A GIS-based application-oriented university talent training quality tracking service system

Abstract

This invention relates to the field of educational evaluation technology and discloses a GIS-based application-oriented university talent training quality tracking service system, comprising: a data acquisition and positioning module for collecting and standardizing information on students, alumni, internships, and employers to generate standardized employment indicator data and spatial element data; a quality scoring module for weighting employment indicators based on preset weights to generate a comprehensive quality score; a spatial analysis module for constructing initial classification units, performing adaptive grid aggregation and kernel density estimation, and extracting employment quality hotspot clusters; a trend monitoring module for summarizing comprehensive quality scores by cohort and calculating year-on-year change rates to generate trend signals; and a gap diagnosis module for comparing the current cohort's average employment indicator with a target threshold, generating an improvement trigger signal and providing feedback on weight correction when discrepancies are found. This invention achieves standardized tracking, spatial cluster identification, and dynamic feedback optimization of talent training quality.

Description

Technical Field

[0001] This invention belongs to the field of educational evaluation technology, specifically relating to a GIS-based application-oriented higher education talent training quality tracking service system. Background Technology

[0002] The current university employment quality tracking system has some technical deficiencies, mainly reflected in the following aspects:

[0003] In existing technologies, employment data collection largely relies on manual entry or aggregation on a single platform, resulting in problems such as inconsistent data formats, ambiguous address information (e.g., only recording the name of the administrative division without precise spatial coordinates), and fragmented indicator dimensions. This makes it difficult to standardize the data and form a unified employment data asset across the entire region.

[0004] At the spatial analysis level, traditional methods can only achieve statistical analysis of employment data at the administrative division level, making it difficult to reveal the distribution patterns of employment quality in micro-geographical space. Existing spatial clustering algorithms often lack the ability to adaptively aggregate multi-scale grids, resulting in insufficient accuracy in identifying hotspot areas. Summary of the Invention

[0005] This invention provides a GIS-based application-oriented university talent training quality tracking service system, which solves the technical problems in related technologies such as scattered employment data structure, lack of spatial analysis in quality evaluation, difficulty in quantifying dynamic trends, and lack of feedback and adjustment mechanisms.

[0006] This invention provides a GIS-based application-oriented higher education talent training quality tracking service system, comprising:

[0007] The data acquisition and positioning module is used to collect student information, alumni information, internship logs and employer information and represent them as employment data. It performs standardized processing on the employment data, converts the included address fields into latitude and longitude coordinates, and generates standardized employment indicator data and spatial element data.

[0008] The quality scoring module is used to weight the employment indicators in the standardized employment indicator data according to preset weights to generate a comprehensive quality score.

[0009] The spatial analysis module generates initial classification units based on spatial element data, categorized by industry and administrative division. It then executes a quadtree-index-driven adaptive grid aggregation algorithm on these initial classification units to generate multi-scale spatial grid units. Based on the comprehensive quality score, it uses a Gaussian kernel function for nonparametric density estimation, outputting an employment quality kernel density point cloud dataset. From this kernel density point cloud dataset, it extracts statistically significant employment quality hotspot clusters. These discrete hotspot clusters are then converted into a continuous 3D point cloud surface model. An octree spatial index is used to adaptively resample the 3D point cloud surface model to generate uniformly distributed employment quality point cloud data. Finally, it estimates the point cloud normal vectors and constructs an enhanced point cloud dataset containing spatial coordinates, density values, and normal vectors, along with their attribute association information.

[0010] The trend monitoring module is used to summarize the comprehensive quality scores by session, calculate the year-on-year change rate of the average comprehensive quality scores of adjacent sessions, and generate trend signals.

[0011] The gap diagnosis module is used to calculate the average value of the current employment indicators and compare it with the corresponding preset target threshold to obtain the difference. When the difference of any employment indicator exceeds the preset tolerance, an improvement trigger signal is generated and the improvement trigger signal containing the weight adjustment is fed back to the quality scoring module.

[0012] Furthermore, the employment indicators include: on-time employment rate, salary index, job matching degree, employer satisfaction, graduate satisfaction, and skills utilization.

[0013] Furthermore, the standardization process includes:

[0014] S201, Perform missing value imputation on the collected employment data;

[0015] S202, perform outlier removal on the filled employment data, and exclude outlier data that exceeds the preset threshold range;

[0016] S203, based on the employment data of the current graduates after removing outliers, calculate the sample mean and sample standard deviation of each employment indicator;

[0017] S204, based on the sample mean and sample standard deviation, normalizes the employment data after outlier removal to zero mean and unit variance to obtain standardized employment indicator data.

