Personalized learning resource recommendation method based on subject knowledge graph and graph neural network

Abstract

This invention relates to a personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks, belonging to the field of educational content recommendation. The method includes: extracting key entities and relationships between them from textbooks to construct a subject knowledge graph; establishing a preliminary learning record table based on students' behavioral data during the learning process; integrating and predicting the distribution of students' mastery of each knowledge point in the learning record table through a knowledge tracking model to construct a student profile; embedding knowledge points and their relationships from the subject knowledge graph into the student profile and updating the knowledge embedding representation in the student profile through a graph neural network; constructing a personalized recommendation model and using a hybrid recommendation method of collaborative filtering and content recommendation to push learning resources that match the student's current learning status. This invention can accurately capture students' personalized needs in multiple dimensions, effectively improving the accuracy, generalization, and interpretability of recommended content.

Description

Technical Field

[0001] This invention belongs to the field of educational content recommendation and relates to a personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks. Background Technology

[0002] Personalized recommendation methods based on subject knowledge graphs have been widely applied in fields such as educational content recommendation, and are particularly suitable for scenarios such as personalized learning systems and intelligent education platforms. Traditional recommendation methods are usually based on students' historical behavioral data, lacking a comprehensive understanding of students' knowledge structures and failing to capture students' deeper learning needs and knowledge mastery, resulting in insufficient accuracy and relevance of recommended content. Furthermore, existing methods have limited understanding of the relationships between knowledge points, and the recommendation results lack contextual information support, making it difficult for the system to provide sufficient generalization and interpretability in multi-knowledge-point environments.

[0003] To address the above issues, knowledge enhancement strategies based on knowledge graphs have gradually become an effective means to improve the performance of personalized recommendation systems. Knowledge enhancement strategies can not only introduce rich semantic information between knowledge points into student profiles, but also propagate the knowledge graph through graph neural networks at multiple levels, thereby enhancing the system's understanding of students' knowledge mastery and improving the context awareness level and the diversity and accuracy of recommended content. Existing knowledge graph-based personalized recommendation methods mainly have two limitations: firstly, some methods suffer from insufficient information fusion in the integration of knowledge graph embedding and student profiles; secondly, the real-time performance and flexibility of the system in dynamically updating student profiles and recommendation results still need improvement. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks. The knowledge-enhanced subject knowledge graph-based personalized recommendation method combines multi-knowledge point fusion, knowledge enhancement strategies and student profiles to improve the accuracy, generalization and interpretability of the system, and is suitable for deep personalized recommendation in the field of intelligent education.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks includes the following steps:

[0007] S1. Extract key entities from the textbook and the relationships between them to construct a subject knowledge graph;

[0008] S2. Based on students' behavioral data during the learning process, establish a preliminary learning record form;

[0009] S3. By integrating and predicting the distribution of students' mastery of each knowledge point in the learning record table through the knowledge tracking model, a student profile is constructed.

[0010] S4. Embed the knowledge points and their relationships in the subject knowledge graph into the student profile, and update the knowledge embedding representation in the student profile through a graph neural network;

[0011] S5. Construct a personalized recommendation model, and use a hybrid recommendation method that combines collaborative filtering recommendation based on group features and content recommendation based on semantic features to push learning resources that match the student's current learning status.

[0012] Furthermore, in step S1, key entities and their relationships in the textbook are extracted through manual and semi-automatic extraction methods. The entities and relationships are then stored in a structured manner in the form of entity-relationship-entity or attribute to construct a subject knowledge graph. The subject knowledge graph contains a networked structure of each knowledge point and its related relationships. The Neo4j graph database is used to store the subject knowledge graph.

[0013] Furthermore, in step S2, behavioral data of students during the learning process is collected. The behavioral data includes the exercises completed by the students, the knowledge points related to the exercises, the difficulty of the exercises, and the answer results.

[0014] The student's behavioral data is preprocessed and normalized to generate a preliminary learning record table. The preprocessing includes standardization and cleansing operations to remove duplicate records and missing values.

[0015] The learning records in the preliminary learning record table are classified by feature, and the learning records are divided into different types of data according to the frequency of answering questions, the difficulty of answering questions, and the degree of mastery of knowledge points.

