Method for evaluating vocational education effect based on learning behavior data mining

By constructing an implicit representation model of cognitive style and a skills knowledge graph, this study addresses the assessment bias caused by neglecting differences in cognitive style in existing vocational education assessment methods. It achieves accurate, objective, and personalized assessment of learners' skill mastery, provides personalized learning suggestions, and improves educational effectiveness.

CN121526857BActive Publication Date: 2026-04-21NANCHONG VOCATIONAL & TECH COLLEGE
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610055557.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-01-16
Publication Date
2026-04-21
Estimated Expiration
2046-01-16

AI Technical Summary

Technical Problem

Existing methods for evaluating the effectiveness of vocational education fail to effectively distinguish the differences in behavioral patterns among learners with different cognitive styles. They treat the same behavioral indicators with different learning intentions indiscriminately, resulting in systematic biases in the evaluation results and failing to accurately reflect learners' true skill mastery levels.

Method used

By constructing an implicit representation model of learners' cognitive styles, the original learning behavior data is reconstructed with style-adaptive features, and combined with a skill knowledge graph for hierarchical evaluation, so as to achieve accurate, objective and personalized measurement of the skill mastery of different types of learners.

Benefits of technology

It achieves a precise characterization of individual learners' differences, eliminates assessment bias caused by differences in cognitive styles, improves the objectivity and fairness of assessment, and generates assessment reports containing personalized learning suggestions, thereby improving the overall effectiveness and efficiency of vocational education.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121526857B_ABST
    Figure CN121526857B_ABST
Patent Text Reader

Abstract

This invention relates to the field of computer technology and discloses a method for evaluating the effectiveness of vocational education based on learning behavior data mining. The method includes: collecting learners' full-domain learning behavior data sequences; generating cognitive style feature vectors through a self-attention sequence encoder; dynamically weighting the behavioral features using style gating units to form a style-weighted effective behavioral feature matrix; combining a pre-constructed vocational skills knowledge graph and aggregating the behavioral features layer by layer through a graph neural network to calculate the mastery score of each skill node; and finally generating an evaluation report containing multi-dimensional evaluation results and personalized suggestions. This invention achieves accurate, objective, and personalized skills assessment for learners with different cognitive styles, improving the fairness and diagnostic value of vocational education evaluation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of computer technology, specifically relating to a method for evaluating the effectiveness of vocational education based on learning behavior data mining. Background Technology

[0002] With the deepening of the digital transformation of vocational education, learning effectiveness evaluation methods based on learning behavior data are gradually becoming a core means of monitoring educational quality and providing personalized teaching support. Current mainstream evaluation systems generally rely on statistical analysis of learners' online behavior logs, quantifying learning effectiveness by constructing a unified behavior-ability mapping model.

[0003] However, such methods implicitly assume key assumptions:

[0004] The same behavioral indicators have homogeneous cognitive meaning and ability representation value among different learners.

[0005] This hypothesis ignores the fundamental differences in cognitive styles among learners. In particular, in skills-based courses, visual learners tend to internalize knowledge by repeatedly watching demonstration videos, while auditory learners may rely more on audio explanations or interactive discussions. This results in the same behavioral indicator showing drastically different correlations between learning intentions and abilities in different cognitive groups.

[0006] Cognitive style, as a key individual difference factor influencing information processing methods, directly determines learners' preferences and usage strategies for multimodal teaching resources. Existing vocational education assessment models generally fail to incorporate cognitive style into their modeling framework. They lack both an automatic mechanism for identifying learners' cognitive types and a logic for dynamically adjusting the weighting of behavioral indicators based on their style characteristics. This one-size-fits-all assessment paradigm is highly susceptible to distorted assessment results due to interpretative biases when facing a highly heterogeneous vocational education population, failing to accurately reflect learners' skill mastery levels.

[0007] While some existing studies have attempted to introduce questionnaires or explicit labeling to obtain cognitive style information, these methods rely on subjective reporting and suffer from problems such as data lag, incomplete coverage, and questionable authenticity. Research on implicitly inferring cognitive style based on behavioral data is still in its early stages and has not been deeply integrated with assessment models.

[0008] More importantly, the current assessment system lacks a closed-loop mechanism that can adaptively feed the cognitive style recognition results back to the assessment weight calculation process in real time, which makes it impossible to achieve true personalization in the assessment process.

[0009] In highly sensitive application scenarios such as vocational skills certification, job matching recommendations, and teaching intervention decisions, such systemic biases may lead to misjudgments, seriously affecting the fairness and effectiveness of vocational education.

