A real-time big data driven labor market supply and demand prediction and intelligent matching system

Abstract

The application relates to the technical field of data management, and discloses a real-time big data driven labor market supply and demand prediction and intelligent matching system, which comprises a feature separation module, a space-time feature matrix acquisition module, a supply and demand gap prediction module, a dynamic knowledge graph generation module, an optimal matching path acquisition module and a labor resource allocation module, wherein: dynamic features of an original data stream are extracted and separated to obtain a standardized feature vector set; the standardized feature vector set is subjected to space-time alignment to obtain a space-time feature matrix; time dependence is captured, space neighborhood information is aggregated, a regionalized supply and demand gap prediction value is obtained to construct a basic graph skeleton, and entity dynamic attributes in the original data stream are mapped to the basic graph skeleton to obtain a dynamic knowledge graph; an optimal matching path set is generated; the optimal matching path set is subjected to desensitization treatment to obtain a labor resource allocation scheme; and the application can improve the rationality of labor resource allocation.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to a real-time big data-driven labor market supply and demand forecasting and intelligent matching system. Background Technology

[0002] In the field of labor market supply and demand management, existing technologies lack systematic methods for extracting and separating dynamic features when processing massive amounts of raw data streams, and have failed to form an effective standardized processing mechanism. This results in poor compatibility between different types of data, such as job postings, job seeker behavior, and salary trends, making it difficult to build a unified analytical foundation. Consequently, the perception of dynamic market changes is lagging and one-sided, failing to accurately reflect the true supply and demand situation.

[0003] Meanwhile, existing technologies have significant limitations in integrating spatiotemporal information and matching supply and demand. On the one hand, their ability to capture time dependence and aggregate spatial neighborhood information is insufficient, resulting in low accuracy in predicting regional supply and demand gaps. On the other hand, the lack of dynamically updated knowledge graph support and multi-dimensional accurate evaluation systems leads to insufficient optimization of matching paths between supply and demand entities, making it difficult to achieve efficient and rational allocation of labor resources, resulting in resource mismatch and low allocation efficiency. Summary of the Invention

[0004] This invention provides a real-time big data-driven labor market supply and demand forecasting and intelligent matching system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a real-time big data-driven labor market supply and demand forecasting and intelligent matching system, characterized in that the system includes a feature separation module, a spatiotemporal feature matrix acquisition module, a supply and demand gap prediction module, a dynamic knowledge graph generation module, an optimal matching path acquisition module, and a labor resource allocation module, wherein:

[0006] The feature separation module is used to extract the dynamic features of the original data stream in the labor market and separate the parallelized dynamic features to obtain a standardized feature vector set of the labor market.

[0007] The spatiotemporal feature matrix acquisition module is used to perform spatiotemporal alignment of the standardized feature vector set based on a sliding time window to obtain the spatiotemporal feature matrix of the labor market.

[0008] The supply-demand gap prediction module is used to capture the time dependence of the spatiotemporal feature matrix and aggregate the spatial neighborhood information of the time dependence to obtain the regionalized supply-demand gap prediction value of the labor market.

[0009] The dynamic knowledge graph generation module is used to construct the basic knowledge graph skeleton of the labor market based on the regionalized supply and demand gap prediction value and the entity relationship triplet of the labor field knowledge base, and to map the dynamic attributes of entities in the original data stream to the basic knowledge graph skeleton to obtain the dynamic knowledge graph of the labor market.

[0010] The optimal matching path acquisition module is used to generate the optimal matching path set of the labor market based on the multi-dimensional matching degree score between supply and demand entities in the dynamic knowledge graph.

[0011] The labor resource allocation module is used to perform desensitization processing on the optimal matching path set to obtain the labor resource allocation scheme of the labor market.

[0012] In a preferred embodiment, when the feature separation module extracts dynamic features from the raw data stream in the labor market and separates the parallelized dynamic features to obtain a standardized feature vector set of the labor market, it is specifically used for:

[0013] Separate job posting data, job seeker behavior data, and salary trend data from the raw data stream of the labor market;

[0014] By statistically analyzing the job posting data in the sliding window, the characteristic vector of job demand fluctuation in the labor market is obtained.

[0015] Semantic analysis is performed on the job seeker behavior data to obtain the job intention intensity feature vector of the labor market;

[0016] The long-term trend component and short-term fluctuation component of the wage trend data are separated to obtain the dynamic feature vector of wages in the labor market.

[0017] The job demand fluctuation feature vector, job intention intensity feature vector, and salary dynamic feature vector are standardized and scaled to obtain the standardized feature vector set of the labor market.

[0018] In a preferred embodiment, when the spatiotemporal feature matrix acquisition module performs spatiotemporal alignment of the standardized feature vector set based on a sliding time window to obtain the spatiotemporal feature matrix of the labor market, it is specifically used for:

[0019] Identify the timestamp and geographic location markers in the standardized feature vector set to obtain the spatiotemporal coordinate index of the standardized feature vector set;

[0020] Based on the regional division rules of the labor market, the spatiotemporal coordinate index is mapped to a predefined spatial grid;

[0021] Within the sliding time window, the feature vectors of the spatial grid are weighted and fused to obtain the grid feature blocks of the spatial grid;

[0022] The grid feature blocks are smoothed according to their spatial topological relationships;

[0023] The grid feature blocks after boundary smoothing are arranged in time window order to obtain the spatiotemporal feature matrix of the labor market.

[0024] In a preferred embodiment, when the supply-demand gap prediction module captures the time dependency of the spatiotemporal feature matrix and aggregates the spatial neighborhood information of the time dependency, it is specifically used for:

[0025] A bidirectional recursive feature scan is performed on the spatiotemporal feature matrix along the time dimension to obtain the temporal context-dependent features of the spatiotemporal feature matrix.

[0026] Using the labor market regions under the aforementioned regional division rules as nodes and the economic correlation strength of the labor market regions as edges, a spatial topology graph of the labor market is constructed.

[0027] On the spatial topology map, the temporal context-dependent features of the core region are iteratively fused with the features of the neighboring regions in the spatial topology map to obtain the spatiotemporal fusion feature tensor of the labor market;

[0028] The spatiotemporal fusion feature tensor is input into the fully connected decision layer to obtain the predicted value of the regional supply and demand gap in the labor market.

[0029] In a preferred embodiment, when the supply-demand gap forecasting module performs the aggregation of the time-dependent spatial neighborhood information to obtain the regionalized supply-demand gap forecast value of the labor market, it is specifically used for:

[0030] Extract feature blocks from the target region and its first-order neighborhood regions in the spatiotemporal feature matrix to obtain the neighborhood feature set of the labor market;

[0031] The feature vectors in the neighborhood feature set are subjected to feature transformation, wherein the calculation formula for the feature transformation is as follows:

[0032] ;

[0033] In the formula, For the region Feature vector after feature transformation It is a non-linear activation function. For the region ordinal number, For the region The set of neighboring regions, For the region and Spatial weighting coefficients between them For the region The time-dependent feature vector, For the region Spatial feature vectors For bias terms, It is an element-wise product operator;

[0034] By performing regression dimensionality reduction on the feature vector after feature transformation, the regionalized supply and demand gap prediction value of the labor market is obtained.

[0035] In a preferred embodiment, when the dynamic knowledge graph generation module constructs the basic knowledge graph skeleton of the labor market based on the predicted regional supply and demand gap and entity relationship triples from the labor sector knowledge base, it is specifically used for:

[0036] Extract the enterprise-job-skill three-level entity relationship chain from the labor force knowledge base to obtain the core triple set of the labor market;

[0037] The predicted value of the regional supply and demand gap is bound to the corresponding geographical entity node in the labor market, and used as the supply and demand attribute label of the geographical entity node.

[0038] Based on the strength of the economic linkage, labor flow edges are established between adjacent geographical region entity nodes;

[0039] By integrating the core triplet set, the supply and demand attribute tags, and the labor mobility edge, the basic graph framework of the labor market is constructed.

[0040] In a preferred embodiment, when the dynamic knowledge graph generation module performs the operation of mapping the dynamic attributes of entities in the original data stream to the basic graph skeleton to obtain the dynamic knowledge graph of the labor market, it is specifically used for:

[0041] Real-time capture of job demand change rate, job seeker skill update markers, and salary fluctuation values ​​in the raw data stream yields a dynamic attribute vector of the labor market.

[0042] Associate the dynamic attribute vector with the corresponding entity node in the basic graph skeleton;

[0043] The current attribute value of the dynamic attribute vector is weighted and fused with the corresponding historical attribute value, and the fused attribute value is injected into the entity node to obtain the dynamic knowledge graph of the labor market.

