed. Based on the aligned behavior sequences, multi-granularity time window and frequent subsequence mining algorithms are used, combined with event importance indicators, to identify recurring patterns and key behavioral events layer by layer. Combined with multi-granularity behavior event identification methods, the behavior sequences are decomposed at different time scales to extract operation event features from fine-grained to coarse-grained. Finally, operation pattern data is formed, which represents the multi-level time series features and event structure of user industrial operation behavior, providing a foundation for subsequent capability vector construction and job matching.

[0058] It should also be noted that user industrial operation behavior data includes operation action sequences, equipment status information, environmental parameters, sensor monitoring data, event logs, and historical operation records, comprehensively reflecting the user's operation behavior and equipment operating status.

[0059] Fine-grained operational event characteristics typically refer to specific, short-term, and single actions or operational steps, such as a single button press, equipment start-up or shutdown, etc.; coarse-grained operational event characteristics refer to complex behavioral patterns formed by a longer time span or a combination of multiple fine-grained events, such as completing a certain process, equipment operation cycle, or a continuous operation process.

[0060] S1.2. Use a non-linear mapping method to normalize and unify the types of operation mode data, and generate standardized fields.

[0061] Furthermore, after extracting the operation mode data, the various behavioral events and corresponding numerical features contained in the operation mode data are first uniformly indexed and structurally labeled to establish the original feature matrix of the operation mode data. For each type of numerical feature in the original feature matrix, a nonlinear mapping method is used for normalization to construct a normalized feature set. The choice of nonlinear mapping method is based on the discreteness and numerical distribution characteristics of the operation behavior. The hyperbolic tangent function is often used to compress and map the numerical feature fields of the operation behavior to maintain the nonlinear distribution characteristics between the normalized features.

[0062] For categorical fields in the operation mode data, such as behavior type, event identifier, and device status (non-numerical attributes), a unified encoding method for categorical fields is used. First, based on the semantic information of each categorical field in the operation mode data, a corresponding independent identifier set is extracted, and a classification dictionary index is constructed. Then, one-hot encoding or embedding vector mapping is performed on all categorical fields to convert non-numerical behavior labels into corresponding sparse vectors. Finally, the positions of all encoded categorical fields in the operation mode data are rearranged and aligned to ensure that all categorical fields have a unified representation structure with numerical fields in the normalized feature set.

[0063] After completing the nonlinear normalization of numerical fields and unifying the types of categorical fields, all processed fields are concatenated and integrated to generate a standardized field set with stable structure and consistent expression. This standardized field set retains all the information structure of the operational pattern data and possesses a unified numerical scale and representation format, providing a basic expressive input for subsequent feature fusion and semantic modeling.

[0064] The duration feature value of the operation behavior is calculated from the start time and end time of the behavior recorded in the industrial operation behavior data. By calculating the difference between the timestamp information of each type of behavior event in the operation mode data, the execution duration of each behavior event is extracted, and the corresponding fields are filled according to the behavior event index to form a set of operation behavior duration feature values.

[0065] S1.3 Align and integrate the alignment behavior sequence, operation mode data and standardized fields along the time axis to form structured industrial behavior data.

[0066] Furthermore, after completing the construction of the alignment behavior sequence, the extraction of operation mode data, and the generation of standardized fields, the operation behaviors in the operation mode data are mapped to the corresponding time periods according to the start and end times, based on the time index in the alignment behavior sequence, and then the standardized fields are aligned to the corresponding time positions according to the behavior identifiers.

[0067] When multiple operational behaviors occur at the same point in time, the primary behavior and standardized fields are determined by behavior priority, while secondary behavior information is retained as a supplement. Finally, the aligned behavior sequences, operation mode data, and standardized fields are concatenated and integrated in chronological order to generate structured industrial behavior data.

[0068] S2. Use behavioral pattern mining algorithms to extract features from structured industrial behavioral data to obtain a multi-dimensional feature set. After normalization and weighted fusion with preset capability dimension weights, construct a multi-dimensional industrial capability profile vector.

