A real-time tracking teaching scoring method based on graph convolutional neural network

Abstract

A real-time tracking teaching scoring method based on graph convolutional neural networks, involving graph convolutional neural networks and the field of educational technology, solves the problems of low efficiency of manual comments and lack of dynamic tracking of static evaluations in teaching evaluation. In the method of the present invention, in terms of the depth of the evaluation dimension, it is different from the traditional coarse-grained analysis based on grade fluctuation signals or weight adjustment. By aligning the answers with the standard knowledge graph, it realizes the generation of interpretable difference subgraphs based on the GCN attention mechanism; in terms of technological integration innovation, through the structured semantic expression of the graph and the dynamic feature propagation ability of GCN, it fundamentally solves the semantic loss problem caused by the reliance on shallow text similarity or static weights in existing methods, and enables the scoring model to have the ability to adapt to the evolution of the knowledge system. This new paradigm of "structured representation-graph neural network reasoning-visual interpretation" has significantly improved the accuracy of teaching evaluation and the value of educational guidance.

Description

Technical Field

[0001] The present invention relates to the fields of graph convolutional neural networks and educational technology, and in particular to a real-time tracking teaching and scoring method based on a graph convolutional neural network. Background Art

[0002] Traditional student response evaluation relies primarily on manual review by teachers, which is not only time-consuming and labor-intensive, but also susceptible to subjective factors, making it difficult to implement large-scale, standardized evaluations. Furthermore, existing methods often focus on single-stage, static assessments and lack dynamic tracking of student learning progress.

[0003] At present, traditional federated learning parameter aggregation and dynamic weighted scoring methods have defects such as insufficient semantic modeling capabilities, limited adaptability and poor interpretability. Its linear calculations at the numerical level are difficult to capture deep semantic relationships, static weight rules cannot adapt to the evolution of the knowledge system, and the scoring process lacks transparency; coarse-grained analysis methods based on score fluctuation signals or weight adjustments are limited by their coarse analysis granularity, reliance on static rules and lack of evolutionary capabilities. They can only focus on macro-score changes but cannot deeply explore knowledge deficiencies.

[0004] Therefore, it is necessary to propose a real-time tracking teaching scoring method that can effectively capture the semantic relationships and structural information in the text and automatically generate evaluation scores in real time to reduce the burden on teachers and improve the scientificity and efficiency of the evaluation. Summary of the Invention

[0005] In order to solve the problems of low efficiency of manual comments and lack of dynamic tracking of static evaluation in teaching evaluation, the present invention provides a real-time tracking teaching scoring method based on graph convolutional neural network.

[0006] A real-time tracking teaching and scoring method based on graph convolutional neural network is implemented by the following steps:

[0007] Step 1: Collect student answer texts at different stages, use a universal large language model to construct a student answer knowledge graph for the student answer texts; generate a student answer node sequence based on the student answer knowledge graph;

[0008] Step 2: Use the GCN network to calculate the features of the current student answer node in the student answer node sequence to obtain an embedded representation of the current node feature; and execute step 3;

[0009] Step 3: Select the next node and determine whether the next node exists in the student's answer node sequence. If so, return to step 2; otherwise, execute step 4.

[0010] Step 4: averaging the feature embedding representations of the student answer node sequence;

[0011] Step 5: Select the standard answer text as the evaluation benchmark, use the general large language model to construct a standard answer knowledge graph for the standard answer text; generate a standard answer node sequence based on the standard answer knowledge graph;

[0012] Step 6: Calculate the features of the current node in the standard answer node sequence to obtain an embedded representation of the features;

[0013] Step 7: Select the next node and determine whether the next node exists in the standard answer node sequence. If yes, return to step 6; otherwise, go to step 8.

[0014] Step 8. Average the feature embedding representation of the standard answer node sequence:

[0015] Step 9. Calculate the similarity between the average value of the feature embedding representation of the student answer node sequence obtained in step 4 and the average value of the feature embedding representation of the standard answer node sequence obtained in step 8, so as to achieve real-time tracking of the student's learning level.

[0016] Furthermore, in step one, a general large language model is used to extract triples from the student answer text, and the extracted triples are converted into a graph structure. The head entity and tail entity in the triple are used as nodes, and the relationship in the triple is used as an edge. The nodes and edges are combined into a graph structure to construct a knowledge graph of student answers.

[0017] Furthermore, in step five, a general large language model is used to extract triples from the standard answer text, and the extracted triples are converted into a graph structure. The head entity and tail entity in the triple are used as nodes, and the relationships in the triple are used as edges. The nodes and edges are combined into a graph structure to construct a standard answer knowledge graph.