[0016] Furthermore, in step S3, the knowledge tracking model includes an embedding layer, a feature concatenation layer, a fully connected layer, a deep neural network layer, and an output layer, which includes the following steps:

[0017] S31. Encode knowledge points into vector representations through an embedding layer:

[0018] e s =Eebedding s (s)

[0019] e q =Eebedding q (q)

[0020] e k =Eebedding k (k)

[0021] ea =Eebedding a (a)

[0022] In the formula, e s e q e k and e a Let be the embedding vectors for student, question, knowledge point, and answer, respectively, and let s, q, k, and a be the data information corresponding to student, question, knowledge point, and answer, respectively. (Eebedding) s (), Eebedding q (), Eebedding k (), Eebedding a () are the embedding mapping functions corresponding to students, questions, knowledge points, and answers, respectively;

[0023] S32. The embedded vectors are concatenated into a single feature vector x through the feature concatenation layer:

[0024] x = [e s e q e k e a ]

[0025] S33. Pass x as input to the fully connected layer. The fully connected layer integrates knowledge point embeddings and student features, fusing multiple knowledge point features into one to generate a feature vector suitable for student knowledge state analysis. Input the fused feature vector into the deep neural network to predict the student's mastery probability on each knowledge point.

[0026] KLinear(x) = W base ·σ(x)+W spline ·B(x)

[0027]

[0028] In the formula, KLinear(x) is the linear computation result of the fully connected layer on the input x, that is, the output value of the fully connected layer, W base W represents the base weights, where σ(x) is the activation function of the base weights. spline B(x) is the spline weight, and B(x) is the basis function of spline interpolation; k and s are the current node index and spline order, respectively, of the grid. k Let represent the value of the k-th grid node used for spline interpolation in a fully connected layer. k+s+1 This represents the (k+s+1)th grid node, used to calculate the distance between the current node and the next node. k+s+2 This represents the (k+s+2)th grid node, used for higher index values ​​needed in interpolation calculations. k+s-2This represents the (k+s-2)th grid node, used to calculate earlier node values ​​and construct the polynomial basis during interpolation. k+1 This represents the (k+1)th grid node, used in the calculation of the interpolation basis function to determine the boundary of the interpolation range;

[0029] S34. The output layer uses the Sigmoid activation function to generate the predicted probability y of students' mastery of knowledge points:

[0030] y=σ(KLinear3(KLinear l (KLinear1(x))))

[0031] Based on the distribution of students' knowledge mastery, a student profile reflecting their learning status is obtained, including the mastery level and progress of each knowledge point.

[0032] Furthermore, step S4 includes the following steps:

[0033] S41. Using graph embedding, vectorize each knowledge point entity in the knowledge graph: vectorize the knowledge point k and the relationship r between the knowledge point and other knowledge points. ij The language model processes the input text description, encoding it into an embedded representation:

[0034] e k =BERT(T) k )

[0035]

[0036] In the formula, T k It is a textual description of knowledge point k; It is a relation r ij The text description, BERT() is a pre-trained bidirectional encoder-representation converter model used to convert input text into embedding vectors;

[0037] S42. Integrate the knowledge graph embedding results into the student profile, including: based on the relational structure in the knowledge graph, using graph neural networks to propagate semantic information of knowledge points, and integrating upstream and downstream information of related knowledge points into the current student profile; concatenating the student's mastery of knowledge points with the vectors embedded in the knowledge graph to form a more complete representation of the student's knowledge state, wherein, combined with the neighborhood aggregation operation of the graph neural network, the embedded representations of knowledge points and relations are integrated into the student profile to obtain the knowledge-enhanced student profile embedding:

[0038] For each knowledge point node k, collect its neighboring nodes. The embedded information is integrated with the neighborhood information through the aggregation function f:

[0039]

[0040] In the formula, W(l) is the weight matrix of the l-th layer, σ() is the activation function, concat() is the concatenation function, and f() is the aggregation function, which aggregates the embedding information of node k and its neighboring nodes. This represents the embedding information of the t-th neighborhood node of node k in layer l. and These represent the embedding information of node k in layer l and layer l+1, respectively.

[0041] By using knowledge graph information embedded in the graph to supplement the background knowledge of student profiles, we can identify the relationships of students' mastery of knowledge points.