[0069] S2.1 Divide the structured industrial behavior data into multiple dynamic sliding time windows, perform time-domain statistical analysis and frequency-domain transformation processing on each dynamic sliding time window, extract time-domain statistical features and frequency-domain features, and perform weighted nonlinear fusion processing to generate a comprehensive behavior feature vector.

[0070] Furthermore, the structured industrial behavior data is divided into sliding windows along the time axis, with dynamic sliding window length and sliding step size set, and the data within each dynamic sliding time window is processed independently. First, within each dynamic sliding time window, statistics are calculated based on indicators such as operation frequency, event duration, and behavior interval, and time-domain statistical features such as mean, variance, extreme values, and kurtosis are extracted.

[0071] For the time series behavior values ​​formed by the structured industrial behavior data within each dynamic sliding time window, the fast Fourier transform method is used to convert the time domain signal into a frequency domain signal. Frequency domain features such as the main frequency reflecting the periodicity of the behavior, the spectral energy distribution characterizing the energy distribution characteristics, and the frequency band amplitude describing the behavior intensity of different frequency bands are extracted to characterize the frequency domain variation law of the structured industrial behavior data.

[0072] After extracting the time-domain statistical features and frequency-domain features, a weighted nonlinear function is used to fuse these features, generating a comprehensive behavioral feature representation through nonlinear interactions between different features. Finally, the fusion result is used to construct the comprehensive behavioral feature vector corresponding to the current dynamic sliding time window.

[0073] S2.2. Concatenate the comprehensive behavioral feature vectors in chronological order to form a multi-dimensional behavioral feature set.

[0074] Furthermore, after generating the comprehensive behavioral feature vectors, all comprehensive behavioral feature vectors are sorted by time according to the start time of the dynamic sliding time window. Each comprehensive behavioral feature vector is then arranged in chronological order to construct a time-ordered feature sequence.

[0075] To maintain behavioral continuity and temporal consistency, the overlapping feature vectors between adjacent dynamic sliding time windows are ensured to remain sequentially connected during the splicing process. Through temporal splicing, the comprehensive behavioral feature vectors corresponding to multiple dynamic sliding time windows are integrated to form a complete multi-dimensional behavioral feature set.

[0076] S2.3. Perform unified normalization on the multi-dimensional behavioral feature set to obtain a unified scale feature set, and calculate the corresponding fusion weights through a nonlinear function.

[0077] Furthermore, after constructing a multi-dimensional behavioral feature set, the behavioral features of each dimension in the set are normalized according to a unified numerical range. This transforms features with different dimensions and numerical ranges to a uniform scale, ensuring the comparability of dimensional behavioral features in subsequent fusion processes. The normalization process can employ existing nonlinear mapping methods, using logarithmic transformations to compress and map the operational behavior numerical feature fields. After normalization, in the unified-scale feature set, the operational behavior numerical feature fields of each dimension are at the same scale and possess a unified numerical structure.

[0078] Based on a unified scale feature set, a nonlinear mapping is performed on the weighted linear combination of features using activation functions or kernel functions to extract the nonlinear interaction effects between features. The fusion weights of each feature are then calculated through normalization, as expressed in the following expression:

[0079] ;

[0080] in, Indicates the first The fusion weights of each feature Represents the normalization function. Indicates the first The original feature values ​​of each feature. This represents the feature index traversed during the summation process. Indicates the total number of features. Indicates the first The original or processed feature values ​​of each feature;

[0081] S2.4. Utilize fusion weights to perform nonlinear interactive fusion on the unified scale feature set to generate a capability expression vector, and perform vector concatenation operations to construct a multidimensional industrial capability profile vector.

[0082] Furthermore, after calculating the fusion weights, the features in each dimension of the unified scale feature set are weighted dimension by dimension based on the fusion weights. Nonlinear interaction fusion between features is achieved through known nonlinear functions to generate an expression vector representing the comprehensive effect of features.