[0018] Furthermore, based on spatial element data, initial classification units are generated according to industry category and administrative division; a quadtree index-driven adaptive grid aggregation algorithm is executed on the initial classification units to generate multi-scale spatial grid units, including:

[0019] S301, Based on industry category and administrative division, a quadtree spatial index is established for spatial element data to generate initial spatial partitions;

[0020] S302, perform a depth-first traversal of the quadtree based on the feature distribution density within the initial spatial partition:

[0021] When the number of features in the grid exceeds the preset density threshold, the quadrature subgrid generation is recursively executed.

[0022] When the number of features within a grid is below the minimum density threshold, an adjacent grid merging operation is performed;

[0023] S303 inherits the original spatial element attributes from the final generated grid cells, aggregates employment indicator statistics by industry-administrative region dimension, and outputs a multi-scale spatial grid cell set with attribute labels.

[0024] Furthermore, based on the comprehensive quality score, a Gaussian kernel function is used for nonparametric density estimation, outputting an employment quality kernel density point cloud dataset; statistically significant employment quality hotspot point cloud clusters are extracted from the kernel density point cloud dataset; the discrete hotspot point cloud clusters are converted into a continuous three-dimensional point cloud surface model, including:

[0025] The overall quality score is mapped to the center point of the corresponding grid cell, and radial density diffusion calculation is performed using the Gaussian kernel function as the basic unit to generate a discrete kernel density point cloud containing latitude and longitude coordinates and density values.

[0026] Perform local spatial autocorrelation analysis on the kernel density point cloud:

[0027] Based on Getis-Ord Gi The statistical measure is used to calculate the hotspot significance Z-value for each point; points with Z-values ​​exceeding the confidence threshold are selected to form a significant hotspot cloud cluster.

[0028] Using salient hotspot clusters as control points:

[0029] The spatial variogram model is fitted using the Kriging algorithm; a continuous three-dimensional point cloud surface model is generated based on semivariance final interpolation, and its elevation dimension represents the employment quality density value.

[0030] Furthermore, an octree spatial index is used to perform density-adaptive resampling of the 3D point cloud surface model to generate uniformly distributed employment quality point cloud data; point cloud normal vectors are estimated, and an enhanced point cloud dataset containing spatial coordinates, density values, and normal vectors, along with its attribute association information, is constructed, including:

[0031] Import the 3D point cloud surface model into the octree spatial index, initialize the cube voxels according to the preset resolution threshold, and retain the representative point with the largest density value for each non-empty voxel.

[0032] The uniformity of point cloud distribution within a voxel is recursively detected. When the variance of the point spacing exceeds the tolerance, the voxel is subdivided into eight equal parts. The process is iterated until the uniformity threshold is met, and then a density-balanced employment quality point cloud is output.

[0033] Perform local neighborhood analysis on the resampled point cloud to extract the K nearest neighbor set of each point and construct the covariance matrix;

[0034] The covariance matrix eigenvectors are obtained by principal component analysis, and the vector corresponding to the smallest eigenvalue is used as the point cloud normal vector.

[0035] An enhanced point cloud dataset is generated by binding spatial coordinates, density values, and normal vectors.

[0036] Furthermore, the overall quality scores are aggregated by session, and the year-on-year change rate of the average overall quality score between adjacent sessions is calculated to generate trend signals, including:

[0037] S401, aggregate the overall quality scores by session and calculate the average overall quality score for that session;

[0038] S402. The year-on-year change rate is obtained by subtracting the average comprehensive quality score of the previous session from the average comprehensive quality score of the current session and dividing by the average comprehensive quality score of the previous session.

[0039] S403 compares the year-on-year change rate with a preset positive threshold and a preset negative threshold respectively. When the year-on-year change rate is not less than the preset positive threshold, an upward trend signal is generated. When the year-on-year change rate is not greater than the preset negative threshold, a downward trend signal is generated. When it is between the preset positive threshold and the preset negative threshold, a stable trend signal is generated.

[0040] Furthermore, all employment indicators with differences exceeding the preset tolerance and their corresponding differences are written into the improvement trigger signal; the ratio of the difference to the largest difference among the employment indicators is multiplied by the preset learning rate to obtain the weight correction amount, which is then written into the improvement trigger signal.

[0041] Furthermore, it also includes a route optimization module, which uses road network data and real-time speeds of various road segments to calculate the travel time of multiple routes from the school to the internship unit based on the geographical location of the school and the internship unit, and selects the shortest route to generate an internship inspection plan.