[0042] Furthermore, in step S5, the personalized recommendation model employs a hybrid recommendation method, combining collaborative filtering recommendation based on group features and content recommendation based on semantic features to recommend learning resources. Specifically, the comprehensive recommendation score r for knowledge points... ik The calculation method is as follows:

[0043]

[0044] In the formula, The collaborative filtering method provides a recommendation for knowledge point k for student i. The content recommendation method is the recommendation degree given to student i for knowledge point k, and α is a weight parameter used to balance the influence of the two recommendation methods;

[0045] Based on the output of the personalized recommendation model, personalized learning content is pushed to students, including: knowledge point materials, specific exercises, and learning paths.

[0046] Furthermore, in personalized recommendation models, the recommendation process of collaborative filtering based on group characteristics is as follows:

[0047] Based on the student's knowledge acquisition and interest characteristics, other knowledge points similar to their existing knowledge are recommended. The similarity is calculated using the vectorized results of the knowledge points and the student's learning records.

[0048]

[0049] In the formula, i and j represent different students; R ik R jk represents the scores given by student i and student j for knowledge point k, respectively, and n is the total number of knowledge points;

[0050] Predicting the recommendation level of student i for unlearned knowledge point k based on similarity.

[0051]

[0052] In the formula, r i r represents the average mastery level of student i. jk Let N(i) represent the degree of mastery of knowledge point k by student i, and N(i) be the set of students most similar to student i.

[0053] Furthermore, in personalized recommendation models, the recommendation process for content recommendation based on semantic features is as follows:

[0054] Semantic enhancement of knowledge graphs is achieved using vector representation and semantic matching degree. Let knowledge point k be embedded into an embedding vector v through the graph embedding model of the knowledge graph. k The embedding vector of the student profile is v. i The semantic similarity is calculated as follows:

[0055]

[0056] In the formula, D is the vector dimension, and v id The student profile embedding vector v i The d-th dimension vector in the vector, v kd The knowledge graph embedding vector v k The d-th dimension vector in;

[0057] Then, based on semantic similarity, predict the recommendation degree of student i for knowledge point k.

[0058]

[0059] In the formula, β is an adjustment parameter used to adjust the impact of content recommendations on the overall recommendation score.

[0060] The beneficial effects of this invention are as follows:

[0061] This invention generates a student's knowledge mastery profile by constructing a multi-knowledge-point fusion model and enhances knowledge using the rich semantic information of a knowledge graph, integrating knowledge points and their relationships into the student profile. Based on a semantic propagation mechanism using graph neural networks, it further enhances the system's ability to understand the context of the knowledge graph. By combining the student profile with enhanced knowledge information, this invention can accurately capture students' personalized needs across multiple dimensions, effectively improving the accuracy, generalization, and interpretability of recommended content, making it suitable for deep personalized recommendations in educational and content recommendation scenarios.

[0062] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0063] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0064] Figure 1 The flowchart shows the personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks of the present invention.

[0065] Figure 2 This is a schematic diagram illustrating the recommendation process of the personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks of the present invention. Detailed Implementation

[0066] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0067] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0068] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0069] Please see Figures 1-2 This is a personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks.

[0070] Example

[0071] This embodiment provides a detailed process for a personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks, such as... Figure 1 As shown, it first extracts key entities and relationships between key entities from the textbook to construct a subject knowledge graph; then it collects and analyzes student behavior data, constructs student profiles through a knowledge tracing model, and uses knowledge enhancement strategies to disseminate knowledge graph information; finally, it constructs a personalized recommendation model to provide personalized recommendation strategies, specifically including the following steps:

[0072] Step 1: Extract the key entities and the relationships between them from the textbook to obtain the entity-relationship-entity (or attribute) data format, and construct a knowledge graph for the subject area;

[0073] Key entities and their relationships in textbooks are extracted using a combination of manual and semi-automatic extraction techniques. These entities and relationships are then structured and stored in an entity-relationship-entity (or attribute) format, constructing a subject-specific knowledge graph. This knowledge graph contains a networked structure of knowledge points and their relationships. By processing the entities and relationships extracted from the subject textbooks, a subject-specific knowledge graph is constructed and stored using the Neo4j graph database. Personalized recommendations for students are then made based on semantic logic and association rules, utilizing the entities and relationships within this knowledge graph.

[0074] Step 2: By collecting students' behavioral data during the learning process, including information such as completed exercises, relevant knowledge points, exercise difficulty, and answer results, the data is preprocessed and normalized to generate a preliminary learning record table for students. The data is then classified and organized according to the students' behavioral characteristics to provide basic data for building student profiles in the future.