[0083] The operational behavior numerical feature field of each dimension in the capability expression vector is calculated by nonlinear function from the operational behavior numerical feature fields in the corresponding uniform scale feature set and the fusion weight, reflecting the capability representation result after weighted fusion of multi-dimensional behavioral features. Multiple capability expression vectors are concatenated according to the order of their respective time windows or behavioral segments to maintain temporal consistency and behavioral structural continuity, thus constructing a multi-dimensional industrial capability profile vector with a complete temporal structure and multi-dimensional information expression.

[0084] S3. Collect job descriptions, job responsibility texts, and job task process data from enterprises, and construct a job semantic knowledge graph using natural language processing methods.

[0085] S3.1 Perform word segmentation and part-of-speech tagging on the enterprise job descriptions, job responsibility texts, and job task process data to generate a tagged word sequence.

[0086] Furthermore, for enterprise job descriptions, job responsibility texts, and job task process data, the text content is first processed by word segmentation, breaking down continuous text strings into independent word units to ensure that each word unit fully expresses the specific semantics.

[0087] Subsequently, part-of-speech tagging (POS) is applied to the word segmentation results. Based on the grammatical function and contextual relationship of each word in a sentence, the POS category of each word is labeled, generating a labeled word sequence. The labeled word sequence not only contains the words themselves but also their corresponding POS information, providing structured input for semantic feature extraction and deep semantic encoding.

[0088] S3.2 Input the labeled word sequence into the context-aware language model to generate multi-dimensional vectors, and form a semantic feature space through deep semantic encoding.

[0089] Furthermore, the annotated word sequence is input into a context-aware language model. Each word is represented as a vector based on its contextual relationship, generating a multi-dimensional word vector sequence. Subsequently, deep semantic encoding methods are used to extract hierarchical features and compress semantic information from this multi-dimensional word vector sequence, capturing deep semantic connections between words and the overall semantic structure of the sentence, forming a semantic feature space with high-dimensional expressive power. The vectors in this semantic feature space not only reflect the semantic content of the words but also contain context-related semantic enhancement information, providing a foundational expression for semantic relationship calculation and knowledge graph construction.

[0090] It should also be noted that the training process of the context-aware language model is based on multi-source semantic data and user behavior features. It uses a self-attention mechanism to capture contextual dependencies, optimizes the parameters of the context-aware language model through supervised learning, combines gradient descent and regularization techniques to prevent overfitting, and adjusts hyperparameters through a validation set to ensure the accuracy and stability of the context-aware language model.

[0091] S3.3. Using multivariate nonlinear interaction functions, calculate the higher-order interaction tensor between entity pairs in the semantic feature space to obtain the weighted entity nodes and relation edges.

[0092] Furthermore, based on the entity node attributes and relation edge weights in the job semantic knowledge graph, a multivariate nonlinear interaction function is employed to uncover the complex hidden relationships and interaction patterns between entity pairs through a holistic observation and fusion of semantic features between entity nodes. This process relies on existing entity semantic descriptions and job attribute information to comprehensively analyze the attribute characteristics and interrelationships of entity nodes from multiple dimensions, extracting interaction features that reflect high-order semantic relationships between entities, and then assigning weighted attributes to entity nodes and relation edges. This approach, through deep semantic understanding and fusion, forms a richer and more accurate representation of weighted entity nodes and relation edges.

[0093] S3.4. Import the weighted entity nodes and relation edges into the weighted entity relation network to construct a weighted directed graph structure, and extract the semantic relation features between entity nodes to construct weighted directed edges.

[0094] Furthermore, by using weighted entity nodes and relational edges as input, a weighted entity-relationship network is constructed. Through the weighted entity nodes and edge weights, a weighted directed graph structure is established. The weighted directed graph structure uses entity nodes as vertices and weighted relational edges as directed edges connecting the corresponding nodes. The edge weights reflect the strength of the semantic association between nodes.