[0010] Therefore, there is an urgent need for an intelligent vocational education effectiveness evaluation method that can integrate cognitive style self-identification and dynamic allocation of evaluation weights. Summary of the Invention

[0011] The technical problem that this invention aims to solve is that existing vocational education effectiveness evaluation methods fail to effectively distinguish the differences in behavioral patterns among learners with different cognitive styles, and treat the same behavioral indicators with different learning intentions indiscriminately, resulting in systematic biases in the evaluation results and failing to accurately reflect the learners' true skill mastery level.

[0012] To overcome the shortcomings of the prior art, this invention provides a method for evaluating the effectiveness of vocational education based on learning behavior data mining. This method constructs an implicit representation model of learners' cognitive styles, performs style-adaptive feature reconstruction on the original learning behavior data, and combines it with a skill knowledge graph for hierarchical evaluation, thereby achieving accurate, objective, and personalized measurement of the skill mastery of different types of learners.

[0013] This invention provides a method for evaluating the effectiveness of vocational education based on learning behavior data mining, the method comprising the following steps:

[0014] Obtain a sequence of learners' full-domain learning behavior data on a vocational education platform. The sequence of full-domain learning behavior data includes timestamps, behavior types, associated learning resource identifiers, and behavior metadata.

[0015] The whole-domain learning behavior data sequence is input into a preset cognitive style implicit representation network. The cognitive style implicit representation network generates a cognitive style feature vector that represents the learner's cognitive style tendency based on the distribution and transformation pattern of interaction behavior of different types of learning resources in the sequence.

[0016] The whole-domain learning behavior data sequence and the cognitive style feature vector are input into the cognitive style adaptive behavior feature weighting module. The module uses the cognitive style feature vector to dynamically weight each behavior indicator in the whole-domain learning behavior data sequence to generate a style-weighted effective behavior feature matrix.

[0017] Construct a predefined occupational skill knowledge graph, which consists of skill nodes and edges representing skill dependencies;

[0018] The style-weighted effective behavioral feature matrix and the professional skills knowledge graph are input into the hierarchical skills mastery assessment network. The hierarchical skills mastery assessment network aggregates and propagates learning behavioral features layer by layer according to the topology of the graph and the dependencies between nodes, calculates the learner's mastery score at each skill node, and forms a multi-dimensional skills mastery assessment vector.

[0019] Based on the multi-dimensional skills mastery assessment vector, a vocational education effectiveness evaluation report is generated.

[0020] As one embodiment of the present invention, the acquisition of the global learning behavior data sequence specifically includes:

[0021] By connecting with the database interface of the learning management system, learners' course access records, learning unit completion status and online time data are collected in real time;

[0022] By deploying event listening scripts in front of the video player, learners' video playback, pause, progress bar dragging, speed-up playback, and loop playback behaviors are captured, and the specific time point and video content segment of each behavior are recorded.

[0023] By interacting with the application programming interface of online document readers, data on learners' opening frequency, reading duration, scrolling depth, annotation and note-taking behavior of teaching documents can be obtained;

[0024] By analyzing the operation logs of online programming environments or simulation training platforms, learners can extract data such as the number of code submissions, compilation error types and frequencies, simulation operation sequences, and task completion time.

[0025] By querying the database of the quiz and assignment system, we can obtain data on learners' submitted answers, scores, answering time, and distribution of incorrect options for questions corresponding to each knowledge node.

[0026] The data obtained above for the whole-domain learning behavior data sequence are strictly sorted according to timestamps, and the heterogeneous data are normalized to form a time series data stream with a unified format. Each data record is a structured data unit containing user identifier, timestamp, behavior type code, resource identifier and detailed parameters.

[0027] In one embodiment of the present invention, the cognitive style implicit representation network is a sequence encoder structure based on a self-attention mechanism. The specific operation of inputting the global learning behavior data sequence into the cognitive style implicit representation network is as follows:

[0028] First, each behavioral data in the global learning behavioral data sequence is vectorized and embedded to generate a behavioral event embedding vector, which is composed of behavioral type embedding and resource type embedding.

[0029] Secondly, embed vector sequences with superimposed positional encodings into the behavioral events to inject temporal sequence information of the behavior;

[0030] Then, the event embedding vector sequence carrying location information is input into the sequence encoder based on the self-attention mechanism. The sequence encoder based on the self-attention mechanism calculates the association weight between any two behavioral events in the sequence through a multi-head self-attention layer, capturing the long-range dependency and switching preference of learners between different learning activities.

[0031] Finally, the aggregated representation vector output by the sequence encoder based on the self-attention mechanism for the entire behavioral sequence is identified as the cognitive style feature vector. This vector quantifies and locates the learner's cognitive style in a pre-defined multi-dimensional vector space, with different dimensions corresponding to the intensity of different cognitive styles such as visual, auditory, and kinesthetic.