[0075] During the student's use of the learning system, data on student learning behavior is collected. For example, the system records detailed information on the various exercises completed by students during the learning process; each exercise is associated with relevant knowledge points, ensuring that each exercise is matched with a specific knowledge point; student performance during answering questions is collected, recording the student's answering behavior for each exercise, including whether the answer is correct or not (1 for a correct answer and 0 for an incorrect answer); the difficulty level of the exercises is recorded as supplementary data for evaluating students' learning abilities.

[0076] The collected behavioral data was standardized and cleaned to ensure its integrity and consistency; duplicate records and missing values ​​were removed to ensure that each record in the dataset was unique and complete. Based on the preprocessed data, a preliminary learning record table for each student was obtained, covering the student's behavioral performance on specific knowledge points, such as answering questions and learning progress. The student's learning records were then categorized by behavioral characteristics, into different types of data such as answering frequency, question difficulty, and knowledge point mastery. This step helps in the subsequent construction of student profiles.

[0077] Step 3: Based on the student's learning record sheet, construct a multi-knowledge point fusion model, integrate and calculate all the knowledge points involved in the student, use deep neural networks to predict the student's mastery probability on each knowledge point, generate an accurate distribution of the student's knowledge mastery level, thereby constructing a personalized profile of the student and comprehensively reflecting the student's learning status in different knowledge areas.

[0078] Following the learning record table obtained in step two, a knowledge tracing model is designed to integrate student data across multiple knowledge points. An embedding layer is used to encode knowledge points into vector representations, providing consistent data for subsequent deep learning network input.

[0079] e s =Eebedding s (s)

[0080] e q =Eebedding q (q)

[0081] e k =Eebedding k (k)

[0082] e a =Eebedding a (a)

[0083] In the formula, e s e q e k and e a Let be the embedding vectors for student, question, knowledge point, and answer, respectively, and let s, q, k, and a be the data information corresponding to student, question, knowledge point, and answer, respectively. (Eebedding) s (), Eebedding q (), Eebedding k (), Eebedding a () are the embedding mapping functions corresponding to students, questions, knowledge points, and answers, respectively.

[0084] By concatenating features, the embedding vectors are combined into a single feature vector x:

[0085] x = [e s e q e k e a ]

[0086] x is passed as input to a fully connected layer, which integrates knowledge point embeddings and student features to fuse multiple knowledge point features into one, generating a feature vector suitable for analyzing students' knowledge status. The fused feature vector is then input into a deep neural network to predict the probability of students mastering each knowledge point.

[0087] KLinear(x) = W base ·σ(x)+W spline ·B(x)

[0088]

[0089] In the formula, KLinear(x) is the linear computation result of the fully connected layer on the input x, that is, the output value of the fully connected layer, W base W represents the base weights, where σ(x) is the activation function of the base weights. spline B(x) is the spline weight, and B(x) is the basis function of spline interpolation; k and s are the current node index and spline order, respectively, of the grid. k Let represent the value of the k-th grid node used for spline interpolation in a fully connected layer. k+s+1 This represents the (k+s+1)th grid node, used to calculate the distance between the current node and the next node. k+s+2 This represents the (k+s+2)th grid node, used for higher index values ​​needed in interpolation calculations. k+s-2 This represents the (k+s-2)th grid node, used to calculate earlier node values ​​and construct the polynomial basis during interpolation. k+1 This represents the (k+1)th grid node, used in the calculation of the interpolation basis function to determine the boundary of the interpolation range.

[0090] The model's output is activated by a sigmoid function to generate a predicted probability y of students' mastery of the knowledge points:

[0091] y=σ(KLinear3(KLinear2(KLinear1(x))))

[0092] Based on the distribution of students' knowledge mastery, a learning profile reflecting their status is generated, including the level of mastery and progress of each knowledge point. This profile provides comprehensive and detailed student learning information support for subsequent knowledge enhancement and personalized recommendations.

[0093] Step 4: Introduce semantic information from the knowledge graph, embedding knowledge points and their relationships into the student profile. Use graph embedding methods to transform entities and relationships in the knowledge graph into vector representations. Combine this with the student profile to enrich its background information and enhance the model's understanding of the student's knowledge structure. Furthermore, apply graph neural networks to the knowledge graph to achieve multi-layered propagation of knowledge graph information. By aggregating neighborhood information of knowledge points and dynamically updating the knowledge embedding representation in the student profile, further enhance the deep knowledge representation capability of the student profile.