[0095] By traversing a weighted directed graph, semantic relationship features between entity nodes are extracted, including adjacency relationships, path weights, and node connection patterns, forming a multi-dimensional feature set describing the semantic connections between nodes. Finally, by combining node features and relationship weights, weighted directed edges are constructed to fully express the semantic structure and association strength between entities.

[0096] S3.5. Perform semantic fusion and graph entity integration on the weighted directed edges, and organize them into a job semantic knowledge graph according to job attributes and relationships.

[0097] Furthermore, semantic fusion is performed on weighted directed edges. First, the semantic information of each relation edge is integrated through methods such as weighted averaging and nonlinear transformation to eliminate redundancy and noise and enhance key semantic features. Then, entity nodes are classified and grouped according to job attributes, and the structured integration of entities is completed by combining the semantic relationships between entities.

[0098] By organizing and optimizing entities and relationships, an orderly hierarchical structure that conforms to job attributes and relationships between jobs is formed, enabling multi-level expression of semantic information. Ultimately, a job semantic knowledge graph that is structurally complete, semantically rich, and reflects job characteristics and interrelationships is constructed.

[0099] S4. Construct a semantic association mapping graph by using multidimensional industrial capability profile vectors and job semantic knowledge graphs.

[0100] S4.1. Perform nonlinear normalization on the multidimensional industrial capability profile vector to obtain the capability feature vector.

[0101] Furthermore, a nonlinear normalization method is used to process the features of each dimension in the multidimensional industrial capability profile vector, mapping the original operational behavior numerical feature fields to a unified numerical range, thus eliminating the problems of dimensional differences and inconsistent numerical ranges between different dimensions.

[0102] During the normalization process, the distribution of the original operational behavior numerical feature fields in each dimension is adjusted using a nonlinear function. This enhances the robustness of the capability feature vector to outliers and extreme values ​​while preserving the original feature information, thereby improving the stability and accuracy of subsequent calculations. The final output capability feature vector maintains the relative relationships of multi-dimensional features and possesses a uniform scale, facilitating fusion and comparison with other semantic vectors.

[0103] S4.2 Generate multi-dimensional semantic vectors for entity nodes in the job semantic knowledge graph, and construct high-dimensional semantic expression vectors by combining job attributes.

[0104] Furthermore, for each entity node in the job semantic knowledge graph, based on semantic description information and contextual relationships, the context-aware language model extracts a multi-dimensional semantic vector of the entity by encoding the semantic description of the entity node and the semantic information of neighboring nodes. The multi-dimensional semantic vector can reflect the semantic features of the entity and its position in the semantic space.

[0105] Subsequently, by combining job attribute information, the multidimensional semantic vector and the multidimensional industrial capability profile vector are fused. Through feature concatenation or weighted fusion methods, a high-dimensional semantic expression vector containing semantic information and job attributes is constructed. This high-dimensional semantic expression vector comprehensively reflects the semantic meaning of entity nodes and job characteristics, providing a basic expression for subsequent semantic association mapping and matching.

[0106] S4.3. Perform temporal and static feature fusion processing on the capability feature vector and the high-dimensional semantic expression vector to output a unified node representation vector.

[0107] Furthermore, the capability feature vector and the high-dimensional semantic expression vector are first preprocessed to make the two types of vectors compatible in terms of dimension and representation. Then, the time series information in the capability feature vector is encoded using a time series feature extraction method to capture dynamic change features; and the static job attribute information in the high-dimensional semantic expression vector is preserved and enhanced.

[0108] Feature fusion technology is used to fuse temporally encoded capability feature vectors with statically processed high-dimensional semantic representation vectors. Fusion methods include feature concatenation, weighted superposition, or nonlinear interaction to ensure the effective combination of dynamic capability features and static job semantic information. The final output unified node representation vector comprehensively reflects the temporal dynamic features of individual capabilities and the semantic static features of the job, providing a unified representation basis for subsequent semantic matching and association analysis.