[0042] The beneficial effects of this invention are as follows:

[0043] The data acquisition and positioning module transforms scattered heterogeneous data such as student information and employer data into standardized employment indicator data and spatial element data in a unified format through address-latitude and longitude coordinate conversion and standardization processing. This solves the problem of data silos caused by the chaotic format and lack of spatial information in traditional employment data.

[0044] The spatial analysis module adopts an adaptive grid aggregation algorithm driven by a quadtree index, which breaks through the limitations of traditional fixed-scale spatial analysis and can automatically generate multi-scale grid cells according to data density. Combining Gaussian kernel function density estimation and 3D point cloud modeling technology, discrete employment data is transformed into a continuous spatial distribution model. Through octree indexing, efficient resampling and feature extraction of point cloud data are achieved, and the spatial distribution patterns of employment quality hotspots are accurately identified. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of a module of an application-oriented higher education talent training quality tracking service system based on GIS, according to the present invention. Detailed Implementation

[0046] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0047] like Figure 1 As shown, an application-oriented higher education talent training quality tracking service system based on GIS includes:

[0048] The data acquisition and positioning module 101 is used to collect student information, alumni information, internship logs and employer information and represent them as employment data. It performs standardized processing on the employment data, converts the included address fields into latitude and longitude coordinates, and generates standardized employment indicator data and spatial element data.

[0049] The quality scoring module 102 is used to perform weighted processing on the employment indicators in the standardized employment indicator data according to preset weights to generate a comprehensive quality score.

[0050] The spatial analysis module 103 is used to generate initial classification units based on spatial element data, categorized by industry and administrative division; execute a quadtree index-driven adaptive grid aggregation algorithm on the initial classification units to generate multi-scale spatial grid units; perform nonparametric density estimation using a Gaussian kernel function based on the comprehensive quality score, and output an employment quality kernel density point cloud dataset; extract statistically significant employment quality hotspot point cloud clusters from the kernel density point cloud dataset; convert discrete hotspot point cloud clusters into a continuous 3D point cloud surface model; perform density adaptive resampling on the 3D point cloud surface model using an octree spatial index to generate uniformly distributed employment quality point cloud data; estimate point cloud normal vectors, and construct an enhanced point cloud dataset containing spatial coordinates, density values, and normal vectors, along with their attribute association information;

[0051] The trend monitoring module 104 is used to summarize the comprehensive quality scores by session, calculate the year-on-year change rate of the average comprehensive quality scores of adjacent sessions, and generate trend signals.

[0052] The gap diagnosis module 105 is used to calculate the average value of the current employment indicators and compare it with the corresponding preset target threshold to obtain the difference. When the difference of any employment indicator exceeds the preset tolerance, an improvement trigger signal is generated and the improvement trigger signal containing the weight adjustment amount is fed back to the quality scoring module.

[0053] In one embodiment of the present invention, the information collected for student information includes, but is not limited to, name, student ID, major, graduation date, academic performance, skill certificates obtained, competitions participated in and awards received, etc.; the information collected for alumni information includes, but is not limited to, name, graduation year, current employer, job title, start date, salary, career path, etc.; the information collected for internship logs includes, internship company name, internship position, internship start and end dates, internship assessment results, etc.; and the information collected for employers includes, full company name, industry category, administrative division code, detailed address, contact person and contact information, job requirements, satisfaction rating, etc. All the information collected is aggregated through structured form input and system interface integration, and uniformly represented as structured employment data.

[0054] In one embodiment of the present invention, the employment indicators include: timely employment rate, salary index, job matching degree, employer satisfaction, graduate satisfaction, and skills utilization; wherein, the timely employment rate is obtained by the ratio of the number of graduates who sign contracts within 3 months of graduation to the total number of current graduates; the salary index is obtained by the ratio of salary to the industry average salary; the job matching degree is obtained by comparing the students' professional direction and curriculum with the employer's job requirements and skill requirements, and calculating using cosine similarity; employer satisfaction is represented by satisfaction scores; and skills utilization is obtained by the ratio of the number of skills actually applied to the total number of skills learned in school.

[0055] In one embodiment of the present invention, the standardization process includes:

[0056] S201. Missing value imputation is performed on the collected employment data. Specifically, for numerical missing data, such as salary, the K-nearest neighbor imputation method is used based on the statistical characteristics of groups of the same major, the same year, or the same industry. The mean of the K samples most similar to the missing sample is selected for imputation. For categorical missing data, such as industry category, the mode imputation method is used to imput the mainstream industry category of employers in the same region, ensuring that the imputed data conforms to the overall distribution characteristics of the sample.