[0032] As one embodiment of the present invention, the cognitive style adaptive behavior feature weighting module integrates a style gating unit. The dynamic weighting of various behavior indicators using cognitive style feature vectors is specifically implemented as follows:

[0033] The cognitive style feature vector is input into the style gating unit, which is composed of a fully connected neural network and an activation function.

[0034] The fully connected neural network takes the cognitive style feature vector as input and outputs a weight adjustment vector with the same dimension as the original behavioral feature vector.

[0035] The weight adjustment vector is subjected to a nonlinear transformation by the activation function, such as a sigmoid function, to generate a gated signal vector with values ​​between 0 and 1.

[0036] The gating signal vector is multiplied element-wise with the original multi-dimensional behavioral feature vector extracted from the global learning behavior data sequence, thereby dynamically enhancing or suppressing different behavioral features. For the cognitive style feature vector of a learner identified as a strong visual learner, the gating unit will automatically increase the weight of behavioral features related to video viewing and interaction with text and image materials, while decreasing the weight of behavioral features related to audio listening and plain text reading.

[0037] In one embodiment of the present invention, the hierarchical skill mastery assessment network is a graph neural network. The step of inputting the style-weighted effective behavioral feature matrix and the vocational skill knowledge graph into the hierarchical skill mastery assessment network specifically includes:

[0038] Each skill node in the vocational skills knowledge graph is initialized as a skill node feature vector. The skill node feature vector is formed by aggregating the static attributes of all learning resources associated with the skill, such as difficulty coefficient and resource type, with the behavioral features corresponding to these resources in the style-weighted effective behavioral feature matrix.

[0039] The graph neural network operates through a multi-layer iterative information propagation mechanism. In each layer, each skill node aggregates information about its preceding dependent skill nodes in the knowledge graph.

[0040] The information aggregation process employs an attention-based aggregation function, which dynamically calculates the weight of each neighbor node's information based on the relationship between the central node and its neighbor nodes, thereby simulating the different contributions of mastering different basic skills to learning advanced skills.

[0041] After multiple layers of information propagation and aggregation, the feature vector of each skill node fully integrates the learning effect information of all its prerequisite dependent skills;

[0042] The feature vectors of each skill node are then input into a shared linear classification layer and passed through an output activation function to calculate the learner's final mastery score for that skill, which ranges from 0 to 1.

[0043] As one embodiment of the present invention, the generation of the vocational education effectiveness evaluation report specifically includes:

[0044] The multi-dimensional skill mastery assessment vector is displayed graphically, such as using radar charts or bar charts, to visually present the distribution of learners' strengths and weaknesses in each skill point.

[0045] For skill nodes whose mastery score is less than a preset threshold, the system automatically traces back their dependency path in the professional skills knowledge graph to locate the prerequisite basic skills that they have weak mastery of.

[0046] By combining learners' cognitive style feature vectors, the system generates personalized learning suggestions. For visual learners, it recommends instructional videos and interactive diagrams related to their weak skills, while for auditory learners, it recommends relevant audio explanations and online discussion groups.

[0047] This invention also provides a vocational education effectiveness evaluation system based on learning behavior data mining, the system comprising:

[0048] The full-domain learning behavior data acquisition module is used to interact with various subsystems of the learning platform, capture and integrate learners' multimodal learning behaviors in real time, and form a unified format of full-domain learning behavior data sequence.

[0049] The cognitive style implicit representation module has a self-attention mechanism-based sequence encoder embedded in it, which is used to receive the global learning behavior data sequence and output a cognitive style feature vector representing the learner's cognitive style.

[0050] A cognitive style adaptive behavior feature weighting module, which is connected to the cognitive style implicit representation module, is used to receive the global learning behavior data sequence and the cognitive style feature vector, and generate a style-weighted effective behavior feature matrix through an internal style gating unit;

[0051] The professional skills knowledge graph storage module is used to store graph-structured data that describes the dependencies between skills, pre-built by domain experts.

[0052] A hierarchical skill mastery assessment module is connected to the behavior feature weighting module and the knowledge graph storage module. It has a graph neural network model embedded in its core, which is used to combine the style-weighted effective behavior feature matrix and the professional skill knowledge graph to calculate and output the learner's multi-dimensional skill mastery assessment vector.

[0053] The assessment report generation module receives the multi-dimensional skill mastery assessment vector and transforms it into a structured assessment report containing diagnostic information and personalized learning suggestions.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. By introducing a cognitive style implicit representation module, this invention can automatically and dynamically mine learners’ inherent learning style preferences from their continuous behavioral sequences, solving the problem of assessment model mismatch caused by traditional assessment methods treating all learners as a homogeneous group, and achieving accurate characterization of learners’ individual differences.