[0094] The graph embedding method is used to vectorize each knowledge point entity in the knowledge graph in order to preserve its semantic features in the knowledge graph; the relations in the knowledge graph are transformed into vectors so that the relations between knowledge points can be represented in the embedding space, which helps the model understand the association between knowledge points.

[0095] Assume knowledge point k and its relationship r with other knowledge points. ij This can be handled using a language model. A language model encodes the input text description into an embedded representation:

[0096] e k =BERT(T) k )

[0097]

[0098] In the formula, T k It is a textual description of knowledge point k; It is a relation r ij The text description is provided by BERT(), a pre-trained bidirectional encoder-representation converter model used to convert input text into embedding vectors.

[0099] By integrating knowledge graph embedding results into student profiles, the knowledge representation capabilities of the profiles are enhanced. Based on the relational structure in the knowledge graph, graph neural networks are used to propagate semantic information of knowledge points, integrating upstream and downstream information of related knowledge points into the current student profile to enhance the understanding of the knowledge mastery structure. The student's mastery of knowledge points is then concatenated with the vectors embedded from the knowledge graph to form a more complete representation of the student's knowledge state.

[0100] When incorporating the embedded representations of knowledge points and relationships into student profiles, we can further combine this with the neighborhood aggregation operation of graph neural networks to obtain a knowledge-enhanced student profile embedding. For each knowledge point node k, we collect its neighboring nodes. The embedded information is then integrated using the aggregation function f to gather this neighborhood information:

[0101]

[0102] In the formula, W(l) is the weight matrix of the l-th layer, σ() is the activation function, concat() is the concatenation function, and f() is the aggregation function, which aggregates the embedding information of node k and its neighboring nodes. This represents the embedding information of the t-th neighborhood node of node k in layer l. and These represent the embedding information of node k in layer l and layer l+1, respectively.

[0103] By supplementing the background knowledge of student profiles with knowledge graph information embedded in the graph, the model can identify the mastery relationships of students on knowledge points. For example, through precursor relationships, it can be determined that a student's mastery of knowledge point A helps them master knowledge point B, enhancing the semantic background and understanding of related knowledge in the profile. This knowledge-enhanced student profile provides a richer, more personalized knowledge structure for subsequent recommendation models, improving the system's understanding and utilization of students' knowledge structures during the recommendation process.

[0104] The student profile at this point not only reflects their current knowledge mastery but also includes contextual information about their potential knowledge needs. This enhances the recommendation model's ability to understand the implicit relationships between students' interests, making personalized recommendations more accurate and effective.

[0105] Step 5: Combine student profiles containing knowledge-enhanced information with student needs to construct a personalized recommendation model. This model employs a hybrid recommendation approach, integrating collaborative filtering based on group characteristics and content recommendation based on semantic features. Through collaborative filtering, the system can analyze student groups with similar learning characteristics and uncover knowledge points that students may be interested in; the content-based recommendation method utilizes semantic information from knowledge graphs to help the system understand students' specific knowledge needs. Personalized learning resources, including personalized knowledge points, exercises, and learning paths, are recommended using these methods, pushing learning resources that match the student's current learning status and helping them develop a suitable learning path, thereby more effectively improving learning outcomes.

[0106] By analyzing student profiles and recording characteristics such as learning behaviors, knowledge mastery levels, and knowledge graph relationships, this information is matched with students' current learning goals and needs. Student needs may include mastery requirements for specific knowledge points, difficulty preferences, etc. The model incorporates multidimensional student characteristics, including their knowledge mastery, learning behavior characteristics, knowledge point preferences, and learning paths. These characteristics help to comprehensively understand student needs. Based on student profiles, content highly relevant to students' current knowledge status is selected through the correlation between knowledge points in the knowledge graph. For example, by analyzing the adjacency relationships of related knowledge points in the knowledge graph, important and relevant knowledge points that students have not yet mastered are included in the recommendation scope.

[0107] This method recommends other knowledge points similar to students' existing knowledge based on their knowledge mastery and interest characteristics. It uses the vectorized results of previously generated knowledge points and students' learning records to calculate similarity (e.g., cosine similarity) for recommendation.