[0057] S202, outlier removal is performed on the filled employment data to exclude abnormal data that exceed the preset threshold range; specifically, based on industry benchmarks and historical data distribution, a reasonable threshold range is preset for each employment indicator, and extreme values ​​that exceed this range are identified and removed using the 3σ principle.

[0058] S203, based on the employment data of the current graduates after removing outliers, calculate the sample mean and sample standard deviation of each employment indicator;

[0059] S204, based on the sample mean and sample standard deviation, performs zero-mean, unit-variance normalization on the employment data after outlier removal, that is, uniformly maps it to the standard normal distribution space with a mean of 0 and a standard deviation of 1, to obtain standardized employment indicator data.

[0060] For address fields in employment data, such as the detailed address of the employer and the current residence address of the alumni, a third-party map interface is called to perform geocoding conversion, parsing the text address into the corresponding latitude and longitude coordinates, and associating the latitude and longitude coordinates with the corresponding employment data and storing them in the spatial database to form spatial element data containing spatial attributes.

[0061] In one embodiment of the present invention, the six employment indicators in the standardized employment indicator data are weighted and summed according to preset weights to generate a comprehensive quality score; wherein, the weight of each employment indicator is determined based on the consensus of education industry experts and historical data analysis.

[0062] In one embodiment of the present invention, the comprehensive quality score and spatial element data are classified based on industry category and administrative division. The administrative division is obtained by calling the administrative division geocoding interface based on the latitude and longitude coordinates contained in the spatial element data.

[0063] In one embodiment of the present invention, initial classification units are generated based on spatial element data, according to industry categories and administrative divisions; an adaptive grid aggregation algorithm driven by a quadtree index is executed on the initial classification units to generate multi-scale spatial grid units, including:

[0064] S301. Based on industry category and administrative division, a quadtree spatial index is established for spatial element data to generate initial spatial partitions. Specifically, firstly, the industry classification standards and administrative division levels are determined, and the spatial element data is classified according to both industry category and administrative division to obtain preliminary grouped data. Then, the smallest bounding rectangle covering all spatial elements is used as the root node of the quadtree. Based on the above classification results, spatial elements belonging to the same industry and located within the same administrative division are assigned to the initial sub-regions under the root node, and the spatial boundaries of each initial sub-region are determined to form the first-level nodes of the quadtree. Next, the spatial element IDs and basic attributes contained in each initial sub-region are recorded to complete the initialization of the quadtree spatial index and generate the initial spatial partitions.

[0065] This step, through the dual dimensions of industry and administrative division, ensures that the initial spatial partitioning retains the relevance of industry attributes while possessing clear spatial boundaries, providing a structured foundation for subsequent grid aggregation.

[0066] S302, based on the feature distribution density within the initial spatial partitions, perform a quadtree depth-first traversal; specifically, first count the number of spatial features within each initial spatial partition, and calculate the feature distribution density (i.e., the number of features per unit area) by combining the partition area. Then, start the quadtree depth-first traversal, visiting each initial spatial partition sequentially starting from the root node;

[0067] When the number of elements in a grid exceeds a preset density threshold, the current grid is divided into four sub-grids along the center point. Each sub-grid inherits the industry and administrative division attributes of the parent grid. The number of elements in each sub-grid is recounted. This division process is repeated until the number of elements in the sub-grids is lower than the density threshold.

[0068] When the number of elements in a grid is lower than the minimum density threshold, the adjacent grids of the same industry and administrative divisions at the same level are retrieved. The total number of elements after merging is calculated to see if it reaches a reasonable density range. If it does, the adjacent grids are merged into a new grid, and the spatial boundary and element attribute set of the new grid are updated.

[0069] This step avoids the problems of over-segmentation of sparse data and under-segmentation of dense data by fixed grids through density-driven dynamic partitioning and merging, enabling the grid cell size to adaptively match the feature distribution characteristics, thereby improving the expression accuracy and computational efficiency of spatial data.

[0070] S303 inherits the attributes of the original spatial elements from the final generated grid cells, aggregates employment indicator statistics by industry-administrative region dimension, and outputs a multi-scale spatial grid cell set with attribute labels. Specifically, it associates the attribute information of all original spatial elements involved in the subdivision or merging process with each final grid cell, such as graduate education level, employment salary, and employer rating. Then, it statistically aggregates the employment indicators within each grid cell by industry category and administrative region dimension, calculates the average, median, and maximum values ​​of the indicators, and generates the employment quality characteristic value of that grid cell. Finally, it adds attribute labels containing industry code, administrative region code, and grid level to each grid cell, integrates all grid cells to form a multi-scale spatial grid cell set, and outputs it.