[0056] 2. The cognitive style adaptive behavioral feature weighting module designed in this invention can dynamically adjust the interpretation of the original behavioral indicators according to the determined learner's cognitive style, so that the same behavior has different evaluation meanings in different learner backgrounds, thereby fundamentally eliminating the evaluation bias caused by differences in cognitive style and improving the objectivity and fairness of the evaluation.

[0057] 3. This invention adopts a hierarchical skill mastery assessment model based on graph neural networks, which closely integrates the assessment process with the inherent logical structure of vocational skills. It can not only assess the mastery of isolated knowledge points, but also accurately diagnose the breakpoints of learners in the skill dependency chain, making the assessment results have stronger diagnostic value.

[0058] 4. The final assessment report generated by this invention not only includes a quantitative mastery score, but also provides highly personalized and actionable learning path suggestions by combining cognitive styles and skill maps, realizing a closed loop from "assessment" to "empowerment", providing learners with precise follow-up learning guidance, thereby comprehensively improving the overall effectiveness and efficiency of vocational education. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the overall technical solution architecture of the vocational education effectiveness evaluation method based on learning behavior data mining proposed in this invention;

[0060] Figure 2 This is a schematic diagram of the core principle framework of the cognitive style implicit representation network in this invention;

[0061] Figure 3 This is a logical flow diagram of the cognitive style adaptive behavioral feature weighting module in this invention;

[0062] Figure 4 This is a schematic diagram illustrating the principle framework of the fusion of hierarchical skill mastery assessment network and vocational skill knowledge graph in this invention.

[0063] Figure 5 This is a flowchart illustrating the multi-source data integration logic of the full-domain learning behavior data collection and normalization processing in this invention.

[0064] Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow of the skill node information propagation and aggregation mechanism based on graph neural networks in this invention;

[0065] Figure 7 This is a flowchart illustrating the diagnostic and recommendation logic of the personalized assessment report generation stage in this invention.

[0066] Figure 8 This is a schematic diagram of the overall data flow and feedback loop of the synergistic effect of learners' cognitive style, behavioral characteristics and skill mastery in this invention. Detailed Implementation

[0067] Please refer to Figures 1 to 8 This invention provides a method for evaluating the effectiveness of vocational education based on learning behavior data mining, aiming to solve the systematic bias problem caused by the neglect of differences in learners' cognitive styles in existing vocational education effectiveness evaluation models.

[0068] This method constructs an implicit representation model of cognitive style, performs style-adaptive feature reconstruction on the original learning behavior data, and combines it with a vocational skills knowledge graph to conduct a hierarchical assessment of skill mastery, thereby achieving a precise, objective, and personalized measurement of the skill mastery level of different types of learners. The following will describe in detail the specific implementation steps of this method.

[0069] The method includes the following steps:

[0070] S1, Obtain the sequence of learners' full-domain learning behavior data on the vocational education platform;

[0071] S2, input the global learning behavior data sequence into a preset cognitive style implicit representation network to generate a cognitive style feature vector;

[0072] S3, input the global learning behavior data sequence and the cognitive style feature vector into the cognitive style adaptive behavior feature weighting module to generate a style-weighted effective behavior feature matrix;

[0073] S4, construct a predefined professional skills knowledge graph;

[0074] S5, input the style-weighted effective behavioral feature matrix and the professional skills knowledge graph into the hierarchical skills mastery assessment network to calculate the multi-dimensional skills mastery assessment vector.

[0075] S6. Based on the multi-dimensional skill mastery assessment vector, generate a vocational education effectiveness evaluation report.

[0076] In step S1, a sequence of learners' full-domain learning behavior data on the vocational education platform is obtained. The full-domain learning behavior data sequence includes timestamps, behavior types, associated learning resource identifiers, and behavior metadata.

[0077] Specifically, this step involves connecting to the database interface of the learning management system to collect learners' course access records, learning unit completion status, and online time data in real time.

[0078] By deploying event listening scripts in front of the video player, learners' video playback, pause, progress bar dragging, speed-up playback, and loop playback behaviors are captured, and the specific time point and video content segment of each behavior are recorded.

[0079] By interacting with the application programming interface of online document readers, data on learners' opening frequency, reading duration, scrolling depth, annotation and note-taking behavior of teaching documents can be obtained;

[0080] By analyzing the operation logs of online programming environments or simulation training platforms, learners can extract data such as the number of code submissions, compilation error types and frequencies, simulation operation sequences, and task completion time.