[0018] Furthermore, in step nine, the cosine similarity is used to calculate the similarity between the average value of the feature embedding representation of the student answer node sequence and the average value of the feature embedding representation of the standard answer node sequence; it can be expressed as follows: ;

[0019] Where, The feature embedding representation of the student answer node sequence is the average value, is the feature embedding representation average of the standard answer node sequence, Represents the magnitude of a vector.

[0020] Furthermore, step five also includes normalizing the similarities of all stages to between 0 and 1 to obtain the normalized scores, which are expressed as follows: ;

[0021] Where, is the similarity of the students at the current stage, and is the maximum and minimum similarity of all stages; the normalized score As the student's personal learning level at that stage.

[0022] Beneficial effects of the present invention:

[0023] The real-time tracking teaching scoring method of the present invention can effectively capture the semantic relationships and structural information in the text, and automatically generate evaluation scores in real time to reduce the burden on teachers and improve the scientificity and efficiency of evaluation. Compared with existing technical solutions that use federated learning parameter aggregation or dynamic weighted scoring, the present invention has made a breakthrough in data modeling by introducing a collaborative modeling mechanism of knowledge graph triple structure (entity-relationship-attribute) and graph convolutional neural network (GCN). It dynamically captures the complex logical associations in teaching through semantic graph structure, rather than just performing parameter fusion or linear weighting at the numerical level. Finally, it is normalized to a score between 0 and 1, automatically evaluating students' learning progress, providing teachers with efficient and quantitative evaluation support, and reducing the burden on teachers' review.

[0024] In terms of depth of assessment dimensions, unlike traditional coarse-grained analysis based on performance fluctuation signals or weight adjustments, this method aligns answers with a standard knowledge graph, enabling the generation of interpretable difference subgraphs based on the GCN attention mechanism. Furthermore, in terms of technological integration and innovation, by combining the graph's structured semantic representation with the GCN's dynamic feature propagation capabilities, it fundamentally addresses the semantic loss problem caused by existing methods' reliance on shallow text similarity or static weights, enabling the scoring model to adapt to the evolution of the knowledge system. This novel paradigm of "structured representation - graph neural network reasoning - visual interpretation" significantly improves the accuracy of teaching evaluation and the value of educational guidance.

[0025] The real-time tracking teaching and scoring method described in this paper transforms student answers and standard answers into a structured knowledge graph through precise triple modeling. It uses entity-relationship-attribute triples to achieve fine-grained semantic expression. Based on GCN semantic enhancement technology, it aggregates adjacent node features through a graph convolutional neural network and dynamically calculates subgraph similarities between answers, significantly improving the ability to capture complex logical relationships and long-tail concepts. These two methods work together to optimize the entire process, from data representation to algorithm calculation to output, balancing scoring accuracy, efficiency, and teaching value. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a real-time tracking teaching and scoring method based on graph convolutional neural network shown in the present invention;

[0027] Figure 2This is a diagram showing the effect of using the real-time tracking teaching scoring method based on graph convolutional neural network described in the present invention. DETAILED DESCRIPTION

[0028] Combine Figure 1 and Figure 2 This embodiment describes a real-time tracking teaching and scoring method based on a graph convolutional neural network. The method is implemented by the following steps:

[0029] Step 1: Publish the topic learning on the system and collect students' response texts at different stages (such as time point 1, time point 2, etc.);

[0030] Step 2: Extract triplets from student response text using a general large language model , the triplet is in the form of (head entity, relation, tail entity);

[0031] Step 3: Extract the triples Convert it into a graph structure, with the head and tail entities in the triple as nodes, and the relationships in the triple as edges. The nodes and edges are combined into a graph structure to represent the semantics of the text. Construct a knowledge graph of student answers ;

[0032] Step 4: Extract student answer knowledge graph Entity nodes in the , and use this to generate student answer node sequence , ,in Answer node for the i-th student, The number of nodes for student responses;

[0033] Step 5: Node sequence for student responses The current student answer node in , use the graph convolutional network (GCN network) to perform feature calculation and obtain the current node embedding representation , proceed to step 6;

[0034] Step 6: Select the next node , if the student answers the node sequence There is a next node in , return to step 5, otherwise go to step 7;

[0035] Step 7: Have students answer the node sequence The embedding representation is the average : ;

[0036] Step 8: Teachers or experts provide standard answer texts to the questions as an evaluation benchmark;

[0037] Step 9: Use a general large language model to extract triplets from the ground truth text , the form is (head entity, relationship, tail entity);