[0044] In a preferred embodiment, when the optimal matching path acquisition module generates the optimal matching path set for the labor market based on the multi-dimensional matching degree scores between supply and demand entities in the dynamic knowledge graph, it is specifically used for:

[0045] Identify key supply and demand entity nodes in the dynamic knowledge graph;

[0046] Establish multi-dimensional evaluation channels among the key supply and demand entities;

[0047] The comprehensive matching score of potential paths in the multi-dimensional evaluation channel is calculated using the path value evaluation function;

[0048] Based on the comprehensive matching score, conflicting path branches in the potential paths are eliminated to obtain the non-dominant paths in the labor market;

[0049] The Pareto front sorting of the non-dominated paths yields the optimal matching path set for the labor market.

[0050] In a preferred embodiment, when the optimal matching path acquisition module calculates the comprehensive matching score of potential paths in the multi-dimensional evaluation channel using the path value evaluation function, it is specifically used for:

[0051] Extract the skill feature vector and demand feature vector of the key supply and demand entity nodes to obtain the feature difference tensor of the key supply and demand entity nodes;

[0052] Based on the feature difference tensor, the basic matching degree of the potential paths in the multi-dimensional evaluation channel is calculated, wherein the formula for calculating the basic matching degree is as follows:

[0053] ;

[0054] In the formula, The score for the basic matching degree. The total number of skill dimensions. For skill dimension ordinal numbers, For job seekers The level values ​​of the feature difference tensor. For the recruiter The required value of the feature difference tensor. For the first Industry weight coefficients of the feature difference tensor;

[0055] An economic environment adjustment factor is introduced to correct the basic matching degree, and the final matching degree score of the potential path is obtained.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. This invention uses a feature separation module to accurately extract and separate the raw data stream of the labor market, obtaining a standardized set of feature vectors. Combined with a spatiotemporal feature matrix acquisition module, spatiotemporal alignment is achieved, providing a high-quality data foundation for subsequent analysis. The supply-demand gap prediction module effectively captures time dependencies and aggregates spatial neighborhood information, improving the accuracy of regionalized supply-demand gap prediction and enabling a more precise grasp of market supply and demand changes.

[0058] 2. This invention constructs and updates a graph containing dynamic attributes of entities through a dynamic knowledge graph generation module, generates the optimal matching path based on multi-dimensional matching degree scores through an optimal matching path acquisition module, and forms a configuration scheme through de-identification processing through a labor resource allocation module. The entire process realizes efficient matching of supply and demand entities and rational allocation of resources, significantly improving the scientificity and rationality of labor resource allocation and enhancing the effectiveness of market supply and demand regulation. Attached Figure Description

[0059] Figure 1 A system architecture diagram of a real-time big data-driven labor market supply and demand forecasting and intelligent matching system provided in an embodiment of the present invention;

[0060] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, 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.

[0062] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0063] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0064] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0065] In practice, the server-side equipment deployed in a real-time big data-driven labor market supply and demand forecasting and intelligent matching system may consist of one or more devices. This system can be implemented as a business instance, a virtual machine, or hardware devices. For example, it can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, it can be understood as software deployed on a cloud node, providing real-time big data-driven labor market supply and demand forecasting and intelligent matching to various users. Alternatively, it can be implemented as a virtual machine deployed on one or more devices in a cloud node, with application software installed to manage various users. Or, it can also be implemented as a server-side system composed of numerous identical or different types of hardware devices, with one or more devices used to provide real-time big data-driven labor market supply and demand forecasting and intelligent matching to various users.

[0066] In terms of implementation, the real-time big data-driven labor market supply and demand forecasting and intelligent matching system and the user terminal are mutually compatible. That is, if the real-time big data-driven labor market supply and demand forecasting and intelligent matching system is implemented as an application installed on a cloud service platform, then the user terminal is implemented as a client that establishes a communication connection with the application; or if the real-time big data-driven labor market supply and demand forecasting and intelligent matching system is implemented as a website, then the user terminal is implemented as a webpage; or if the real-time big data-driven labor market supply and demand forecasting and intelligent matching system is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0067] like Figure 1 The figure shown is a system architecture diagram of a real-time big data-driven labor market supply and demand forecasting and intelligent matching system provided in an embodiment of the present invention.

[0068] The real-time big data-driven labor market supply and demand forecasting and intelligent matching system 100 described in this invention can be set up in a cloud server. In terms of implementation, it can be implemented as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed as a website. Depending on the functions implemented, the real-time big data-driven labor market supply and demand forecasting and intelligent matching system 100 may include a feature separation module 101, a spatiotemporal feature matrix acquisition module 102, a supply and demand gap prediction module 103, a dynamic knowledge graph generation module 104, an optimal matching path acquisition module 105, and a labor resource allocation module 106. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by an electronic device's processor and perform a fixed function, stored in the electronic device's memory.

[0069] In this embodiment of the invention, in the real-time big data-driven labor market supply and demand forecasting and intelligent matching system, each of the above modules can be implemented independently and can call other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. In the real-time big data-driven labor market supply and demand forecasting and intelligent matching system provided by this embodiment of the invention, the applicable scope of the system architecture can be adjusted by adding modules and directly calling them without modifying the program code, achieving cluster-based horizontal expansion to quickly and flexibly expand the real-time big data-driven labor market supply and demand forecasting and intelligent matching system. In practical applications, the above modules can be set in the same device or different devices, or they can be set in virtual devices, such as service instances in a cloud server.

[0070] The following describes the components and specific workflow of a real-time big data-driven labor market supply and demand forecasting and intelligent matching system, using specific examples:

[0071] The feature separation module 101 is used to extract the dynamic features of the original data stream in the labor market and separate the parallelized dynamic features to obtain a standardized feature vector set of the labor market.

[0072] In this embodiment of the invention, when the feature separation module extracts the dynamic features of the raw data stream in the labor market and separates the parallelized dynamic features to obtain a standardized feature vector set of the labor market, it is specifically used for:

[0073] Separate job posting data, job seeker behavior data, and salary trend data from the raw data stream of the labor market;

[0074] By statistically analyzing the job posting data in the sliding window, the characteristic vector of job demand fluctuation in the labor market is obtained.

[0075] Semantic analysis is performed on the job seeker behavior data to obtain the job intention intensity feature vector of the labor market;

[0076] The long-term trend component and short-term fluctuation component of the wage trend data are separated to obtain the dynamic feature vector of wages in the labor market.

[0077] The job demand fluctuation feature vector, job intention intensity feature vector, and salary dynamic feature vector are standardized and scaled to obtain the standardized feature vector set of the labor market.

[0078] Specifically, when separating job posting data, job seeker behavior data, and salary trend data from the raw data stream of the labor market, it is necessary to first sort out all the field information contained in the raw data. Job posting data should include fields such as job title, full name of the recruiting company, industry of the company, work location (accurate to city or district), required education level (e.g., high school, bachelor's, master's), required years of work experience, job description, application deadline, and planned number of hires. Job seeker behavior data should extract fields such as the job seeker's registration ID, keywords used in each search (e.g., "marketing specialist" or "software engineer"), search time (accurate to hour), browsing time for each job page (time difference from clicking to leaving), whether the job was saved, the specific job ID submitted to the resume, submission time, expected job type and expected salary range stated in the resume. Salary trend data should include fields such as the salary range for each job posting (e.g., "8000-12000 yuan / month"), salary calculation method (e.g., pre-tax or post-tax), whether performance bonuses are included and their distribution rules, welfare and subsidy information (e.g., housing allowance, meal allowance), and the release time of the salary standard. By matching the field attributes one by one, the data should be accurately divided into three categories.

[0079] Furthermore, when statistically analyzing the recruitment position data in a sliding window to obtain the job demand fluctuation feature vector, first determine the duration of the sliding window. For example, set a 7-day window in units of natural weeks, and the windows do not overlap. Sort the recruitment position data according to their release time and include them in the corresponding windows in sequence. If the release time of a certain position is at the junction of two windows, it is classified into the earlier window. For each window, count the total number of new positions in the window, the number of positions in each industry (such as Internet, manufacturing, finance), the number of positions corresponding to each educational requirement (such as the number of positions with a bachelor's degree or above), the number of positions with each work experience requirement (such as the number of positions with 3-5 years of experience), the number of positions with the "urgent recruitment" label, etc. Then arrange the statistical results of all windows in chronological order, and each index of each window serves as a dimension of the vector, thus forming the job demand fluctuation feature vector.