[0109] S4.4 Calculate the similarity and association weights between entity nodes in the unified node representation vector to obtain the entity node similarity matrix. Then, filter the entity node pairs in the entity node similarity matrix that exceed the dynamic adjustment threshold, construct weighted directed edges to connect the nodes, and generate a semantic association mapping graph.

[0110] Furthermore, a unified node representation vector is used to calculate the similarity and association weights between entity nodes. The entity node similarity matrix is ​​obtained through a similarity calculation method, expressed as:

[0111] ;

[0112] in, Represents the entity node similarity matrix. This represents a matrix consisting of vectors representing all entity nodes, where rows represent different entity nodes. This represents the transpose of the entity node vector matrix. This represents the dot product matrix between entity node vectors. This indicates that cosine similarity is calculated for each pair of entity vectors. This represents the matrix transpose operation;

[0113] Each element in the entity node similarity matrix reflects the semantic similarity of the corresponding node pair.

[0114] A dynamic threshold is used to filter the entity node similarity matrix. The dynamic threshold is adjusted dynamically according to the actual semantic distribution to filter out entity node pairs whose similarity exceeds the dynamic threshold. For the selected node pairs, weighted directed edges are established to connect the entity nodes, and the edge weight is the similarity of the corresponding node pair. Finally, a semantic association mapping graph with weighted directed edges is formed to express the strength and direction of the semantic relationship between entity nodes.

[0115] It should also be noted that the dynamic threshold setting process is based on the statistical characteristics of the entity node similarity matrix. The mean and standard deviation of all elements in the similarity matrix are calculated, and an initial threshold is set based on this. The initial threshold is jointly determined by the mean and standard deviation of the similarity matrix. Specifically, the standard deviation is mapped proportionally and combined with the mean to obtain an initial boundary value reflecting the overall distribution level. According to actual business needs and semantic distribution, the dynamic threshold range is dynamically adjusted. By monitoring changes in the distribution of the similarity matrix in real time, the dynamic threshold is adjusted upwards or downwards as needed to ensure that the quantity and quality of the selected entity nodes meet expectations. The adjustment process combines a sliding window statistical method to continuously observe the similarity distribution, ensuring that the dynamic threshold setting is adaptable and stable, and achieving effective control and optimization of the semantic association mapping graph construction.

[0116] S5. The mismatch reasoning algorithm is used to map and reverse reason the semantic association map, identify potential undermatched job areas, and generate user job matching scores.

[0117] S5.1 Map the multidimensional deviations between user capability vectors and entity nodes in the semantic association mapping graph to form a job capability mismatch graph structure.

[0118] Furthermore, the user capability vector and the multi-dimensional semantic vector of the entity node in the semantic association mapping graph are mapped through vector difference calculation. Specifically, this includes calculating the deviation value between the capability vector and the corresponding dimension of the entity node in each dimension, and constructing a deviation matrix to characterize the difference between the user's capability and the job requirements.

[0119] By combining the deviation matrix with the entity node relationship structure, a job competency mismatch graph structure is constructed based on the magnitude of the deviation and the association weight. This clearly shows the mismatch between user competencies and job requirements, thereby providing decision support for subsequent competency compensation and job adjustments. This step is closely linked to the construction of the semantic association mapping graph, achieving a deep expression of the competency-job matching status through multi-dimensional deviation mapping.

[0120] S5.2. Using the semantic path feedback mechanism, reverse reasoning is performed on the job competency mismatch graph structure to obtain the propagation path and intensity distribution of job competency mismatch.