[0081] By querying the database of the quiz and assignment system, we can obtain data on learners' submitted answers, scores, answering time, and distribution of incorrect options for questions corresponding to each knowledge node.

[0082] All data from the above sources are strictly sorted according to a unified time base, and heterogeneous data are normalized to form a time series data stream in a unified format.

[0083] Each data record is a structured data unit, containing user identifier, timestamp, behavior type code, resource identifier, and detailed parameters.

[0084] The behavior type encoding adopts a predefined enumeration value system. For example, video playback is encoded as 001, document reading as 002, code submission as 003, and quiz answering as 004.

[0085] Resource identifiers are globally unique to ensure that different learning resources are uniquely identifiable within the system.

[0086] Detailed parameter fields are dynamically filled based on the behavior type. For example, for video playback behavior, parameter fields include start time, end time, playback speed, and whether to loop; for quiz answering behavior, parameter fields include question number, selected option, correct or incorrect, and answering time.

[0087] This data sequence serves as the raw input for all subsequent analysis modules, and its integrity and temporal consistency are fundamental to ensuring the accuracy of the evaluation results.

[0088] In step S2, the global learning behavior data sequence is input into a preset cognitive style implicit representation network. The cognitive style implicit representation network generates a cognitive style feature vector that represents the learner's cognitive style tendency based on the distribution and transformation patterns of interaction behaviors of different types of learning resources in the sequence.

[0089] The cognitive style implicit representation network is a sequence encoder structure based on a self-attention mechanism.

[0090] First, each behavioral data point in the global learning behavioral data sequence is vectorized and embedded to generate a behavioral event embedding vector. The behavioral event embedding vector is composed of behavioral type embedding and resource type embedding.

[0091] Behavior type embedding is implemented through a trainable embedding lookup table that maps discrete behavior type encodings to dense vectors of fixed dimensions; resource type embedding is also implemented through another embedding lookup table that maps resource identifiers to another set of dense vectors.

[0092] The two are concatenated to form a high-dimensional behavior event embedding vector.

[0093] Next, the embedding vector sequence of the behavioral event is superimposed with position encoding to inject the temporal sequence information of the behavior.

[0094] Position encoding uses a combination of sine and cosine functions to ensure that the model can distinguish events at different positions in the sequence.

[0095] Then, the sequence of event embedding vectors carrying location information is input into a sequence encoder based on a self-attention mechanism.

[0096] This self-attention-based sequence encoder contains multiple stacked encoding layers, each consisting of a multi-head self-attention sublayer and a feedforward neural network sublayer.

[0097] Multi-head self-attention sublayers calculate the association weights between any two behavioral events in a sequence, capturing learners' long-range dependencies and switching preferences across different learning activities.

[0098] For example, if learners frequently engage in programming practice immediately after watching videos, the model will learn a strong correlation between the video-programming behavior transition pattern, which may indicate that they are kinesthetic learners.

[0099] After multiple layers of encoding, the aggregated representation vector of the entire sequence is taken, usually by average pooling all output vectors or by using output vectors with special labels such as [CLS], as the cognitive style feature vector.

[0100] This cognitive style feature vector quantifies and locates learners' cognitive styles within a pre-defined multi-dimensional vector space. Different dimensions correspond to the intensity of different cognitive styles, such as visual, auditory, and kinesthetic. The cognitive style feature vector does not rely on explicit questionnaires but is implicitly inferred entirely from behavioral data, thus possessing higher objectivity and real-time accuracy.

[0101] In step S3, the global learning behavior data sequence and the cognitive style feature vector are input into the cognitive style adaptive behavior feature weighting module. The module uses the cognitive style feature vector to dynamically weight each behavior indicator in the global learning behavior data sequence to generate a style-weighted effective behavior feature matrix.

[0102] The cognitive style adaptive behavioral feature weighting module integrates a style gating unit.

[0103] First, raw multi-dimensional behavioral feature vectors are extracted from the global learning behavior data sequence. This vector is obtained by statistically aggregating the sequence, such as calculating the total video viewing time, average document reading depth, code submission success rate, and average quiz score, to form a fixed-dimensional feature vector.

[0104] The cognitive style feature vector is then input into the style gating unit.

[0105] The style gating unit consists of a fully connected neural network and an activation function. The fully connected neural network takes the cognitive style feature vector as input and outputs a weight adjustment vector with the same dimensions as the original behavioral feature vector.

[0106] This fully connected neural network contains at least one hidden layer, whose weights are optimized during the model training phase using the backpropagation algorithm.

[0107] The weight adjustment vector is subjected to a nonlinear transformation by the activation function, such as a sigmoid function, to generate a gated signal vector with values ​​between 0 and 1.