[0080] Furthermore, when performing semantic analysis on the job seeker behavior data to obtain the job intention intensity feature vector, first extract the text information from the job seeker behavior data, including the job keywords searched, the expected job descriptions filled in the resume, the personal profiles when applying for jobs, etc. Process this text, remove meaningless function words (such as "of", "in"), and retain the core nouns and verbs (such as "accounting", "full-time", "promotion"). Determine the high-frequency core words by counting the occurrence times of the same words (such as "five social insurances and one housing fund", "weekends off" appear the most times), and at the same time analyze the associations between high-frequency words (such as "software development" often appears with "Java"). Combine the behavior indicators, including the proportion of the search times of a certain type of job keyword in the total search times, the average browsing duration of jobs containing specific core words, the ratio of the application times to the browsing times for a certain type of job (application conversion rate), etc. Sort these word association data and behavior indicators according to the preset dimensions (such as job type, welfare requirements, working mode), and the combined vector is the job intention intensity feature vector.

[0081] Furthermore, to obtain the salary dynamic feature vector by separating the long-term trend component and short-term fluctuation component of salary trend data, the salary trend data is first sorted by salary release time, with the time unit accurate to the month. When calculating the long-term trend component, a moving average method is used, with a continuous three-month period as one cycle. For example, when calculating the long-term trend component for March, the average salary of January, February, and March is used (averaging the median of the salary range for each month); the long-term trend component for April is the average salary of February, March, and April, and so on, ensuring that each month has a corresponding long-term trend component. The short-term fluctuation component is the actual average salary for each month (the average of the median of the salary range for all positions in that month) minus the long-term trend component for that month. If the actual average salary is higher than the long-term trend component, the short-term fluctuation component is positive; otherwise, it is negative. The long-term trend components of all months are arranged in chronological order, and the corresponding short-term fluctuation components are arranged in the same order. The vector formed by combining these two sequences is the salary dynamic feature vector.

[0082] Furthermore, when standardizing and scaling the job demand fluctuation feature vector, job intention intensity feature vector, and salary dynamic feature vector to obtain a standardized feature vector set, each vector is processed separately. Taking the job demand fluctuation feature vector as an example, first find the maximum and minimum values ​​of all dimensions in this vector. For example, in the dimension of "number of jobs in the Internet industry," the maximum value for each window is 500, and the minimum value is 100. Calculate for each data point under this dimension: subtract the minimum value from the value of each window, and then divide by the difference between the maximum and minimum values ​​(e.g., if the value of this dimension for a certain window is 300, then calculate (300-100) / (500-100) = 0.5), so that all data in this dimension are between 0 and 1. Following the same method, process the other dimensions of the job demand fluctuation feature vector, all dimensions of the job intention intensity feature vector, and all dimensions of the salary dynamic feature vector respectively. After processing, these three vectors together constitute the standardized feature vector set of the labor market.

[0083] In summary, by separating job posting data, job seeker behavior data, and salary trend data, key features such as fluctuations in job demand, intensity of job intentions, and salary dynamics can be accurately extracted, enabling targeted analysis of core elements of the labor market and ensuring the professionalism and relevance of the feature data.

[0084] In summary, standardizing and scaling the extracted feature vectors eliminates the differences in dimensions and scales between different types of data. This allows features such as job requirements, job seekers' intentions, and salary trends to be compared and integrated under a unified dimension, providing a consistent data foundation for subsequent spatiotemporal alignment, supply and demand gap prediction, and other processes, thereby improving the accuracy and effectiveness of data processing and analysis.

[0085] In summary, this process provides high-quality input data for subsequent modules of the entire system. The standardized set of feature vectors can be more efficiently utilized by the spatiotemporal feature matrix acquisition module, laying the foundation for building an accurate spatiotemporal feature matrix, thereby promoting the smooth operation of the entire labor market supply and demand forecasting and intelligent matching process.

[0086] The spatiotemporal feature matrix acquisition module 102 is used to perform spatiotemporal alignment of the standardized feature vector set based on a sliding time window to obtain the spatiotemporal feature matrix of the labor market.

[0087] In this embodiment of the invention, when the spatiotemporal feature matrix acquisition module performs spatiotemporal alignment of the standardized feature vector set based on a sliding time window to obtain the spatiotemporal feature matrix of the labor market, it is specifically used for:

[0088] Identify the timestamp and geographic location markers in the standardized feature vector set to obtain the spatiotemporal coordinate index of the standardized feature vector set;

[0089] Based on the regional division rules of the labor market, the spatiotemporal coordinate index is mapped to a predefined spatial grid;

[0090] Within the sliding time window, the feature vectors of the spatial grid are weighted and fused to obtain the grid feature blocks of the spatial grid;

[0091] The grid feature blocks are smoothed according to their spatial topological relationships;

[0092] The grid feature blocks after boundary smoothing are arranged in time window order to obtain the spatiotemporal feature matrix of the labor market.

[0093] Specifically, the time and geographic location information associated with each vector is extracted from the standardized feature vector set. The timestamp is the minute-accurate time when the data was recorded, such as "2025-08-10 09:30". The geographic location is the specific administrative region code of the job or job seeker, such as the six-digit county-level administrative region code "310104". The timestamp and geographic location code corresponding to each standardized feature vector are combined one-to-one to form the spatiotemporal coordinate index of the vector. For example, the spatiotemporal coordinate index of a vector is "2025-08-10 09:30+310104".

[0094] Furthermore, the labor market's regional division rule involves dividing the target area into fixed-size square spatial grids based on latitude and longitude. Each grid has a side length of 1 kilometer, and each grid is assigned a unique grid number, such as "G001, G002...G100" from left to right and top to bottom. Based on the geographic location markers in the spatiotemporal coordinate index, the spatiotemporal coordinate index is mapped to the grid number by querying which predefined spatial grid the latitude and longitude coordinates of that geographic location fall into. For example, if the geographic location with latitude and longitude coordinates (121.45°E, 31.23°N) falls into grid "G056", then the corresponding spatiotemporal coordinate index is mapped to "G056".

[0095] Furthermore, the sliding time window is set to a duration of 24 hours, with windows being consecutive and non-overlapping. Each window starts at 00:00 daily and ends at 00:00 the following day. Within each sliding time window, all standardized feature vectors mapped to the same spatial grid are collected. Weights are assigned based on the actual influence of each dimension's feature value within the feature vector, with the weights for the job demand fluctuation feature vector (0.4), job intention intensity feature vector (0.3), and salary dynamic feature vector (0.3) all assigned to it. After multiplying each dimension's feature value of each feature vector by its corresponding weight, the arithmetic mean of the same dimension's feature values ​​for all vectors within the same spatial grid is taken to obtain the grid feature block for that spatial grid within the current time window.

[0096] Furthermore, spatial topological relationships refer to the geographical connections between adjacent spatial grids, meaning each spatial grid forms a topological association with its four adjacent grids in the four directions (up, down, left, and right). When performing boundary smoothing on grid feature blocks, the dimensional feature values ​​closest to the boundary in each grid feature block are extracted. These values ​​are then averaged with the dimensional feature values ​​of the corresponding boundaries in adjacent grid feature blocks. This average value replaces the feature values ​​at the boundaries of the original grid feature block. For example, the feature value of the right boundary of grid "G056" is averaged with the feature value of the left boundary of its right-adjacent grid "G057," and then the feature value of the right boundary of "G056" is updated, ensuring a continuous transition of boundary feature values ​​between adjacent grid feature blocks.

[0097] Furthermore, the grid feature blocks of all spatial grids after boundary smoothing are arranged in the order of their corresponding sliding time windows. All grid feature blocks within each time window are taken as a row of the matrix, and the feature values ​​of each dimension in each grid feature block are taken as the column elements of that row. The resulting matrix containing time and space dimensions is the spatiotemporal feature matrix of the labor market.

[0098] In summary, by identifying timestamp and geolocation markers in a standardized feature vector set, mapping them to a predefined spatial grid, and performing feature fusion and boundary smoothing within a sliding time window, the originally scattered feature vectors are transformed into a matrix structure with clear spatiotemporal coordinates. This strengthens the inherent correlation between labor market data in terms of time series and spatial distribution, and provides a structured data foundation for subsequently capturing time dependencies and aggregating spatial neighborhood information.

[0099] In summary, the dynamic adjustment mechanism of the sliding time window can adapt to real-time changes in the labor market and ensure that the feature vectors included in the matrix always reflect the latest market dynamics. The division of spatial grids and feature weighting fusion not only preserve the feature differences between different regions, but also reduce the interference of feature mutations between regions through boundary smoothing. This enables the spatiotemporal feature matrix to more stably and comprehensively represent the overall state of the labor market, providing reliable support for accurately predicting regional supply and demand gaps.