[0108]

[0109] In the formula, i and j represent different students; R ik R jk represents student i's score on knowledge point k; student j's score on knowledge point k; and n is the total number of knowledge points.

[0110] Similarity can be used to predict the recommendation level of student i for knowledge point k that has not been learned.

[0111]

[0112] In the formula, r i r represents the average mastery level of student i. jk Let N(i) represent the degree of mastery of knowledge point k by student i, and N(i) be the set of students most similar to student i.

[0113] In the content recommendation section, vector representation and semantic matching degree are used to enhance the semantics of the knowledge graph. Assume that knowledge point k obtains an embedding vector v through the graph embedding model of the knowledge graph. k The embedding vector of the student profile is v. i The semantic similarity is calculated as follows:

[0114]

[0115] In the formula, D is the vector dimension, and v id The student profile embedding vector v i The d-th dimension vector in the vector, v kd The knowledge graph embedding vector v k The d-th dimension vector in;

[0116] Then, based on semantic similarity, predict the recommendation degree of student i for knowledge point k.

[0117]

[0118] In the formula, β is an adjustment parameter used to adjust the impact of content recommendations on the overall recommendation score.

[0119] Combining the results of collaborative filtering and content recommendation, we can obtain the overall recommendation score r for each knowledge point. ik :

[0120]

[0121] In the formula, α is a weighting parameter used to balance the influence of the two recommendation methods.

[0122] Based on the combined results of student profiles and knowledge point correlation analysis, a personalized recommendation model is constructed. This model inputs the knowledge point features and multi-dimensional needs from the student profiles into the recommendation algorithm, outputting a set of learning content that best matches the student's current learning stage and needs. Finally, based on the output of the personalized recommendation model, personalized learning content, including knowledge point materials, specific exercises, and learning paths, is pushed to students, ensuring that the recommended content not only matches their knowledge level but also satisfies their learning interests and needs.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks, characterized in that: The method includes the following steps: S1. Extract key entities from the textbook and the relationships between them to construct a subject knowledge graph; S2. Based on students' behavioral data during the learning process, establish a preliminary learning record form; S3. By integrating and predicting the distribution of students' mastery of each knowledge point in the learning record table through the knowledge tracking model, a student profile is constructed. S4. Embed the knowledge points and their relationships in the subject knowledge graph into the student profile, and update the knowledge embedding representation in the student profile through a graph neural network; S5. Construct a personalized recommendation model and use a hybrid recommendation method that combines collaborative filtering recommendation based on group characteristics and content recommendation based on semantic features to push learning resources that match the student's current learning status. In step S3, the knowledge tracing model includes an embedding layer, a feature concatenation layer, a fully connected layer, a deep neural network layer, and an output layer, which includes the following steps: S31. Encode knowledge points into vector representations through an embedding layer: In the formula, , , and These are the embedding vectors for the student, the question, the knowledge point, and the answer, respectively. The data information includes students, questions, knowledge points, and answers. These are the embedding mapping functions corresponding to students, questions, knowledge points, and answers, respectively. S32. The embedded vectors are concatenated into a single feature vector through a feature concatenation layer. : S33, will The features are passed as input to a fully connected layer, which integrates knowledge point features with student features, fusing multiple knowledge point features into a single feature vector suitable for analyzing student knowledge status. This fused feature vector is then input into a deep neural network layer to predict the student's mastery probability for each knowledge point. In the formula, For fully connected layers to input The linear calculation result, i.e., the output value of the fully connected layer, It is the base weight. It is the activation function of the base weights; It is the spline weight. These are the basis functions of spline interpolation; These are the current node index and the order of the spline interpolation, respectively. This represents the value of the m-th grid node used for spline interpolation in a fully connected layer. This represents the (m+s+1)th grid node, used to calculate the distance between the current node and the next node. This represents the (m+s+2)th grid node, used for higher index values ​​required in interpolation calculations. This represents the (m+s−2)th grid node, used to calculate earlier node values ​​and construct the polynomial basis during interpolation. This represents the (m+1)th grid node, used in the calculation of the interpolation basis function to determine the boundary of the interpolation range; S34. The output layer uses the Sigmoid activation function to generate the predicted probability of students' mastery of knowledge points. : Based on the distribution of students' knowledge mastery, a student profile reflecting their learning status is obtained, including the mastery level and progress of each knowledge point; Step S4 includes the following steps: S41. Using graph embedding, vectorize each knowledge point entity in the knowledge graph: vectorize the knowledge points... And the relationship between this knowledge point and other knowledge points. The language model processes the input text description, encoding it into an embedded representation: In the formula, It is a knowledge point The text description; It is a relationship The text description, This is a pre-trained bidirectional encoder-representer model used to convert input text into embedding vectors; S42. Integrate the knowledge graph embedding results into the student profile, including: based on the relational structure in the knowledge graph, using graph neural networks to propagate semantic information of knowledge points, and integrating upstream and downstream information of related knowledge points into the current student profile; concatenating the student's mastery of knowledge points with the vectors embedded in the knowledge graph to form a more complete representation of the student's knowledge state, wherein, combined with the neighborhood aggregation operation of the graph neural network, the embedded representations of knowledge points and relations are integrated into the student profile to obtain the knowledge-enhanced student profile embedding: For each knowledge point node Collect its neighboring nodes Embedded information, through aggregation functions Integrate neighborhood information: In the formula, It is The weight matrix of the layer, It is an activation function. For concatenation functions, This is an aggregation function, its function is to aggregate nodes. The embedded information of its neighboring nodes, For the first Nodes in the layer The Embedded information of each domain node and Representing nodes respectively exist Layer and first Layer embedding information; By supplementing the background knowledge of student profiles with knowledge graph information embedded in the graph, the mastery relationship of students on knowledge points can be identified. In step S5, the personalized recommendation model employs a hybrid recommendation method, combining collaborative filtering recommendation based on group features and content recommendation based on semantic features to recommend learning resources. Specifically, the overall recommendation score for knowledge points... The calculation method is as follows: In the formula, For collaborative filtering methods targeting students The given knowledge points Recommendation level , Content recommendation methods are targeted at students The given knowledge points Recommendation level These are weighting parameters used to balance the influence of the two recommendation methods; Based on the output of the personalized recommendation model, personalized learning content is pushed to students, including: knowledge point materials, specific exercises, and learning paths.