[0071] This step inherits the original attributes to ensure the traceability of grid cells and basic data. The aggregation of indicators at the industry-administrative region level enables grid cells to not only have spatial characteristics but also contain quantifiable employment quality information, providing a structured, multi-scale data foundation for subsequent hotspot analysis and trend monitoring.

[0072] In one embodiment of the present invention, based on the comprehensive quality score, a Gaussian kernel function is used to perform nonparametric density estimation, outputting an employment quality kernel density point cloud dataset; from the kernel density point cloud dataset, statistically significant employment quality hotspot point cloud clusters are extracted; the discrete hotspot point cloud clusters are converted into a continuous three-dimensional point cloud surface model, including:

[0073] The overall quality score is mapped to the center point of the corresponding grid cell, and radial density diffusion calculation is performed using a Gaussian kernel function as the primitive to generate a discrete kernel density point cloud containing latitude and longitude coordinates and density values. Specifically, firstly, the generated multi-scale spatial grid cell set is obtained, and the latitude and longitude coordinates of the center point of each grid cell are extracted. The generated overall quality score is used as the density value and associated with the center point of the corresponding grid cell. With each center point as the origin, the bandwidth parameter of the Gaussian kernel function is set, and radial density diffusion is performed in space: the closer the location is to the center point, the greater the density contribution received; the farther away, the smaller the contribution, until it becomes negligible after exceeding the bandwidth range. For each calculation point in space, the density contribution values ​​of all grid center points are accumulated to obtain the final density value of that point, forming a kernel density point cloud composed of a large number of discrete points, each point containing latitude and longitude coordinates and the corresponding density value.

[0074] Perform local spatial autocorrelation analysis on the kernel density point cloud:

[0075] Based on Getis-Ord Gi The statistical measure is used to calculate the hotspot significance Z-value for each point; points with Z-values ​​exceeding the confidence threshold are selected to form a significant hotspot cloud cluster.

[0076] Specifically, for each point in the kernel density point cloud, its neighborhood is defined, and the density values ​​of all points within that neighborhood are calculated. This is based on Getis-Ord Git. The statistical calculation method compares the density value of the target point with the average density value of points in its neighborhood to assess whether the target point is a locally high-density or low-density area. The Gi statistic calculated for each point is standardized and converted into a Z-value. A larger Z-value indicates that the point is more likely to be a hotspot (high-density area), while a smaller Z-value indicates that it is more likely to be a coldspot (low-density area). A confidence threshold is set, and points with Z-values ​​exceeding the threshold are selected. These points and their density values ​​are then combined into a significant hotspot cluster.

[0077] This step, through local spatial autocorrelation analysis, distinguishes between random high-density regions and statistically significant hotspot regions, avoiding misjudgment of accidental high-density points; the extraction of hotspot cloud clusters focuses on high-quality employment areas that truly have spatial clustering characteristics, providing precise targets for university resource allocation and employment guidance.

[0078] Using salient hotspot clusters as control points:

[0079] The spatial variogram model is fitted using the Kriging algorithm; a continuous three-dimensional point cloud surface model is generated based on semivariance final interpolation, and its elevation dimension represents the employment quality density value.

[0080] Specifically, the latitude and longitude coordinates and density values ​​of control points are extracted from salient hotspot point cloud clusters and used as known sample points. The spatial distance and density value differences between sample points are calculated, the semivariogram is analyzed as a function of distance, a suitable spatial variogram model is fitted, and model parameters, such as nugget value, sill value, and range, are determined. Based on the fitted spatial variogram model, Kriging interpolation is performed on the entire study area: for any point to be estimated, the weight of each control point to that point is calculated according to its spatial relationship with the control points and the variogram model, and a weighted average is used to obtain the predicted density value of that point. The predicted density values ​​of all points to be estimated are combined with their corresponding latitude and longitude coordinates to generate a continuous three-dimensional point cloud surface model, where the X and Y coordinates represent geographical location, and the Z coordinate represents the employment quality density value.