[0100] The supply and demand gap prediction module 103 is used to capture the time dependence of the spatiotemporal feature matrix and aggregate the spatial neighborhood information of the time dependence to obtain the regionalized supply and demand gap prediction value of the labor market.

[0101] In this embodiment of the invention, when the supply-demand gap prediction module captures the time dependency of the spatiotemporal feature matrix and aggregates the spatial neighborhood information of the time dependency, it is specifically used for:

[0102] A bidirectional recursive feature scan is performed on the spatiotemporal feature matrix along the time dimension to obtain the temporal context-dependent features of the spatiotemporal feature matrix.

[0103] Using the labor market regions under the aforementioned regional division rules as nodes and the economic correlation strength of the labor market regions as edges, a spatial topology graph of the labor market is constructed.

[0104] On the spatial topology map, the temporal context-dependent features of the core region are iteratively fused with the features of the neighboring regions in the spatial topology map to obtain the spatiotemporal fusion feature tensor of the labor market;

[0105] The spatiotemporal fusion feature tensor is input into the fully connected decision layer to obtain the predicted value of the regional supply and demand gap in the labor market.

[0106] When the supply-demand gap prediction module aggregates the time-dependent spatial neighborhood information to obtain the regionalized supply-demand gap prediction value of the labor market, it is specifically used for:

[0107] Extract feature blocks from the target region and its first-order neighborhood regions in the spatiotemporal feature matrix to obtain the neighborhood feature set of the labor market;

[0108] The feature vectors in the neighborhood feature set are subjected to feature transformation, wherein the calculation formula for the feature transformation is as follows:

[0109] ;

[0110] In the formula, For the region Feature vector after feature transformation It is a non-linear activation function. For the region ordinal number, For the region The set of neighboring regions, For the region and Spatial weighting coefficients between them For the region The time-dependent feature vector, For the region Spatial feature vectors For bias terms, It is an element-wise product operator;

[0111] By performing regression dimensionality reduction on the feature vector after feature transformation, the regionalized supply and demand gap prediction value of the labor market is obtained.

[0112] Specifically, when performing a bidirectional recursive feature scan of the spatiotemporal feature matrix along the time dimension, the scan first proceeds forward from the first window, linking the spatiotemporal features of the current window with the features of the previous window and recording the correlation information of the features changing forward over time. After completing the forward scan, the scan proceeds backward from the last window, linking the spatiotemporal features of the current window with the features of the next window and recording the correlation information of the features changing backward over time. The correlation information obtained from the forward and backward scans is then merged. The features of each time window contain the feature influence information of its preceding and following adjacent windows. The resulting feature set containing temporal dependencies is the temporal context-dependent feature of the spatiotemporal feature matrix.

[0113] Furthermore, each spatial grid under the regional division rules is used as a node of the labor market region. Each node is assigned a unique identifier containing the grid's latitude and longitude range information. The economic linkage strength between labor market regions is determined by statistically analyzing the number of cross-regional job postings, the number of job seekers submitting resumes across regions, and the number of job postings jointly released by companies in both regions. The sum of these quantities is taken as the numerical value of the economic linkage strength between the two regions; the higher the value, the stronger the linkage. Nodes represent regions, and line segments with this numerical weight connect nodes with economic linkages. The thickness of the line segments corresponds to the magnitude of the linkage strength, thereby constructing a spatial topology map of the labor market.

[0114] Furthermore, in the spatial topology graph, the core region is defined as the top 20% of nodes with the highest sum of economic linkage strength. First, the temporal context-dependent features of the core region are extracted, along with the features of its directly adjacent neighboring regions (nodes directly connected by an edge). The core region features and neighboring region features are then fused at a 50 / 50 ratio to obtain the first fused feature. Next, the first fused feature is used as the new core region feature, and fused with the features of the neighboring regions (nodes connected two steps away from the core region) at the same ratio. This process is repeated three times, with each fusion incorporating features from more distant neighboring regions. The resulting multidimensional data structure, containing the fused temporal features of the core region and multiple layers of neighboring regions, is the spatiotemporal fused feature tensor of the labor market.

[0115] Furthermore, the spatiotemporal fusion feature tensor is input into the fully connected decision layer, which consists of three processing levels. The first level receives the data of each dimension of the spatiotemporal fusion feature tensor, multiplies the data of each dimension with preset fixed coefficients, and then adds them together to obtain the output of the first level. The second level receives the output of the first level, and uses the same processing method as the first level, multiplying the result with another set of fixed coefficients and then adding them together to obtain the output of the second level. The third level receives the output of the second level, and by judging the sign and magnitude of the result, directly outputs the difference between the number of job vacancies and the number of job seekers in each spatial grid area. This difference is the predicted value of the regional supply and demand gap in the labor market. A positive value indicates that the job vacancies are greater than the number of job seekers, and a negative value indicates that the number of job seekers is greater than the job vacancies.

[0116] Specifically, the target region is determined from the spatiotemporal feature matrix as a pre-defined spatial grid to be analyzed. This grid corresponds to specific row (time window) and column (spatial dimension) positions in the matrix. The first-order neighborhood regions are the four grids directly adjacent to the target region in space, namely the grids directly above, below, to the left, and to the right of the target region, and these neighborhood regions also have corresponding positions in the spatiotemporal feature matrix. Feature blocks of the target region are extracted across all time windows, and each feature block contains all dimensional feature values ​​of the region for each time window. Simultaneously, feature blocks of the four first-order neighborhood regions within the same time window are extracted, and each neighborhood region's feature block also contains all dimensional feature values ​​of its respective time window. The feature blocks of the target region and the feature blocks of the four first-order neighborhood regions are combined to form a set of five feature blocks, which constitutes the neighborhood feature set of the labor market.

[0117] Furthermore, feature transformation is performed on each feature vector in the neighborhood feature set. The feature vector is a sequence of feature values ​​of each dimension arranged in the time window order in each feature block. The transformation method is to reorganize the feature values ​​of each dimension in each feature vector according to a preset category. The feature values ​​of dimensions related to job demand fluctuations are grouped into one category and their sum is calculated. The feature values ​​of dimensions related to job intention intensity are grouped into another category and their sum is calculated. The feature values ​​of dimensions related to salary dynamics are grouped into another category and their sum is calculated. Then, the sums of these three categories are rearranged in the order of "total sum of job demand - total sum of job intention - total sum of salary dynamics" to form a new feature vector, thus completing the feature transformation.

[0118] Furthermore, the feature vector after feature transformation is subjected to regression dimensionality reduction. The core factors influencing the regional supply-demand gap are first identified as the total job demand and the total job intention, ignoring the dimension of dynamic total salary, retaining only the two dimensions of total job demand and total job intention. The difference between these two dimensions is calculated by subtracting the total job intention from the total job demand for each time window; the difference is the supply-demand gap value for the target region within that time window. The supply-demand gap values ​​for all time windows are arranged chronologically, and the value of the latest time window is taken as the predicted value of the regional supply-demand gap in the labor market. A positive value indicates that job demand exceeds the number of job seekers, and a negative value indicates that the number of job seekers exceeds job demand.

[0119] Specifically, As a non-linear activation function, it is derived from a pre-defined standard function. The sigmoid function is selected, which can map the input value to the range of 0 to 1. Specifically, the output is obtained by substituting the input value into the calculation rules of the sigmoid function. That is, for any input value, it is calculated in the manner of "1 divided by (1 plus the negative power of the natural constant)".

[0120] Furthermore, As a regional ordinal number, its origin is based on the region Neighborhood region set Each region in the list is sequentially numbered, starting from 1, according to neighboring regions and regions. The distances increase sequentially from near to far, and each region corresponds to a unique ordinal number.

[0121] Furthermore, As a region The set of neighboring regions is derived by selecting regions from the set of neighborhood features. Areas that are directly adjacent in space, specifically regions The four areas directly above, directly below, directly to the left, and directly to the right constitute... .

[0122] Furthermore, As a region and The spatial weighting coefficients between regions are derived from regional factors. With the region The strength of the economic ties between the two regions is determined by summing the number of cross-regional recruitments, the number of cross-regional resumes submitted, and the number of job postings by cooperating companies. The greater the total, the stronger the economic ties. The larger the value, specifically, the larger the sum is when divided by all neighboring regions. The ratio obtained by summing the strengths of economic ties is _____. .

[0123] Furthermore, As a region The time-dependent feature vectors originate from the region. Extracted from temporal context dependency features, which are obtained by analyzing the region... The spatiotemporal feature matrix is ​​obtained by bidirectional recursive feature scanning, containing the region. The correlation information between the features of each time window and the features of the preceding and following windows is used to extract the feature values ​​of each dimension in the order of the time windows to form... .