[0121] Furthermore, the nodes and edges in the job competency mismatch graph structure undergo reverse reasoning based on a semantic path feedback mechanism. Starting from the entity node with a high mismatch intensity, the propagation proceeds backward along the weighted directed edges. The reverse propagation path of the mismatch impact factor along the directed edges is analyzed in the job semantic topology graph. During propagation, the propagation intensity is dynamically adjusted based on the edge weights, and the mismatch impact value transmitted along each reverse propagation path is accumulated. Finally, the set of mismatch propagation paths and their corresponding intensity distribution are output, reflecting the diffusion range and influence of job competency mismatch in the semantic structure, providing a basis for accurately locating competency gaps and making adjustments. Based on the job competency mismatch graph structure, a deep analysis of mismatch propagation is achieved through a semantic path feedback mechanism.

[0122] S5.3. Associate the propagation path and intensity distribution of job competency mismatch with the high-dimensional semantic expression vector, and generate a user job matching score by weighted summation and nonlinear activation function mapping.

[0123] Furthermore, the propagation path and intensity distribution of job competency mismatch are correlated with the corresponding high-dimensional semantic expression vectors in the job semantic knowledge graph. By weighted summation of the intensity distribution in the propagation path and the corresponding high-dimensional semantic expression vector weights, job semantic information and mismatch impact are integrated. A non-linear activation function is used to map the weighted results to obtain the user job matching score, expressed as: ;

[0124] in, This represents the final calculated user job matching score. Indicates the total number of characteristic components. Indicates the feature component index. Indicates the first The weights of each feature component, Indicates the first The input values ​​of each feature component, Indicates the bias term. Represents a nonlinear activation function;

[0125] User job matching scores reflect the degree of match between user skills and job requirements. Accurate calculation of matching scores is achieved through deep fusion of job skill mismatch propagation paths and high-dimensional semantic representation vectors.

[0126] S6. Sort the user's job matching score and combine the job priority with the user's preferences to generate a personalized job recommendation list.

[0127] S6.1 Collect job priority information and user preferences, and construct a set of ranking factors together with user job matching scores.

[0128] Furthermore, job priority information and user preferences are acquired through appropriate data collection methods to ensure the timeliness and accuracy of the information; the job priority information, user preferences, and previously generated user job matching scores are integrated to form a set of ranking factors;

[0129] The set of ranking factors covers three dimensions: job importance, user personalized needs, and matching score. It supports multi-dimensional weight adjustment in subsequent ranking processes, enabling a comprehensive consideration of job priority and user matching preferences.

[0130] S6.2 Input the set of ranking factors into the job ranking decision function, calculate the ranking score of each job, filter jobs with ranking scores lower than the dynamic quantile threshold by setting a dynamic quantile threshold for ranking scores, normalize the ranking scores of the remaining jobs, and generate a job recommendation priority ranking list.

[0131] Furthermore, the set of ranking factors is input into the job ranking decision function to calculate the ranking score for each job. The ranking score reflects the comprehensive performance of the job across all dimensions of the ranking factors. A dynamic quantile threshold is set based on the ranking score to filter out jobs with ranking scores below the dynamic quantile threshold, thus eliminating poorly performing jobs. The ranking scores of the remaining jobs are normalized to ensure that the ranking scores are distributed within a uniform scale. Finally, a job recommendation priority ranking list is generated, in which jobs are arranged from high to low according to their normalized ranking scores, reflecting the order of job recommendation priority.

[0132] It should also be noted that the process of setting the dynamic quantile threshold for ranking scores first involves collecting the overall distribution of current job ranking scores and analyzing the distribution characteristics and fluctuation range of the ranking scores. Based on the specific requirements of the ranking task and the actual effect of job recommendations, a suitable quantile interval is determined as the initial threshold range to distinguish between high and low ranking scores. The threshold is then adjusted in conjunction with feedback from job ranking and user preferences. The dynamic quantile threshold can dynamically respond to changes in ranking scores under different scenarios, effectively filtering low-scoring jobs. Applying the dynamic quantile threshold to the ranking score filtering process, real-time updates of the dynamic quantile threshold ensure the accuracy of the ranking results and the effectiveness of personalized recommendations. The setting of the dynamic quantile threshold should be continuously optimized and adjusted based on historical ranking data and current business needs to improve the adaptability and accuracy of job ranking.