[0081] In one embodiment of the present invention, an octree spatial index is used to perform density adaptive resampling on a 3D point cloud surface model to generate uniformly distributed employment quality point cloud data; the point cloud normal vector is estimated, and an enhanced point cloud dataset containing spatial coordinates, density values, and normal vectors, along with its attribute association information, is constructed, including:

[0082] The 3D point cloud surface model is imported into an octree spatial index. Cube voxels are initialized according to a preset resolution threshold, and the representative point with the highest density value is retained for each non-empty voxel. Specifically, the spatial boundary of the 3D point cloud surface model is determined, and a cube space that completely encloses the model is constructed. Based on the preset initial resolution threshold, this cube space is divided into multiple cube voxels of equal size. All points in the 3D point cloud surface model are traversed, and each point is assigned to a corresponding voxel. For each non-empty voxel, the density values ​​of all points within it are compared, and the point with the highest density value is selected as the representative point of that voxel. The spatial coordinates and density value of this point are recorded, and other points are discarded, completing the initialization of the octree spatial index.

[0083] The uniformity of point cloud distribution within a voxel is recursively checked. When the variance of the point spacing exceeds the tolerance, the voxel is subdivided into eight equal parts, iterating until the uniformity threshold is met, and outputting a density-balanced employment quality point cloud. Specifically, for each initialized voxel, the distance between its representative points is calculated, and the variance of the point spacing is statistically analyzed. The variance of the point spacing is compared with a preset uniformity tolerance. If the variance exceeds the tolerance, it indicates that the point cloud distribution within the voxel is not uniform. The voxel is then divided into eight smaller sub-voxels along each of the three coordinate axes. The points within the original voxel are redistributed to the sub-voxels. The above process of calculating and comparing the variance of the point spacing is repeated for each non-empty sub-voxel. If the sub-voxels still do not meet the uniformity threshold, the subdivision continues until the variance of the point spacing of all voxels is within the tolerance range. The representative points of all voxels that meet the uniformity requirements are collected to form a density-balanced employment quality point cloud.

[0084] Local neighborhood analysis is performed on the resampled point cloud to extract the K nearest neighbors for each point and construct a covariance matrix. Specifically, for each point in the resampled employment quality point cloud, its neighborhood is determined, typically using the K nearest neighbors method, which finds the K nearest points to that point to form its local neighborhood set. A local coordinate system is established with this point as the origin, and the coordinates of all points within the neighborhood are transformed to this local coordinate system. Based on the transformed coordinate values, the mean of the neighborhood set along the three coordinate axes is calculated. The mean of the corresponding coordinate axis is subtracted from the coordinate value of each point to obtain the centered coordinate values. Using the centered coordinate values, the covariance between the three coordinate axes is calculated, constructing a 3×3 covariance matrix, where each element represents the covariance value between two coordinate axes.

[0085] Principal component analysis (PCA) is used to solve for the eigenvectors of the covariance matrix. The vector corresponding to the smallest eigenvalue is then used as the normal vector of the point cloud. Specifically, for each constructed point, the eigenvalues ​​and eigenvectors of the covariance matrix are solved. Eigenvalue decomposition is typically used to obtain three eigenvalues ​​and three corresponding eigenvectors, each representing a principal direction. The magnitudes of the three eigenvalues ​​are compared, and the smallest eigenvalue is identified. Its corresponding eigenvector is the normal vector of the point cloud surface at that point. The direction of the normal vector is determined by choosing the direction pointing outward from the point cloud surface as the positive direction, which can be determined by the sign of the dot product with the global reference direction.

[0086] An enhanced point cloud dataset is generated by binding spatial coordinates, density values, and normal vectors. For each point in the resampled point cloud, its spatial coordinates, density values, and calculated normal vectors are combined to form a point data structure containing seven-dimensional information. Attribute association information, such as the industry category, administrative division code, and original grid cell ID, is added to each point for subsequent querying and analysis. The seven-dimensional information and attribute association information of all points are integrated to generate an enhanced point cloud dataset that includes not only spatial location and employment quality density information but also surface orientation features.

[0087] The enhanced point cloud dataset in this step enriches the information dimensions of the original point cloud. The combination of spatial coordinates, density values, and normal vectors enables the point cloud to more comprehensively express the spatial distribution characteristics and surface morphology of employment quality, providing a richer data foundation for subsequent 3D visualization, spatial analysis, and decision support.

[0088] In one embodiment of the present invention, the comprehensive quality scores are aggregated by session, the year-on-year change rate of the average comprehensive quality score of adjacent sessions is calculated, and a trend signal is generated, including:

[0089] S401, aggregate the overall quality scores by session and calculate the average overall quality score for that session;

[0090] S402. The year-on-year change rate is obtained by subtracting the average comprehensive quality score of the previous session from the average comprehensive quality score of the current session and dividing by the average comprehensive quality score of the previous session.