[0124] Furthermore, As a region The spatial feature vectors originate from the region. The mesh feature block is a region within a sliding time window. The standardized feature vectors are obtained by weighted fusion of feature values, and the region is included. Each feature index in the spatial dimension is used to extract feature values ​​for each dimension to form... .

[0125] Furthermore, As a bias term, it is derived from a pre-set fixed value, which is determined based on the benchmark value of regional feature transformation in historical data. It is used to adjust the benchmark for sum calculation so that the result is closer to the actual feature distribution.

[0126] Furthermore, As an element-wise product operator, its function is to... and Multiply elements that are in the same position in the matrix, i.e. The first element and Multiply the first element, and multiply the second element by the first element. Multiply by the second element, and so on, to obtain a new vector where each element is a corresponding position. and The product of elements.

[0127] Furthermore, the significance of the formula lies in calculating the region. All neighboring areas Time-dependent feature vectors With spatial feature vectors The element-wise product is then multiplied by the region. and Spatial weight coefficient Multiply them, then sum the results of all neighboring regions, and add the bias term. Finally, the sum is substituted into the nonlinear activation function. Processing is performed to obtain the region. Feature vector after feature transformation This process integrates the spatiotemporal characteristics of neighboring regions and adjusts the influence of each neighboring region based on the strength of spatial correlation, thereby achieving regional... Effective transformation of features.

[0128] Furthermore, the trend of the formula is that when the neighboring region With the region The stronger the economic ties, the greater the corresponding The larger the value, and The higher the proportion of element-wise product in the sum, the greater its impact on the final result. The features are closer to the spatiotemporal features of the neighborhood region; when and When the element-wise product is large, the sum will increase, plus... After deal with, The values ​​of each element will increase accordingly (approaching 1 under the sigmoid function); when the neighboring region and The correlation strength is weak, or and When the product value is small, its contribution to the sum is small. The element values ​​will decrease accordingly (approaching 0 under the sigmoid function), overall... The changing trend is consistent with the spatiotemporal characteristic changing trend of neighboring regions with strong correlation.

[0129] In summary, improving the time dynamic accuracy of forecasts: By performing bidirectional recursive feature scanning of the spatiotemporal feature matrix along the time dimension, it is possible to effectively capture the dependencies of labor market data over time, such as the periodic fluctuations in job demand and the trend changes in job search intentions, making the forecast results more in line with the dynamic evolution of the market and avoiding the lag error caused by static analysis.

[0130] In summary, a spatial topology map is constructed with regions as nodes and the intensity of economic linkages as edges. By iteratively integrating the characteristics of core regions and neighboring regions, the spatial spillover effect of the labor market (such as the mutual influence of supply and demand in adjacent regions) is fully considered.

[0131] In summary, by accurately processing neighborhood features based on feature transformation formulas and combining them with regionalized prediction values ​​obtained through regression dimensionality reduction, the supply and demand differences in different regions can be accurately reflected, providing a precise quantitative basis for subsequent targeted resource allocation.

[0132] The dynamic knowledge graph generation module 104 is used to construct the basic knowledge graph skeleton of the labor market based on the regionalized supply and demand gap prediction value and the entity relationship triplet of the labor field knowledge base, and to map the dynamic attributes of entities in the original data stream to the basic knowledge graph skeleton to obtain the dynamic knowledge graph of the labor market.

[0133] In this embodiment of the invention, when the dynamic knowledge graph generation module executes the construction of the basic knowledge graph skeleton of the labor market based on the predicted regional supply and demand gap and the entity relationship triples of the labor field knowledge base, it is specifically used for:

[0134] Extract the enterprise-job-skill three-level entity relationship chain from the labor force knowledge base to obtain the core triple set of the labor market;

[0135] The predicted value of the regional supply and demand gap is bound to the corresponding geographical entity node in the labor market, and used as the supply and demand attribute label of the geographical entity node.

[0136] Based on the strength of the economic linkage, labor flow edges are established between adjacent geographical region entity nodes;

[0137] By integrating the core triplet set, the supply and demand attribute tags, and the labor mobility edge, the basic graph framework of the labor market is constructed.

[0138] When the dynamic knowledge graph generation module maps the dynamic attributes of entities in the original data stream to the basic graph skeleton to obtain the dynamic knowledge graph of the labor market, it is specifically used for:

[0139] Real-time capture of job demand change rate, job seeker skill update markers, and salary fluctuation values ​​in the raw data stream yields a dynamic attribute vector of the labor market.

[0140] Associate the dynamic attribute vector with the corresponding entity node in the basic graph skeleton;

[0141] The current attribute value of the dynamic attribute vector is weighted and fused with the corresponding historical attribute value, and the fused attribute value is injected into the entity node to obtain the dynamic knowledge graph of the labor market.

[0142] Specifically, the labor force knowledge base stores structured information about companies, positions, and skills, including the full name of the company and its industry, the name of the position, the company it belongs to, and its job content, and the name of the skill and the applicable positions. The relationship between companies and positions is extracted from this knowledge base, that is, the specific positions that each company recruits, forming a binary relationship of (company, recruitment, position). Then, the relationship between positions and skills is extracted, that is, the specific skills required for each position, forming a binary relationship of (position, requirement, skill). These two binary relationships are combined into a three-level entity relationship chain of company-position-skill. Each relationship chain contains three entities: company, position, and skill, and the relationships between them. The set formed by summing up all such relationship chains is the core triplet set of the labor market.

[0143] Furthermore, a geographic region entity node refers to a regional unit in the labor market with a clear geographic boundary, such as a district or county divided by administration. Each node has a unique geographic code. The regional supply and demand gap forecast value is matched with the corresponding geographic code, and a new field named "supply and demand attribute tag" is added to the attribute information of the geographic region entity node. The value of the field is the regional supply and demand gap forecast value of the region, so that each geographic region entity node is bound to the corresponding forecast value, thus completing the binding of the supply and demand attribute tag.

[0144] Furthermore, adjacent geographical region entity nodes refer to areas that are spatially adjacent. For example, if the administrative boundaries of region A and region B share a common part, then region A and region B are adjacent nodes. The strength of economic linkage between these adjacent nodes is calculated. This strength is determined by the sum of the number of cross-regional job postings and the number of cross-regional resumes submitted each month between the two locations. When the strength value is greater than 0, a connecting line segment is established between the two adjacent geographical region entity nodes. This line segment is the labor mobility edge, and the specific economic linkage strength value is recorded in the edge's attributes.

[0145] Furthermore, when integrating the core triple set, supply and demand attribute tags, and labor mobility edges, enterprises, positions, and skills in the core triple set are used as basic entity nodes, and the "recruitment" and "need" relationships between them are used as connecting edges between entities; geographical region entity nodes with supply and demand attribute tags are added to the system as nodes in the spatial dimension; and labor mobility edges are used as spatial connecting edges connecting adjacent geographical region entity nodes, so that entity nodes, attribute tags, and connecting edges form an interconnected network structure, which is the basic graph skeleton of the labor market.

[0146] Specifically, when capturing the job demand change rate in the raw data stream in real time, the number of newly added jobs each day is extracted from the raw data stream and compared with the total number of jobs the previous day. The ratio of the number of newly added jobs on that day to the total number of jobs on the previous day is calculated, and this ratio is the job demand change rate for that day. Job seeker skill update tags are obtained by monitoring modification records of skill fields in job seeker resume data. When a job seeker adds new skills, deletes old skills, or modifies skill proficiency in their resume, the system automatically records the operation and generates a skill update tag. The tag content includes the skill differences before and after the update and the update time. Salary fluctuation value is calculated by tracking the average salary of the same position over two consecutive weeks. The difference between the average salary of the current week and the average salary of the previous week is the salary fluctuation value for that position. Combining the daily job demand change rate, the weekly job seeker skill update tags, and the salary fluctuation value of the position in the order of "job demand change rate - skill update tag - salary fluctuation value" forms a vector that represents the dynamic attribute vector of the labor market.

[0147] Furthermore, when associating dynamic attribute vectors with corresponding entity nodes in the basic graph skeleton, the entity nodes in the basic graph skeleton include job nodes, skill nodes, and company nodes. The job demand change rate corresponds to a specific job node, found by uniquely combining the job name and company name in the basic graph skeleton; the job seeker skill update marker corresponds to a skill node, found by matching the updated skill name in the basic graph skeleton; the salary fluctuation value corresponds to the company node that posted the job, found by matching the company node in the basic graph skeleton using the company's full name. Each dynamic attribute vector's three components are associated with their corresponding entity nodes, and the association information includes the generation time of the dynamic attribute.