[0038] Step 10: Extract the triples Convert it into a graph structure, with the head and tail entities in the triple as nodes, and the relationships in the triple as edges. Combine the nodes and edges into a graph structure to represent the semantics of the text. Construct a knowledge graph of standard answers ;

[0039] Step 11: Based on the knowledge graph of standard answers Generate standard answer node sequence , ,in is the jth standard node, is the number of standard nodes;

[0040] Step 12: For the standard answer node sequence The current node in , use the GCN network to perform feature calculation and obtain the embedded representation , go to step 13;

[0041] Step 13: Select the next node , if the standard answer node sequence There are nodes in , then return to step 12, otherwise, go to step 14;

[0042] Step 14: Standard answer node sequence The embedding representation is the average : ;

[0043] Step 15: Use cosine similarity to calculate the average of the embedding representations of students’ answers at each stage and the standard answer embedding representation to find the average The similarity of the values ​​is calculated for each stage of the students’ answers, thus enabling real-time tracking of students’ learning levels. ;

[0044] in, Represents the magnitude of a vector.

[0045] Step 16: Normalize the similarities of all stages to between 0 and 1;

[0046] ;

[0047] in, is the similarity of the students at the current stage, and is the maximum and minimum similarity of all stages. As the student's personal learning level at that stage.

[0048] like Figure 2 As shown, Figure 2 This is a diagram showing the effects achieved using the method described in this embodiment. The blue box in the figure shows the comparison score of each student's answer against the standard answer. The left side of the system interface shows the student's answer content. Once the student completes and submits the answer, the right area automatically generates a knowledge graph based on the answer content. This design enables real-time comparison and analysis of student answers with the standard answer, and intuitively presents knowledge mastery through a visual knowledge graph.

[0049] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0050] The above-described embodiments merely illustrate several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A real-time tracking teaching and scoring method based on graph convolutional neural network, characterized by: The method is implemented by the following steps: Step 1: Collect student answer texts at different stages, use a universal large language model to construct a student answer knowledge graph for the student answer texts; generate a student answer node sequence based on the student answer knowledge graph; Step 2: Use the GCN network to calculate the features of the current student answer node in the student answer node sequence to obtain the embedded representation of the current node feature; Execute step 3; Step 3: Select the next node and determine whether the next node exists in the student's answer node sequence. If yes, return to step 2. Otherwise, go to step 4; Step 4: averaging the feature embedding representations of the student answer node sequence; Step 5: Select the standard answer text as the evaluation benchmark, use the general large language model to construct a standard answer knowledge graph for the standard answer text; generate a standard answer node sequence based on the standard answer knowledge graph; Step 6: Calculate the features of the current node in the standard answer node sequence to obtain an embedded representation of the features; Step 7: Select the next node and determine whether the next node exists in the standard answer node sequence. If yes, return to step 6. Otherwise, go to step eight; Step 8: Calculate the average of the feature embedding representations of the standard answer node sequence; Step 9. Calculate the similarity between the average value of the feature embedding representation of the student answer node sequence obtained in step 4 and the average value of the feature embedding representation of the standard answer node sequence obtained in step 8, so as to achieve real-time tracking of the student's learning level.

2. The real-time tracking teaching and scoring method based on graph convolutional neural network according to claim 1 is characterized by: In step one, a general large language model is used to extract triples from the student answer text, and the extracted triples are converted into a graph structure. The head entity and tail entity in the triple are used as nodes, and the relationships in the triple are used as edges. The nodes and edges are combined into a graph structure to construct a knowledge graph of student answers.

3. The real-time tracking teaching and scoring method based on graph convolutional neural network according to claim 1 is characterized in that: In step five, a general large language model is used to extract triples from the standard answer text, and the extracted triples are converted into a graph structure. The head entity and tail entity in the triple are used as nodes, and the relationships in the triple are used as edges. The nodes and edges are combined into a graph structure to construct a standard answer knowledge graph.

4. The real-time tracking teaching and scoring method based on graph convolutional neural network according to claim 1 is characterized in that: In step nine, cosine similarity is used to calculate the similarity between the average value of the feature embedding representation of the student answer node sequence and the average value of the feature embedding representation of the standard answer node sequence; it can be expressed as follows: ； Where, The feature embedding representation of the student answer node sequence is the average value, is the feature embedding representation average of the standard answer node sequence, Represents the magnitude of a vector.

5. The real-time tracking teaching and scoring method based on graph convolutional neural network according to claim 1 is characterized in that: Step 5 also includes normalizing the similarities of all stages to between 0 and 1 to obtain the normalized scores, which are expressed as follows: ； Where, is the similarity of the students at the current stage, and is the maximum and minimum similarity of all stages; the normalized score As the student's personal learning level at that stage.

Images