[0091] S403 compares the year-on-year change rate with a preset positive threshold and a preset negative threshold respectively. When the year-on-year change rate is not less than the preset positive threshold, an upward trend signal is generated; when the year-on-year change rate is not greater than the preset negative threshold, a downward trend signal is generated; and when it is between the preset positive threshold and the preset negative threshold, a stable trend signal is generated. Here, the preset positive threshold represents the critical value for a significant increase, and the preset negative threshold represents the critical value for a significant decrease. An upward trend signal indicates a significant increase in employment quality, and a downward trend signal indicates a significant decrease in employment quality.

[0092] This embodiment eliminates the interference of individual differences on the overall trend by aggregating the average values ​​of each session, making changes in employment quality more representative. The calculation of the year-on-year change rate enables quantitative comparison between adjacent sessions, avoiding trend misjudgments caused by absolute value comparisons. The generation of trend signals based on thresholds transforms the abstract change rate into an intuitive decision-making basis, facilitating schools to quickly identify the rising and falling trends in employment quality. Simultaneously, the trend signal can be directly fed back to the gap diagnosis module, providing a time-dimensional reference for the formulation of subsequent improvement measures, enabling the system to dynamically monitor the quality of talent cultivation.

[0093] In one embodiment of the present invention, all employment indicators with differences exceeding a preset tolerance and their corresponding differences are written into an improvement trigger signal; the ratio of the difference to the largest difference among the employment indicators is multiplied by a preset learning rate to obtain a weight correction amount, which is then written into the improvement trigger signal; the improvement trigger signal carries the above information and feeds it back to the quality scoring module, which dynamically updates the original indicator weights according to the weight correction amount, thereby achieving adaptive adjustment of the evaluation system.

[0094] This embodiment refines the over-tolerance index and the difference, making the improvement measures clearly targeted and avoiding blind adjustments. The calculation of the weight correction amount is based on the product of the relative difference and the learning rate, which not only ensures the correlation between the correction magnitude and the index gap, but also avoids drastic fluctuations in weights through learning rate control, thus ensuring the stability of the evaluation system.

[0095] In one embodiment of the present invention, a route optimization module is also included, which is used to calculate the travel time of multiple routes from the school to the internship unit based on the geographical location of the school and the internship unit, call road network data and the real-time speed of each road segment, select the shortest route to generate an internship inspection plan.

[0096] Specifically, the route optimization module receives the school's geographical location and the student's selected learning unit's geographical location as the starting and ending points. This module calls road network data provided by a third-party map service via an interface, including road segment lengths, road types, and real-time traffic speeds. Based on the road network data, it generates multiple candidate routes, each composed of continuous road segments, and calculates their travel time. By traversing all candidate routes, it selects the route with the shortest travel time as the optimal route. Then, based on the optimal route, the module generates an internship inspection plan, marks key nodes on the optimal route, and, combined with a preset inspection duration, plans the inspection departure time, arrival time at intermediate nodes, and key points for on-site inspection, forming a structured inspection plan document, which is then simultaneously pushed to the responsible teacher's terminal device.