[0148] Furthermore, when weighted and fused with the current attribute value and corresponding historical attribute value of the dynamic attribute vector, the current attribute value is the value in the newly generated dynamic attribute vector, and the historical attribute values ​​are the dynamic attribute values ​​recorded in the past three times for that entity node. The current attribute value is assigned a weight of 60%, the first historical attribute value a weight of 20%, and the second and third historical attribute values ​​each a weight of 10%. These attribute values ​​are multiplied by their respective weights and then summed to obtain the fused attribute value. The fused attribute value is then used to update the attribute field of the corresponding entity node, overwriting the original attribute value. This allows the attributes of the entity nodes to change dynamically over time. After all entity nodes have completed the attribute updates, the resulting network structure containing dynamic attributes constitutes the dynamic knowledge graph of the labor market.

[0149] In summary, the structured knowledge framework is constructed by extracting the three-level entity relationship chain of enterprise-job-skill to form a core triplet set, binding geographical entity nodes with regional supply and demand gap forecasts, and establishing labor flow edges between adjacent regions. The constructed basic graph skeleton can clearly present the core entity relationships and regional supply and demand characteristics of the labor market, providing a structured knowledge foundation for market analysis.

[0150] In summary, by mapping the dynamic attributes of entities in the original data stream, such as the rate of change in job requirements, job seeker skill update markers, and salary fluctuation values, to the basic knowledge graph skeleton, and by injecting current and historical attribute values ​​into entity nodes through weighted fusion, the knowledge graph can reflect market dynamics in real time. It not only preserves the historical characteristics of entities but also captures the latest market information in a timely manner, providing dynamic and comprehensive knowledge support for supply and demand matching and improving the timeliness and accuracy of matching decisions.

[0151] The optimal matching path acquisition module 105 is used to generate the optimal matching path set of the labor market based on the multidimensional matching degree score between supply and demand entities in the dynamic knowledge graph.

[0152] In this embodiment of the invention, when the optimal matching path acquisition module generates the optimal matching path set for the labor market based on the multi-dimensional matching degree scores between supply and demand entities in the dynamic knowledge graph, it is specifically used for:

[0153] Identify key supply and demand entity nodes in the dynamic knowledge graph;

[0154] Establish multi-dimensional evaluation channels among the key supply and demand entities;

[0155] The comprehensive matching score of potential paths in the multi-dimensional evaluation channel is calculated using the path value evaluation function;

[0156] Based on the comprehensive matching score, conflicting path branches in the potential paths are eliminated to obtain the non-dominant paths in the labor market;

[0157] The Pareto front sorting of the non-dominated paths yields the optimal matching path set for the labor market.

[0158] When the optimal matching path acquisition module calculates the comprehensive matching score of potential paths in the multi-dimensional evaluation channel using the path value evaluation function, it is specifically used for:

[0159] Extract the skill feature vector and demand feature vector of the key supply and demand entity nodes to obtain the feature difference tensor of the key supply and demand entity nodes;

[0160] Based on the feature difference tensor, the basic matching degree of the potential paths in the multi-dimensional evaluation channel is calculated, wherein the formula for calculating the basic matching degree is as follows:

[0161] ;

[0162] In the formula, The score for the basic matching degree. The total number of skill dimensions. For skill dimension ordinal numbers, For job seekers The level values ​​of the feature difference tensor. For the recruiter The required value of the feature difference tensor. For the first Industry weight coefficients of the feature difference tensor;

[0163] An economic environment adjustment factor is introduced to correct the basic matching degree, and the final matching degree score of the potential path is obtained.

[0164] Specifically, key supply and demand entity nodes are selected from the dynamic knowledge graph. The criteria for selection are the influence of entity nodes in the supply and demand relationship. Job nodes are sorted by the number of job openings in the past 30 days, and the top 20% of job nodes are selected. Enterprise nodes are sorted by the total number of job openings in the past 30 days, and the top 20% of enterprise nodes are selected. Skill nodes are sorted by the total number of times they are mentioned in job requirements in the past 30 days, and the top 20% of skill nodes are selected. Job seeker-related nodes (associated with skill nodes) are sorted by the number of valid resume submissions in the past 30 days, and the top 20% of related nodes are selected. These selected nodes together constitute the key supply and demand entity nodes.

[0165] Furthermore, a multi-dimensional assessment channel is established between key supply and demand entities. The dimensions include skill matching, salary matching, geographical distance matching, and job urgency. Skill matching assesses the overlap between skill requirements and skill supply between nodes. Salary matching assesses the degree of overlap between the salary range of the job and the job seeker's expected salary. Geographical distance matching assesses the straight-line distance between the location of the job and the job seeker's desired location. Job urgency is determined based on the number of "urgently hiring" tags marked by the company. Each dimension corresponds to an assessment sub-channel. All sub-channels between the same pair of nodes are integrated to form a multi-dimensional assessment channel connecting the pair of nodes.

[0166] Furthermore, the comprehensive matching score of potential paths is calculated using a path value assessment function. First, a scoring range of 0-100 is set for each dimension. Skill matching score is converted according to the overlap ratio (e.g., 80 points for 80% overlap), salary matching score is converted according to the salary overlap ratio (e.g., 60 points for 60% overlap), geographical distance matching score is converted according to distance inverse (e.g., 100 points for within 10 kilometers, 20 points deducted for every additional 10 kilometers), and job demand urgency score is converted according to the number of "urgently hiring" tags (e.g., 100 points for 3 tags, 30 points deducted for each less than 1). Then, weights are assigned to each dimension (skills 40%, salary 30%, geography 15%, urgency 15%). The scores of each dimension are multiplied by their corresponding weights and then summed to obtain the comprehensive matching score of the potential path.

[0167] Furthermore, conflicting path branches are excluded based on the comprehensive matching score. A conflicting path branch refers to another path whose score is no lower than that of the branch in all evaluation dimensions, and whose score is higher than that of the branch in at least one dimension. For example, if path A has a skill score of 80 and a salary score of 70, and path B has a skill score of 85 and a salary score of 75, then path A is a conflicting path branch. All such branches are removed from the potential paths, and the remaining paths are all non-dominated paths, that is, there is no other path that is better than it in all dimensions.

[0168] Furthermore, the non-dominated paths are sorted by Pareto front, arranged from high to low according to the comprehensive matching score, and those with the same score are sorted by skill matching score, and those with the same score are sorted by salary matching score. The top 10% of the non-dominated paths are selected. These paths together constitute the optimal matching path set in the labor market. Each path contains the key supply and demand entity nodes connected and the evaluation scores of each dimension.

[0169] Specifically, feature vectors need to be extracted from job nodes and job seeker-related nodes within key supply and demand entity nodes. Skill feature vectors are extracted using all skills in the labor force knowledge base as dimensions. In the skill feature vector of a job node, a value of 1 for each dimension indicates that the skill is required for the position, and 0 indicates that it is not required. In the skill feature vector of a job seeker-related node, a value of 1 for each dimension indicates that the job seeker possesses the skill, and 0 indicates that they do not. Demand feature vectors are extracted using the number of openings, educational requirements, and work experience requirements as dimensions. The number of openings dimension is a specific number; the educational requirements dimension is assigned values ​​as follows: high school 1, bachelor's degree 2, master's degree 3, doctorate 4; and the work experience requirements dimension is assigned values ​​as follows: 0-1 years 1, 2-3 years 2, 4-5 years 3, 6 years or more 4. Demand feature vectors are generated for both job nodes and job seeker-related nodes according to these rules. Calculate the difference in the corresponding dimension of the skill feature vector of the same pair of nodes (job value minus job seeker value, the result is 0 or 1), and the difference in the corresponding dimension of the demand feature vector (take the absolute value of job value and job seeker value). Arrange these differences in the order of skill dimension first and demand dimension last. The resulting multidimensional data structure is the feature difference tensor of key supply and demand entity nodes.

[0170] Furthermore, the basic matching degree of the potential path is calculated based on the feature difference tensor. A scoring transformation is applied to the difference values ​​of each dimension in the feature difference tensor: a skill dimension difference of 0 receives 100 points, and a difference of 1 receives 0 points; a difference in the number of recruits of 0 receives 100 points, a difference of 1 receives 80 points, a difference of 2 receives 60 points, and a difference of 3 or more receives 0 points; a difference in education requirements of 0 receives 100 points, a difference of 1 receives 50 points, and a difference of 2 or more receives 0 points; the scoring rules for work experience requirement differences are the same as for education requirements. A 50% weight is assigned to the skill dimension, and the weights for the number of recruits, education requirements, and work experience requirements are assigned 20%, 20%, and 10% respectively. The scores of each dimension are multiplied by their corresponding weights and then summed to obtain the basic matching degree of the potential path.