[0108] Finally, the gated signal vector is multiplied element-wise with the original multidimensional behavioral feature vector to achieve dynamic enhancement or suppression of different behavioral features.

[0109] For the cognitive style feature vector of a learner identified as a strong visual learner, the gating unit will automatically increase the weight of behavioral features related to video viewing and interaction with text and image materials, while decreasing the weight of behavioral features related to audio listening and plain text reading.

[0110] This process ensures that the same behavioral indicators are given different assessment meanings by learners with different cognitive styles, fundamentally eliminating assessment bias caused by style differences.

[0111] In step S4, a predefined occupational skill knowledge graph is constructed, which consists of skill nodes and edges representing skill dependencies.

[0112] This professional skills knowledge graph was pre-constructed by domain experts based on professional standards and curriculum outlines.

[0113] Each skill node represents a specific and assessable professional skill, such as circuit fault diagnosis, CNC machine tool programming, and customer communication skills.

[0114] Node attributes include skill name, difficulty level, knowledge domain, and a list of associated learning resources.

[0115] Edges represent the dependencies between skills; for example, mastering Ohm's law is a prerequisite for diagnosing circuit faults.

[0116] The professional skills knowledge graph is stored in the system database in the form of an adjacency list or adjacency matrix for subsequent evaluation modules to access.

[0117] In step S5, the style-weighted effective behavioral feature matrix and the vocational skills knowledge graph are input into the hierarchical skills mastery assessment network. The hierarchical skills mastery assessment network aggregates and propagates learning behavior features layer by layer according to the topology of the graph and the dependencies between nodes, calculates the learner's mastery score at each skill node, and forms a multi-dimensional skills mastery assessment vector.

[0118] The hierarchical skill mastery assessment network is a graph neural network.

[0119] First, each skill node in the professional skills knowledge graph is initialized as a skill node feature vector.

[0120] The node feature vector is formed by aggregating the static attributes of all learning resources associated with the skill, such as difficulty coefficient and resource type, with the behavioral features corresponding to these resources in the style-weighted effective behavioral feature matrix.

[0121] For example, if skill A is associated with 3 video resources and 2 programming exercises, then the behavioral feature values ​​corresponding to these 5 resources are extracted from the style weighting matrix, weighted, and then concatenated with static attributes to form the initial node features. Then, the graph neural network performs computation through a multi-layer iterative information propagation mechanism.

[0122] In each layer, each skill node aggregates information about its preceding dependent skill nodes in the knowledge graph. The information aggregation process employs an aggregation function based on an attention mechanism.

[0123] The aggregation function based on the attention mechanism calculates the attention coefficient between the center node and each neighbor node, as shown in the following formula:

[0124] ;

[0125] and These are the feature vectors of the center node and its neighboring nodes, respectively. The weight matrix is ​​a learnable matrix. For the parameter vector of the attention mechanism, As the central node The neighborhood group, LeakyReLU represents vector concatenation and is a linear rectified function with leakage.

[0126] This attention mechanism dynamically calculates the weight of each neighbor node's information based on the relationship between the central node and its neighbor nodes, thereby simulating the different contributions of mastering different basic skills to learning advanced skills.

[0127] The formula for updating the aggregated node features is:

[0128] ;

[0129] For activation functions, ReLU is typically used. After multiple layers of information propagation and aggregation, the feature vector of each skill node fully integrates the learning effect information of all its preceding dependent skills. It is the updated feature vector of the central node i.

[0130] Finally, the feature vectors of each skill node are input into a shared linear classification layer and passed through an output activation function, such as a sigmoid function, to calculate the learner's final mastery score for that skill, which ranges from 0 to 1.

[0131] The scores of all skill nodes together constitute the multi-dimensional skill mastery assessment vector.

[0132] In step S6, a vocational education effectiveness evaluation report is generated based on the multi-dimensional skill mastery assessment vector.

[0133] The report presents the multi-dimensional skill mastery assessment vector in a graphical way, such as using radar charts or bar charts, to visually show the distribution of learners' strengths and weaknesses in each skill point.

[0134] For skill nodes whose mastery score is less than a preset threshold, the system automatically traces back their dependency paths in the professional skills knowledge graph to locate the prerequisite basic skills in which they are weak.

[0135] For example, if a user has a low level of proficiency in "complex circuit design", the system will check their proficiency in all prerequisite skills such as "basic circuit analysis" and "component identification" to identify the weakest link.

[0136] By combining learners' cognitive style feature vectors, the system generates personalized learning suggestions.

[0137] For visual learners, instructional videos and interactive diagrams related to their weak skills are recommended; for auditory learners, relevant audio explanations and online discussion groups are recommended.