[0133] The process of setting user preference adjustment thresholds includes collecting users' historical choices and feedback, analyzing preference changes, adjusting thresholds in conjunction with job attributes, dynamically correcting them to match real-time feedback, and continuously updating them to improve recommendation relevance and satisfaction.

[0134] S6.3 Calculate the similarity between the target positions in the priority ranking list and the ability feature vectors, and combine them with the ranking weights to generate the final position score.

[0135] Furthermore, the similarity between the target job in the priority ranking list and the ability feature vector is calculated. The similarity reflects the degree of matching between the target job and the ability feature vector. The similarity results are then weighted and fused based on the ranking weight. The fusion process takes into account the influence and contribution of the ranking weight in different jobs. Finally, a job score is generated. The job score comprehensively reflects the matching effect between the target job and the ability feature vector and the importance of the ranking weight, providing a quantitative basis for job matching.

[0136] S6.4 Sort the final job scores in descending order, select the job with the highest final job score, and generate a personalized job recommendation list.

[0137] Furthermore, the final job scores are sorted in descending order to ensure that higher-scoring jobs are at the top of the list; the highest-scoring job is selected from the sorted results as the primary recommended job; personalized job recommendation lists are generated sequentially based on the score ranking, reflecting the matching priority between user skills and job requirements, facilitating accurate matching and recommendations.

[0138] This embodiment also provides a computer device applicable to the employment matching method based on data analysis, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the employment matching method based on data analysis as proposed in the above embodiment.

[0139] The computer device can be a terminal, comprising a processor, memory, communication 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 communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0140] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the data analysis-based job matching method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0141] In summary, this invention achieves a deep connection between user capabilities and job semantics by constructing a semantic association mapping graph, effectively improving the accuracy and robustness of multimodal information matching; furthermore, it introduces a mismatch inference algorithm to identify fine-grained difference paths between capabilities and job requirements, and generates accurate job matching scores through a semantic path feedback mechanism, enhancing the personalization and interpretability of recommendations. Overall, it forms an intelligent employment matching system with semantic understanding and inference evaluation capabilities, significantly improving matching effectiveness and system practicality.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for employment matching based on data analysis, characterized by: Comprising, Collect user industrial operation behavior data, perform time series arrangement, event labeling and field standardization to form structured industrial behavior data; Use behavior pattern mining algorithm to extract features from structured industrial behavior data to obtain a multi-dimensional feature set, perform normalization processing, and combine preset ability dimension weights for weighted fusion to construct a multi-dimensional industrial ability profile vector; Collect enterprise post description, post responsibility text and post task process data, and use natural language processing method to construct post semantic knowledge graph; Use multi-dimensional industrial ability profile vector and post semantic knowledge graph to construct semantic association mapping graph, the specific steps are as follows, Perform nonlinear normalization processing on the multi-dimensional industrial ability profile vector to obtain an ability feature vector; Generate a multi-dimensional semantic vector for the entity nodes in the post semantic knowledge graph, and construct a high-dimensional semantic expression vector in combination with post attributes; Fuse the ability feature vector and the high-dimensional semantic expression vector in terms of time sequence and static features to output a unified node representation vector; Calculate the similarity and association weight between entity nodes in the unified node representation vector to obtain an entity node similarity matrix, and filter entity node pairs in the entity node similarity matrix that exceed a dynamic adjustment threshold to construct a weighted directed edge connection node and generate a semantic association mapping graph; Use mismatch degree reasoning algorithm to map and reason the semantic association mapping graph to identify potential insufficiently matched post areas, generate user post matching scores, and the specific steps are as follows, Map the multi-dimensional deviation between the user ability vector and the entity nodes in the semantic association mapping graph to form a post ability mismatch graph structure; Use semantic path feedback mechanism to perform reverse reasoning on the post ability mismatch graph structure to obtain the propagation path and intensity distribution of post ability mismatch; Associate the propagation path and intensity distribution of post ability mismatch with the high-dimensional semantic expression vector, perform weighted summation, and then perform nonlinear activation function mapping to generate user post matching scores; Sort the user post matching scores and combine post priority and user preferences to generate a personalized post recommendation list.

