Adaptive learning resource recommendation method and system based on knowledge graph

Abstract

The application discloses a kind of based on knowledge graph's self-adapting learning resource recommendation method and system, belong to intelligent learning dynamic optimization field.Method includes to the educational information dataset obtained to build student-learning resource-knowledge concept three-layer knowledge graph;According to student-learning resource-knowledge concept three-layer knowledge graph, obtain the degree of mastery of student to knowledge concept and the learning resource node set containing difficulty parameter;According to the degree of mastery of student to knowledge concept, determine learning path sequence;Combining the learning path sequence of student, according to the difficulty parameter of learning resource, obtain the learning resource recommendation list of student.The present application considers learning risk score, priority order to thereby construct learning path sequence, ensure that learning path follows the priority order of knowledge concept dependence, and can be dynamically adjusted sequence by risk quantification mechanism, adapt to individual cognitive difference;Through multidimensional matching function, the difference of multiple student types is recommended.

Description

Technical Field

[0001] This invention relates to an adaptive learning resource recommendation method and system based on knowledge graphs, belonging to the field of intelligent learning dynamic optimization. Background Technology

[0002] In the field of dynamic optimization for intelligent learning, existing technologies mainly focus on processing time-series data and interactive behavior data to achieve personalized planning of learning paths and recommendation of learning resources. For example, patent application CN120744832A first obtains a set of learning behavior data through multi-source data fusion, then divides knowledge nodes into static attributes (such as logical relationships between knowledge points) and dynamic attributes (adjusting logical relationships based on time thresholds). Next, it calculates the association strength of knowledge nodes based on time sequence, establishes an intelligent learning knowledge graph, and uses an improved Harris Eagle optimization algorithm for dynamic path planning. By balancing global exploration and local development through chaotic fusion parameters and nonlinear control factors, it ultimately generates the optimal learning path. The learning resource matching module then generates learning resource configurations based on the path and performs resource matching through a feature library.

[0003] Although CN120744832A introduces dynamic optimization of time sequence, it still has shortcomings, as follows:

[0004] 1. The relationship modeling has a single dimension and lacks the ability to integrate multi-level heterogeneous data.

[0005] Knowledge graphs are built using time-series and interactive behavior data, but they can only handle simple static and dynamic attribute relationships (such as the time-dependent adjustment of logical relationships between knowledge points). For example, the system filters strongly associated triples by calculating the association strength of knowledge nodes, but this method cannot capture the complex multidimensional relationships between "students-learning resources-knowledge concepts." The construction of knowledge graphs is based solely on the logical connections between knowledge points, ignoring the non-linear interaction patterns in student behavior data (such as the evolution of cognitive impairment behind incorrect answers), leading to one-sided diagnostic results.

[0006] 2. The path planning algorithm is rigid and lacks dynamic weight learning and risk perception mechanisms.

[0007] Existing systems employ an improved Harris Eagle optimization algorithm for path planning. While this avoids local optima through chaotic parameters and nonlinear control factors, path generation relies on a pre-defined topological structure (such as a sequence of knowledge nodes). The algorithm cannot dynamically adjust weights based on real-time learning data; for example, path risk is controlled solely by a simple association strength decay threshold (e.g., 0.5), lacking a quantitative assessment of learning risks (such as the risk of insufficient prior knowledge). This makes path planning difficult to conform to the zone of proximal development theory and unable to adapt to the discontinuous nature of students' cognitive development.

[0008] 3. Neglecting multi-dimensional balance and interpretability.

[0009] The resource matching module of the existing system calculates the matching coefficient only based on the overlap of knowledge points and the degree of urgency. The recommendation goal focuses on knowledge coverage and time sequence adaptation, ignoring key factors such as knowledge coverage and learning path matching. For example, the matching function simply uses weights (such as weights of 0.6 and 0.4), which cannot differentiate the needs of basic learners, benchmark learners and advanced learners.

[0010] In view of this, the present invention is hereby proposed. Summary of the Invention

[0011] This invention proposes an adaptive learning resource recommendation method and system based on knowledge graphs. By constructing a three-layer knowledge graph of students, learning resources, and knowledge concepts, it effectively alleviates the problem of insufficient modeling of the heterogeneous relationship among these three layers in existing systems. Furthermore, based on the above, it considers learning risk scores and priority ranking to construct a learning path sequence that conforms to the zone of proximal development theory. Moreover, through a multi-dimensional matching degree function and combined with ZPD theory, it achieves differentiated recommendations for multiple student types.

[0012] The technical solution of this invention is:

[0013] According to a first aspect of the present invention, an adaptive learning resource recommendation method based on a knowledge graph is provided, comprising: constructing a three-layer knowledge graph of students, learning resources, and knowledge concepts from an acquired educational information dataset; obtaining, based on the three-layer knowledge graph, the students' mastery of knowledge concepts and a set of learning resource nodes including difficulty parameters; determining a learning path sequence based on the students' mastery of knowledge concepts; and obtaining a recommended list of learning resources for the students based on the students' learning path sequence and the difficulty parameters of the learning resources.