2. The personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks according to claim 1, characterized in that: In step S1, key entities and their relationships in the textbook are extracted through manual and semi-automatic extraction methods. The entities and relationships are then stored in a structured manner in the form of entity-relationship-entity or attribute to construct a subject knowledge graph. The subject knowledge graph contains a networked structure of each knowledge point and its related relationships. The Neo4j graph database is used to store the subject knowledge graph.

3. The personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks according to claim 2, characterized in that: In step S2, behavioral data of students during the learning process is collected. The behavioral data includes the exercises completed by the students, the knowledge points related to the exercises, the difficulty of the exercises, and the answer results. The student's behavioral data is preprocessed and normalized to generate a preliminary learning record table. The preprocessing includes standardization and cleansing operations to remove duplicate records and missing values. The learning records in the preliminary learning record table are classified by feature, and the learning records are divided into different types of data according to the frequency of answering questions, the difficulty of answering questions, and the degree of mastery of knowledge points.

4. The personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks according to claim 1, characterized in that: In personalized recommendation models, the recommendation process of collaborative filtering based on group features is as follows: Based on the student's knowledge acquisition and interest characteristics, other knowledge points similar to their existing knowledge are recommended. The similarity is calculated using the vectorized results of the knowledge points and the student's learning records. In the formula, , They represent different students; , Representing students ,student Knowledge points The rating, This represents the total number of knowledge points. Predicting students based on similarity For knowledge points not yet learned Recommendation level : In the formula, For students Average mastery level For students Knowledge points The degree of mastery, To be with students The set of most similar students.

5. The personalized learning resource recommendation method based on subject knowledge graphs and graph neural networks according to claim 4, characterized in that: In personalized recommendation models, the recommendation process for content recommendation based on semantic features is as follows: Knowledge graph semantic enhancement is achieved using vector representation and semantic matching degree, assuming knowledge points... Embedding vectors are obtained through the graph embedding model of knowledge graphs. The embedding vector of the student profile is The semantic similarity is calculated as follows: In the formula, For vector dimensions, Represents the embedding vector of student profiles The d-th dimension vector in Represents the embedding vector of a knowledge graph The d-th dimension vector in; Then predict students based on semantic similarity Knowledge points Recommendation level : In the formula, It is an adjustment parameter used to adjust the impact of content recommendations on the overall recommendation score.

Images