[0097] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0098] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. A GIS-based application-oriented higher education talent training quality tracking service system, characterized in that, include: The data acquisition and positioning module is used to collect student information, alumni information, internship logs and employer information and represent them as employment data. It performs standardized processing on the employment data, converts the included address fields into latitude and longitude coordinates, and generates standardized employment indicator data and spatial element data. The quality scoring module is used to weight the employment indicators in the standardized employment indicator data according to preset weights to generate a comprehensive quality score. The spatial analysis module is used to build a quadtree spatial index for spatial element data based on industry category and administrative division, and generate initial spatial partitions; Based on the feature distribution density within the initial spatial partition, a quadtree depth-first traversal is performed: when the number of features in a grid exceeds a preset density threshold, the quadtree subgrid is recursively divided into subgrids, with each subgrid inheriting the industry and administrative division attributes of the parent grid; when the number of features in a grid is below the minimum density threshold, adjacent grids are merged. The final generated grid cells inherit the attributes of the original spatial elements, aggregate employment indicator statistics by industry-administrative region dimension, generate the employment quality feature value of the grid cell, and output a multi-scale spatial grid cell set with attribute labels; The overall quality score is mapped to the center point of the corresponding grid cell, and radial density diffusion calculation is performed using the Gaussian kernel function as the basic unit to generate a discrete kernel density point cloud containing latitude and longitude coordinates and density values. Perform local spatial autocorrelation analysis on kernel density point clouds: calculate the hotspot significance Z-value of each point in the kernel density point cloud based on the Getis-Ord Gi statistic; select points with Z-values ​​exceeding the confidence threshold to form significant hotspot point cloud clusters; The latitude and longitude coordinates and density values ​​of control points are extracted from salient hot spot point cloud clusters and used as known sample points. The spatial variogram model is fitted using the Kriging algorithm. For any point to be estimated, the weight of each control point to the point to be estimated is calculated based on its spatial position relationship with the control points and the variogram model. The weighted average is used to obtain the predicted density value of the point to be estimated. The predicted density values ​​of all points to be estimated are combined with the corresponding latitude and longitude coordinates to generate a continuous three-dimensional point cloud surface model. Its elevation dimension represents the employment quality density value. Import the 3D point cloud surface model into the octree spatial index, initialize the cube voxels according to the preset resolution threshold, and retain the representative point with the largest density value for each non-empty voxel. The uniformity of point cloud distribution within a voxel is recursively checked. When the variance of the point spacing exceeds the tolerance, the voxel is subdivided into eight equal parts. The process is iterated until the uniformity threshold is met. The set of representative points in all voxels that meet the uniformity requirements is output to form a density-balanced employment quality point cloud. Perform local neighborhood analysis on the employment quality point cloud, and extract the K nearest neighbor set of each point to construct the covariance matrix; The covariance matrix eigenvectors are obtained by principal component analysis, and the vector corresponding to the smallest eigenvalue is used as the point cloud normal vector. An enhanced point cloud dataset is generated by binding spatial coordinates, density values, and normal vectors. The enhanced point cloud dataset includes spatial location and employment quality density information. The trend monitoring module is used to summarize the comprehensive quality scores by session, calculate the year-on-year change rate of the average comprehensive quality scores of adjacent sessions, and generate trend signals. The gap diagnosis module is used to calculate the average value of the current employment indicators and compare it with the corresponding preset target threshold to obtain the difference. When the difference of any employment indicator exceeds the preset tolerance, an improvement trigger signal is generated and the improvement trigger signal containing the weight adjustment is fed back to the quality scoring module.

2. The GIS-based application-oriented higher education talent training quality tracking service system according to claim 1, characterized in that, The employment indicators include: on-time employment rate, salary index, job matching degree, employer satisfaction, graduate satisfaction, and skills utilization.

3. The GIS-based application-oriented higher education talent training quality tracking service system according to claim 1, characterized in that, The standardization process includes: S201, Perform missing value imputation on the collected employment data; S202, perform outlier removal on the filled employment data, and exclude outlier data that exceeds the preset threshold range; S203, based on the employment data of the current graduates after removing outliers, calculate the sample mean and sample standard deviation of each employment indicator; S204, based on the sample mean and sample standard deviation, normalizes the employment data after outlier removal to zero mean and unit variance to obtain standardized employment indicator data.

4. The GIS-based application-oriented higher education talent training quality tracking service system according to claim 1, characterized in that, The overall quality scores are aggregated by session, and the year-on-year change rate of the average overall quality score between adjacent sessions is calculated to generate trend signals, including: S401, aggregate the overall quality scores by session and calculate the average overall quality score for that session; S402. The year-on-year change rate is obtained by subtracting the average comprehensive quality score of the previous session from the average comprehensive quality score of the current session and dividing by the average comprehensive quality score of the previous session. S403 compares the year-on-year change rate with a preset positive threshold and a preset negative threshold respectively. When the year-on-year change rate is not less than the preset positive threshold, an upward trend signal is generated. When the year-on-year change rate is not greater than the preset negative threshold, a downward trend signal is generated. When it is between the preset positive threshold and the preset negative threshold, a stable trend signal is generated.

5. The GIS-based application-oriented higher education talent training quality tracking service system according to claim 1, characterized in that, Write all employment indicators with differences exceeding the preset tolerance and their corresponding differences into the improvement trigger signal; multiply the ratio of the difference to the largest difference among the employment indicators by the preset learning rate to obtain the weight correction amount and write it into the improvement trigger signal.

6. The GIS-based application-oriented higher education talent training quality tracking service system according to claim 1, characterized in that, It also includes a route optimization module, which uses road network data and real-time speeds of various road segments to calculate the travel time of multiple routes from the school to the internship unit based on the geographical location of the school and the internship unit, and selects the shortest route to generate an internship inspection plan.

Images