2. The data analysis based employment matching method as claimed in claim 1, wherein: The specific steps of collecting user industrial operation behavior data, performing time series arrangement, event labeling and field standardization to form structured industrial behavior data are as follows, Use dynamic time warping algorithm to align the time series of user industrial operation behavior data to generate aligned behavior sequences, and perform hierarchical time series pattern mining and multi-granularity behavior event recognition to obtain operation mode data; Use nonlinear mapping method to normalize and unify the types of operation mode data to generate standardized fields; Align and integrate the aligned behavior sequences, operation mode data and standardized fields according to the time axis to form structured industrial behavior data.

3. The data analysis based employment matching method as claimed in claim 2, wherein: The specific steps of using behavior pattern mining algorithm to extract features from structured industrial behavior data to obtain a multi-dimensional feature set are as follows, The structured industrial behavior data is divided into multiple dynamic sliding time windows, time domain statistical analysis and frequency domain transformation are performed on each dynamic sliding time window, time domain statistical features and frequency domain features are extracted, and weighted nonlinear fusion processing is performed to generate a comprehensive behavior feature vector; The comprehensive behavior feature vector is spliced in time sequence to form a multi-dimensional behavior feature set.

4. The data analysis based employment matching method as claimed in claim 3, wherein: The steps of constructing the multi-dimensional industrial capability profile vector by normalizing and combining the preset capability dimension weight for weighted fusion are as follows, The multi-dimensional behavior feature set is uniformly normalized to obtain a uniform scale feature set, and the corresponding fusion weight is calculated by a nonlinear function; The fusion weight is used for nonlinear interactive fusion of the uniform scale feature set to generate a capability expression vector, and vector splicing operation is performed to construct a multi-dimensional industrial capability profile vector.

5. The data analysis based employment matching method as claimed in claim 4, wherein: The steps of collecting enterprise post description, post responsibility text and post task process data, and constructing a post semantic knowledge graph using natural language processing method are as follows, The enterprise post description, post responsibility text and post task process data are subjected to word segmentation and part-of-speech tagging to generate a tagged word sequence; The tagged word sequence is input into a context-aware language model to generate a multi-dimensional vector, and a semantic feature space is formed through deep semantic coding; A high-order interaction tensor between entity pairs in the semantic feature space is calculated using a multi-element nonlinear interaction function to obtain weighted entity nodes and relationship edges; The weighted entity nodes and relationship edges are imported into a weighted entity relationship network to construct a weighted directed graph structure, and semantic relationship features between entity nodes are extracted to construct a weighted directed edge. The weighted directed edge is subjected to semantic fusion and graph entity integration, and is organized into a post semantic knowledge graph according to post attributes and relationships.

6. The data analysis based employment matching method as claimed in claim 1, wherein: The steps of sorting user post matching scores and combining post priority and user preferences to generate a personalized post recommendation list are as follows, Post priority information and user preferences are collected and used together with user post matching scores to construct a sorting factor set; The sorting factor set is input into a post sorting decision function to calculate the sorting score of each post, a dynamic quantile threshold of the sorting score is set, posts with a sorting score lower than the dynamic quantile threshold of the sorting score are filtered, and the remaining post sorting scores are normalized to generate a post recommendation priority sorting list; The target post in the post recommendation priority sorting list is subjected to similarity calculation with the capability feature vector, and weighted fusion is performed in combination with the sorting weight to generate a final post score; The final post scores are sorted in descending order, and the post with the highest final post score is selected to generate a personalized post recommendation list.

7. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor executes the computer program to realize the steps of the employment matching method based on data analysis in any one of claims 1-6.

8. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the employment matching method based on data analysis in any one of claims 1-6.

Images