[0138] This report not only provides diagnostic results, but also actionable follow-up learning pathways, forming a closed loop of assessment and intervention.

[0139] The system includes a full-domain learning behavior data acquisition module, a cognitive style implicit representation module, a cognitive style adaptive behavior feature weighting module, a professional skills knowledge graph storage module, a hierarchical skills mastery assessment module, and an assessment report generation module.

[0140] The full-domain learning behavior data acquisition module is responsible for interacting with various subsystems of the learning platform, capturing and integrating learners' multimodal learning behaviors in real time, and forming a unified format of full-domain learning behavior data sequence.

[0141] The cognitive style implicit representation module has a self-attention mechanism-based sequence encoder embedded inside, which is used to receive the global learning behavior data sequence and output a cognitive style feature vector representing the learner's cognitive style.

[0142] The cognitive style adaptive behavior feature weighting module is connected to the cognitive style implicit representation module. It is used to receive the global learning behavior data sequence and the cognitive style feature vector, and generate a style-weighted effective behavior feature matrix through the internal style gating unit.

[0143] The professional skills knowledge graph storage module is used to store graph-structured data that describes the dependencies between skills, pre-built by domain experts.

[0144] The hierarchical skill mastery assessment module is connected to the behavioral feature weighting module and the knowledge graph storage module. It contains a graph neural network model, which is used to combine the style-weighted effective behavioral feature matrix and the professional skill knowledge graph to calculate and output the learner's multi-dimensional skill mastery assessment vector.

[0145] The assessment report generation module receives the multi-dimensional skill mastery assessment vector and transforms it into a structured assessment report that includes diagnostic information and personalized learning suggestions.

[0146] The modules communicate with each other through standardized data interfaces to ensure efficient and consistent data flow.

[0147] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the effectiveness of vocational education based on learning behavior data mining, characterized in that, include: Obtain a sequence of learners' full-domain learning behavior data on a vocational education platform. The sequence of full-domain learning behavior data includes timestamps, behavior types, associated learning resource identifiers, and behavior metadata. The whole-domain learning behavior data sequence is input into a preset cognitive style implicit representation network. The cognitive style implicit representation network generates a cognitive style feature vector that represents the learner's cognitive style tendency based on the distribution and transformation pattern of interaction behavior of different types of learning resources in the sequence. The global learning behavior data sequence and the cognitive style feature vector are input into the cognitive style adaptive behavior feature weighting module to generate a style-weighted effective behavior feature matrix, including: Extract the original multi-dimensional behavioral feature vector from the global learning behavior data sequence; The cognitive style feature vector is input into a style gating unit, which consists of a fully connected neural network and an activation function. The fully connected neural network takes the cognitive style feature vector as input and outputs a weight adjustment vector with the same dimension as the original behavioral feature vector. The weight adjustment vector is nonlinearly transformed by the activation function to generate a gated signal vector with values ​​between 0 and 1. The gating signal vector is multiplied element-wise with the original multi-dimensional behavioral feature vector to generate the style-weighted effective behavioral feature matrix; Construct a predefined occupational skill knowledge graph, which consists of skill nodes and edges representing skill dependencies; The style-weighted effective behavioral feature matrix and the vocational skill knowledge graph are input into a hierarchical skill mastery assessment network, which is a graph neural network, to calculate a multi-dimensional skill mastery assessment vector, including: Each skill node in the vocational skill knowledge graph is initialized as a skill node feature vector. The skill node feature vector is formed by aggregating the static attributes of all learning resources associated with the skill and the behavioral features corresponding to these resources in the style-weighted effective behavioral feature matrix. Graph neural networks perform computations through a multi-layered iterative information propagation mechanism. In each layer, each skill node aggregates information about its preceding dependent skill nodes in the knowledge graph. The information aggregation process employs an attention-based aggregation function, which dynamically calculates the weight of each neighbor node's information based on the relationship between the central node and its neighbor nodes. After multiple layers of information propagation and aggregation, the feature vector of each skill node fully integrates the learning effect information of all its prerequisite dependent skills; The feature vectors of each skill node are input into a shared linear classification layer and passed through an output activation function to calculate the learner's final mastery score for that skill, thus forming the multi-dimensional skill mastery assessment vector. Based on the multi-dimensional skills mastery assessment vector, a vocational education effectiveness evaluation report is generated.