[0171] Furthermore, the economic environment adjustment factor is determined based on industry prosperity and regional economic growth. Industry prosperity is judged by the number of newly established enterprises in the industry over the past three months: 1.1 is used when the year-on-year growth rate of newly established enterprises exceeds 10%, 1.0 is used when the growth rate is 0-10%, and 0.9 is used when the growth rate is negative. Regional economic growth is based on the local quarterly GDP year-on-year growth rate: 1.1 is used when the growth rate is above 5%, 1.0 is used when the growth rate is between 3% and 5%, and 0.9 is used when the growth rate is below 3%. The arithmetic mean of the industry prosperity adjustment factor and the regional economic growth adjustment factor is taken as the final economic environment adjustment factor. The basic matching degree of the potential path is multiplied by this adjustment factor, and the result is the final matching degree score of the potential path.

[0172] Specifically, The score, which serves as the basic matching score, is derived from the calculation result of a formula and ranges from 0 to 1. The higher the score, the higher the matching degree between the job seeker and the recruiter; the lower the score, the lower the matching degree.

[0173] Furthermore, The total number of skill dimensions is derived from the total number of skill categories in the workforce knowledge base. For example, if the knowledge base contains 100 skills such as "programming languages," "communication skills," and "data analysis," then... The value is 100, covering all skill types involved in the job search and recruitment process.

[0174] Furthermore, As an ordinal number for the skill dimension, its source is... Each skill dimension is numbered in a fixed order, starting from 1, and corresponding sequentially to each skill category in the knowledge base. Each skill dimension has a unique ordinal number; for example, "programming language" corresponds to... =1, corresponding to "communication skills" =2, and so on.

[0175] Furthermore, As the first The industry weight coefficient for each skill is determined based on the industry's demand for that skill, calculated by analyzing the number of job postings in that industry over the past six months. The ratio of the number of times a skill is mentioned to the total number of times all skills in the industry are mentioned is called the percentage. For example, if "programming language" is mentioned 30% of the total number of times in the internet industry, then the corresponding... It is 0.3.

[0176] Furthermore, As the job seeker The level values ​​of the feature difference tensor are derived from the job seeker's position in the feature difference tensor. The actual level of a skill is determined by how it is marked on the job seeker's resume; if it is marked as "proficient," then... =3, "Master" means =2, "understand" then =1, "Not mentioned" means =0.

[0177] Furthermore, As the recruiter The required value of the feature difference tensor is derived from the employer's input in the feature difference tensor. The required skill level is determined by the skill description in the job posting; if "proficient" is required, then... =3, "Master" means =2, "understand" then =1, "No requirements" =0.

[0178] Furthermore, the meaning of the formula is to calculate the skill level difference between job seekers and recruiters in each skill dimension, multiply these differences by the industry weight coefficient of the corresponding feature difference tensor, sum them, and then divide by the sum of all skill weight coefficients to obtain the weighted ratio of skill differences. Subtracting this ratio from 1 gives the basic matching degree, which intuitively reflects the overall matching degree between the job seeker's skill level and the recruiter's skill requirements.

[0179] Furthermore, the trend of the formula is that when the job seeker... With the recruiter The closer the values ​​are on a certain skill dimension, the greater the absolute difference between the two. The smaller the value, the smaller its contribution to the total molecule, and the lower the basic matching degree. The closer the value is to 1, the greater the numerical difference between the two values, and the larger the absolute difference, the greater the contribution to the total numerator. The closer it is to 0. At the same time, the first... Industry weight coefficients of feature difference tensors The greater the skill dimension, the more... Size pair The more significant the impact, the more important it is to determine the overall matching degree of high-weight skills, meaning that the matching degree of high-weight skills plays a major role in the overall matching degree, while the impact of low-weight skills is relatively small.

[0180] In summary, by identifying key supply and demand entity nodes in a dynamic knowledge graph and establishing multi-dimensional evaluation channels between nodes, the degree of supply and demand matching can be comprehensively considered from multiple dimensions such as skills, needs, and regions. The basic matching degree is calculated based on the feature difference tensor, and the score is corrected by introducing an economic environment adjustment factor, making the matching evaluation more in line with actual market conditions, avoiding the limitations of single-dimensional evaluation, and ensuring the accuracy of the matching results.

[0181] In summary, by calculating the comprehensive matching score of potential paths using a path value assessment function, eliminating conflicting path branches, and ranking non-dominated paths using the Pareto front, the optimal set of matching paths can be selected. This process considers both the direct matching needs of supply and demand entities and the rational allocation of overall market resources, reducing resource mismatch and providing an efficient and scientific path basis for the subsequent formulation of labor resource allocation plans, thereby improving the overall efficiency of labor market resource allocation.

[0182] The labor resource allocation module 106 is used to perform desensitization processing on the optimal matching path set to obtain the labor resource allocation scheme of the labor market.

[0183] In this embodiment of the invention, when the labor resource allocation module performs desensitization processing on the optimal matching path set to obtain the labor resource allocation scheme of the labor market, it is specifically used for:

[0184] Identify sensitive fields in the optimal matching path set, wherein the sensitive fields include personal identification identifiers, corporate trade secrets, and precise salary values;

[0185] The personal identification identifier is hashed to obtain the anonymous identifier code of the optimal matching path set;

[0186] Replace the enterprise trade secret item with the industry classification code;

[0187] The exact salary value is converted into an interval to obtain the salary level label of the optimal matching path set;

[0188] By recombining the anonymous identifier, the industry classification code, and the salary grade label, a labor resource allocation scheme for the labor market is obtained.

[0189] Specifically, all field information is extracted from the optimal matching path set, and the content attributes of each field are checked one by one. Personal identification identifiers include job seekers' ID card number, mobile phone number, home address, bank account number and other information that can be directly associated with the individual; corporate trade secrets include the company's core technical parameters, undisclosed project names, customer lists, internal cost structures and other information that is only known to the company internally; salary precision values ​​include monthly salary, annual salary and bonus amount down to the single digit, such as "12,568 yuan / month". In this way, all sensitive fields are clearly identified.

[0190] Furthermore, the identified personal identifiers are hashed. Specifically, the text content of the personal identifier is converted into a string of characters of fixed length. During the conversion, specific calculation rules are used to make it impossible to deduce the original identifier from the result. For example, "410102199001" is converted into "a7b3c9d2e5f8g1h4j". Each different personal identifier corresponds to a unique string, and these strings are the anonymous identifiers of the optimal matching path set.

[0191] Furthermore, the replacement of enterprise trade secret items is based on the industry classification standards issued by the state. These standards divide all industries into different categories and assign them unique codes. For example, "software development" corresponds to code "6510" and "automobile manufacturing" corresponds to code "3611". Based on the business content involved in the enterprise trade secret items, the corresponding industry classification code is found in the standard, and the original text content of the trade secret items is directly replaced with this code. For example, "XX Company's autonomous driving core algorithm project" is replaced with "3650".

[0192] Furthermore, when performing interval conversion on the precise salary values, multiple salary intervals are pre-defined and each interval is assigned a unique label. The interval division rules are as follows: 5,000 yuan and below is L1, 5,001-8,000 yuan is L2, 8,001-12,000 yuan is L3, 12,001-20,000 yuan is L4, and 20,001 yuan and above is L5. Each precise salary value is compared with these intervals to determine its interval, and then the original precise value is replaced with the corresponding label. For example, "12,568 yuan / month" is converted to "L4". These labels are the salary level labels of the optimal matching path set.

[0193] Furthermore, when reorganizing the anonymous identifier, industry classification code, and salary grade label, the anonymous identifier, the replaced industry classification code, and the converted salary grade label corresponding to each path are arranged sequentially according to the original path order in the optimal matching path set. At the same time, non-sensitive fields in the path, such as job name, skill requirements, and regional information, are retained to form structured data containing desensitized sensitive information and complete non-sensitive information. This data is the labor resource allocation plan for the labor market.

[0194] In summary, by identifying sensitive fields such as personal identification identifiers, corporate trade secrets, and precise salary values ​​in the optimal matching path set, the personal identification identifiers are hashed to generate anonymous identifiers, corporate trade secrets are replaced with industry classification codes, and precise salary values ​​are range-based to obtain salary grade labels. This effectively avoids the leakage of personal privacy and the dissemination of corporate business information, and strengthens the information security defense while realizing resource allocation.