2. The method for evaluating the effectiveness of vocational education based on learning behavior data mining according to claim 1, characterized in that, Obtain the full-domain learning behavior data sequence of learners on the vocational education platform, including: By connecting with the database interface of the learning management system, learners' course access records, learning unit completion status and online time data are collected in real time; By deploying event listening scripts in front of the video player, learners' video playback, pause, progress bar dragging, speed-up playback, and loop playback behaviors are captured, and the specific time point and video content segment of each behavior are recorded. By interacting with the application programming interface of online document readers, data on learners' opening frequency, reading duration, scrolling depth, annotation and note-taking behavior of teaching documents can be obtained; By analyzing the operation logs of online programming environments or simulation training platforms, learners can extract data such as the number of code submissions, compilation error types and frequencies, simulation operation sequences, and task completion time. By querying the database of the quiz and assignment system, we can obtain data on learners' submitted answers, scores, answering time, and distribution of incorrect options for questions corresponding to each knowledge node. The acquired full-domain learning behavior data sequence is strictly sorted according to timestamps, and heterogeneous data is normalized to form a time series data stream with a unified format. Each data record is a structured data unit containing user identifier, timestamp, behavior type code, resource identifier and detailed parameters.

3. The method for evaluating the effectiveness of vocational education based on learning behavior data mining according to claim 2, characterized in that, The global learning behavior data sequence is input into a preset cognitive style implicit representation network to generate a cognitive style feature vector, including: Each behavioral data point in the global learning behavioral data sequence is vectorized and embedded to generate a behavioral event embedding vector, which is composed of behavioral type embedding and resource type embedding. The behavioral event is embedded with a vector sequence superimposed with position encoding to inject temporal sequence information of the behavior; The event embedding vector sequence carrying location information is input into a sequence encoder based on a self-attention mechanism. The sequence encoder based on a self-attention mechanism calculates the association weight between any two behavioral events in the sequence through a multi-head self-attention layer, thereby capturing the long-range dependency and switching preference of learners in different learning activities. The aggregated representation vector output by the sequence encoder based on the self-attention mechanism for the entire behavioral sequence is identified as the cognitive style feature vector.

4. The method for evaluating the effectiveness of vocational education based on learning behavior data mining according to claim 3, characterized in that, Construct a predefined occupational skills knowledge graph, including: Skill nodes and skill dependencies are pre-constructed by domain experts based on professional standards and curriculum outlines; Each skill node represents a specific and assessable professional skill. Skill node attributes include skill name, difficulty level, knowledge domain, and a list of associated learning resources. Edges represent dependencies between skills and are used to describe the logical dependency structure between prerequisite skills and subsequent skills; The professional skills knowledge graph is stored in the system database in the form of an adjacency list or an adjacency matrix.

5. The method for evaluating the effectiveness of vocational education based on learning behavior data mining according to claim 4, characterized in that, Each skill node in the vocational skills knowledge graph is initialized as a skill node feature vector, including: Extract the static attributes of all learning resources associated with the skill node, including difficulty level and resource type; Extract behavioral feature values ​​corresponding to these learning resources from the style-weighted effective behavioral feature matrix; The behavioral feature values ​​are weighted and averaged, and then concatenated with the static attributes to form an initial skill node feature vector.

6. The method for evaluating the effectiveness of vocational education based on learning behavior data mining according to claim 5, characterized in that, The information aggregation process uses an attention-based aggregation function to calculate the attention coefficient between the central node and each neighboring node, as shown in the formula: ; and These are the feature vectors of the center node and its neighboring nodes, respectively. The weight matrix is ​​a learnable matrix. For the parameter vector of the attention mechanism, As the central node The neighborhood group, LeakyReLU represents vector concatenation and is a linear rectified function with leakage.

7. The method for evaluating the effectiveness of vocational education based on learning behavior data mining according to claim 6, characterized in that, The formula for updating the aggregated node features is: ; For activation function, It is the updated feature vector of the central node i.

8. The method for evaluating the effectiveness of vocational education based on learning behavior data mining according to claim 7, characterized in that, Based on the aforementioned multi-dimensional skill mastery assessment vector, a vocational education effectiveness evaluation report is generated, including: The multi-dimensional skill mastery assessment vector is displayed graphically, using radar charts or bar charts to show the distribution of learners' strengths and weaknesses in each skill point; For skill nodes whose mastery score is less than a preset threshold, the system automatically traces back their dependency path in the professional skills knowledge graph to locate the prerequisite basic skills that they have weak mastery of. By combining learners' cognitive style feature vectors, the system generates personalized learning suggestions. For visual learners, it recommends instructional videos and interactive diagrams related to their weak skills, while for auditory learners, it recommends relevant audio explanations and online discussion groups.

Citation Information

Patent Citations

  • Online education learning path recommendation system for environmental art design major

    CN119721510A

  • Personalized learning path planning system and method

    CN121169644A