[0195] In summary, the de-identified labor resource allocation scheme removes the specific details of sensitive information, retaining only the key elements used for resource allocation (such as anonymous identifiers, industry classifications, and salary grades). This not only meets the actual needs of labor market supply and demand matching and resource allocation, but also complies with relevant data security and privacy protection regulations. This makes the allocation scheme legal, compliant, and operable in practical applications, promoting the standardized and orderly development of labor resource allocation work.

[0196] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0197] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0198] Finally, 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.

Claims

1. A real-time big data driven labor market supply and demand forecasting and intelligent matching system, characterized in that, The system comprises a feature separation module, a space-time feature matrix acquisition module, a supply-demand gap prediction module, a dynamic knowledge graph generation module, an optimal matching path acquisition module and a labor resource allocation module, wherein: The feature separation module is configured to extract dynamic features of original data streams in a labor market, and separate parallelized dynamic features to obtain a standardized feature vector set of the labor market. The space-time feature matrix acquisition module is configured to perform space-time alignment on the standardized feature vector set based on a sliding time window to obtain a space-time feature matrix of the labor market. The supply-demand gap prediction module is configured to capture time dependence of the space-time feature matrix, and aggregate spatial neighborhood information of the time dependence to obtain regionalized supply-demand gap prediction values of the labor market, including: extracting feature blocks of a target region and its first-order neighborhood region in the space-time feature matrix to obtain a neighborhood feature set of the labor market; performing feature transformation on the feature vectors in the neighborhood feature set, wherein the calculation formula of the feature transformation is as follows: ； wherein, is a region feature vector after feature transformation, is a nonlinear activation function, is a region ordinal, is a region neighbor region set of is a region spatial weight coefficient between and is a region time-dependent feature vector of is a region spatial feature vector, is a bias term, is an element-wise multiplication operator; performing regression dimension reduction on the feature vectors after the feature transformation to obtain the regionalized supply-demand gap prediction values of the labor market; The dynamic knowledge graph generation module is configured to construct a basic graph skeleton of the labor market based on the regionalized supply-demand gap prediction values and entity relationship triples in a labor field knowledge base, and map dynamic attributes of entities in the original data streams to the basic graph skeleton to obtain a dynamic knowledge graph of the labor market, including: extracting enterprise-post-skill three-level entity relationship chains from the labor field knowledge base to obtain a core triple set of the labor market; binding the regionalized supply-demand gap prediction values to corresponding geographical region entity nodes in the labor market as supply-demand attribute labels of the geographical region entity nodes; establishing labor flow edges between adjacent geographical region entity nodes according to economic correlation strength; fusing the core triple set, the supply-demand attribute labels and the labor flow edges to construct the basic graph skeleton of the labor market; The optimal matching path acquisition module is configured to generate an optimal matching path set of the labor market according to multi-dimensional matching degree scores between supply-demand entities in the dynamic knowledge graph; The labor resource allocation module is configured to perform desensitization processing on the optimal matching path set to obtain a labor resource allocation scheme of the labor market.

2. The real-time big data driven labor market demand and supply forecasting and intelligent matching system of claim 1, wherein, When the feature separation module performs extraction of dynamic features of original data streams in a labor market, and separation of parallelized dynamic features to obtain a standardized feature vector set of the labor market, it is specifically configured to: separate recruitment post data, job seeker behavior data and salary trend data from the original data streams of the labor market; statistically analyze the recruitment post data in a sliding window to obtain a job demand fluctuation feature vector of the labor market; perform semantic analysis on the job seeker behavior data to obtain a job intention strength feature vector of the labor market; Separate the long-term trend component and the short-term fluctuation component of the salary trend data to obtain a salary dynamic feature vector of the labor market; Perform standardization scaling processing on the job demand fluctuation feature vector, the job-seeking intention strength feature vector and the salary dynamic feature vector to obtain a standardized feature vector set of the labor market.

3. The real-time big data driven labor market demand and supply forecasting and intelligent matching system of claim 1, wherein, When the spatio-temporal feature matrix acquisition module performs spatio-temporal alignment on the standardized feature vector set based on a sliding time window to obtain a spatio-temporal feature matrix of the labor market, it is specifically used for: Identifying the time stamp mark and the geographical position mark in the standardized feature vector set to obtain a spatio-temporal coordinate index of the standardized feature vector set; Mapping the spatio-temporal coordinate index to a predefined spatial grid according to the regional division rule of the labor market; Within the sliding time window, performing eigenvalue weighted fusion on the feature vectors of the spatial grid to obtain a grid feature block of the spatial grid; Performing boundary smoothing processing on the grid feature block according to the spatial topological relationship; Arranging the grid feature block after the boundary smoothing processing in the order of the time window to obtain the spatio-temporal feature matrix of the labor market.

4. The real-time big data driven labor market demand and supply forecasting and intelligent matching system of claim 3, wherein, When the supply-demand gap prediction module captures the time dependence of the spatio-temporal feature matrix and aggregates the spatial neighborhood information of the time dependence, it is specifically used for: Performing bidirectional recursive feature scanning on the spatio-temporal feature matrix along the time dimension to obtain a time context dependent feature of the spatio-temporal feature matrix; Constructing a spatial topology graph of the labor market by taking the labor market area under the regional division rule as a node and taking the economic correlation strength of the labor market area as an edge; On the spatial topology graph, iteratively fusing the time context dependent feature of the core area with the neighborhood area features in the spatial topology graph to obtain a spatio-temporal fusion feature tensor of the labor market; Inputting the spatio-temporal fusion feature tensor into a fully connected decision layer to obtain a regionalized supply-demand gap prediction value of the labor market.

5. The real-time big data driven labor market demand and supply forecasting and intelligent matching system of claim 1, wherein, When the dynamic knowledge graph generation module maps the entity dynamic attributes in the original data stream to the basic graph skeleton to obtain a dynamic knowledge graph of the labor market, it is specifically used for: Real-time capturing the job demand change rate, the job seeker skill update mark and the salary fluctuation value in the original data stream to obtain a dynamic attribute vector of the labor market; Associating the dynamic attribute vector to the corresponding entity node in the basic graph skeleton; Performing weighted fusion on the current attribute value and the corresponding historical attribute value of the dynamic attribute vector, and injecting the fused attribute value into the entity node to obtain the dynamic knowledge graph of the labor market.

6. The real-time big data driven labor market demand and supply forecasting and intelligent matching system as claimed in claim 1, wherein, When the optimal matching path acquisition module generates an optimal matching path set of the labor market according to the multi-dimensional matching degree score between the supply-demand entities in the dynamic knowledge graph, it is specifically used for: Identifying the key supply-demand entity nodes in the dynamic knowledge graph; Establishing a multi-dimensional evaluation channel between the key supply-demand entity nodes; Calculating the comprehensive matching score of the potential path in the multi-dimensional evaluation channel through a path value evaluation function; According to the comprehensive matching score, a conflict path branch in the potential path is excluded, and a non-dominated path of the labor market is obtained; The non-dominated path is sorted by a Pareto frontier, and a set of optimal matching paths of the labor market is obtained.

7. The real-time big data driven labor market demand and supply forecasting and intelligent matching system of claim 6, wherein, When the optimal matching path acquisition module performs the calculation of the comprehensive matching score of the potential path in the multi-dimensional evaluation channel by the path value evaluation function, it is specifically used for: extracting the skill feature vector and the demand feature vector of the key supply-demand entity node to obtain a feature difference tensor of the key supply-demand entity node; Based on the feature difference tensor, the basic matching degree of the potential path in the multi-dimensional evaluation channel is calculated, and the calculation formula of the basic matching degree is as follows: ； wherein, is the score of the base match degree, is the total number of skill dimensions, is the ordinal number of skill dimensions, is the industry weight coefficient of the th skill, is the level value of the th feature difference tensor of the job seeker, is the requirement value of the th feature difference tensor of the recruiter, is the industry weight coefficient of the th feature difference tensor. An economic environment adjustment factor is introduced to modify the basic matching degree, and a final matching degree score of the potential path is obtained.

8. The real-time big data driven labor market demand and supply forecasting and intelligent matching system as claimed in claim 1, wherein, When the labor resource allocation module performs the desensitization processing on the set of optimal matching paths, a labor resource allocation scheme of the labor market is obtained, and it is specifically used for: Identify the sensitive fields in the set of optimal matching paths, wherein the sensitive fields include personal identity identifiers, enterprise business confidential items, and accurate salary values; Hash transformation is performed on the personal identity identifier to obtain an anonymous identification code of the set of optimal matching paths; The enterprise business confidential item is replaced by an industry classification code; The accurate salary value is converted into an interval to obtain a salary level label of the set of optimal matching paths; The anonymous identification code, the industry classification code, and the salary level label are recombined to obtain a labor resource allocation scheme of the labor market.

Images