[0014] Furthermore, the step of obtaining the student's mastery of knowledge concepts and the set of learning resource nodes including difficulty parameters based on the student-learning resource-knowledge concept three-layer knowledge graph includes: obtaining weighted feature vectors, fused feature vectors, and behavioral feature vectors based on each subgraph in the student-learning resource-knowledge concept three-layer knowledge graph; wherein, the subgraphs include concept relationship graphs, learning resource-knowledge concept association graphs, and student-learning resource interaction graphs; using the weighted feature vectors, fused feature vectors, and behavioral feature vectors as inputs to the trained student-learning resource-knowledge concept three-layer GAT network, and outputting a fused knowledge concept vector; using the fused knowledge concept vector as input to the word embedding layer, and outputting a standardized feature vector; inputting the standardized feature vector to the diagnostic result output layer to obtain the student's mastery of knowledge concepts and the set of learning resource nodes including difficulty parameters.

[0015] Furthermore, determining the learning path sequence based on students' mastery of knowledge concepts includes: obtaining a set of students' knowledge weaknesses based on their mastery of knowledge concepts; constructing a directed acyclic graph (DAG) based on the dependencies between knowledge concepts in the concept relationship graph; calculating students' learning risk scores for knowledge concepts based on their set of knowledge weaknesses and the DAG; and determining the learning path sequence based on their set of knowledge weaknesses, the DAG, and the learning risk scores.

[0016] Furthermore, obtaining the set of students' knowledge weaknesses based on their mastery of knowledge concepts includes: calculating knowledge gaps for each knowledge concept in the set of knowledge concepts in the concept relationship diagram based on the students' mastery of knowledge concepts, and taking the knowledge gaps with non-zero values ​​as the first set of knowledge gaps; constructing the set of students' knowledge weaknesses by constructing the knowledge concepts corresponding to the first set of knowledge gaps.

[0017] Furthermore, determining the learning path sequence based on the student's set of knowledge weaknesses, the directed acyclic graph (DAG), and the learning risk score includes: adding all nodes in the DAG with an in-degree of 0 that belong to the set of knowledge weaknesses or are connected by a path to knowledge concepts in the set of knowledge weaknesses to a candidate node set; calculating a priority parameter for each knowledge concept in the candidate node set; using the magnitude of the priority parameter as the student's priority for learning the knowledge concept; selecting the knowledge concept with the highest current priority from the candidate node set, removing it from the candidate node set, and outputting it to the learning path sequence; removing the knowledge concept with the highest current priority from the DAG, and adding the nodes with an in-degree of 0 that are the direct successors of the knowledge concept with the highest current priority to the candidate node set; repeating this process until all knowledge concepts in the set of knowledge weaknesses have been output to the learning path sequence.

[0018] Furthermore, the step of combining the student's learning path sequence and obtaining the student's learning resource recommendation list based on the difficulty parameters of the learning resources includes: constructing a ZPD interval expression based on the student's mastery of knowledge concepts; obtaining the student's ZPD interval based on the ZPD interval expression; defining a comprehensive matching score function based on the student's learning path sequence, the ZPD interval, and the difficulty parameters of the learning resources; obtaining the comprehensive matching score between the student and the learning resources based on the comprehensive matching score function; obtaining the strategic score between the student and each learning resource based on the comprehensive matching score between the student and each learning resource; and obtaining the student's learning resource recommendation list based on the strategic score between the student and each learning resource.

[0019] Furthermore, the ZPD interval expression is:

[0020] ;

[0021] in, Students ZPD interval, Students knowledge concepts The degree of mastery; For students The personalized zone of proximal development radius, whose value is determined by the base radius parameter. With adjustment coefficient The product is determined, and the adjustment coefficient is adjusted. The learning process is dynamically adjusted based on student type to suit the learning needs of students with different cognitive levels.

[0022] .

[0023] Furthermore, the comprehensive matching degree scoring function is:

[0024] ;

[0025] in, Indicates learning resources Difficulty parameters, Represents students and difficulty parameters Corresponding learning resources The overall matching score, express and ZPD matching degree express and Knowledge coverage express and The degree of matching of the learning path; ZPD matching degree weight coefficient, For knowledge coverage weighting coefficient, Let be the learning path matching degree weight coefficient, and satisfy . .

[0026] Furthermore, the step of obtaining a strategic score for the student and each learning resource based on the overall matching score between the student and each learning resource, and obtaining a recommended list of learning resources for the student based on the strategic score, includes: defining a strategy optimization function according to the student type; applying the corresponding strategy optimization function to adjust the overall matching score between the student and each learning resource to obtain the strategic score between the student and each learning resource; sorting the strategic scores of each student and each learning resource in descending order, and selecting the top... Each learning resource is used to create a recommended list of learning resources for students; among them... express The number of learning resources selected by students of different student types.

[0027] According to a second aspect of the present invention, an adaptive learning resource recommendation system based on knowledge graphs is provided, comprising modules of any of the methods described above.

[0028] The effective effects of this invention are:

[0029] 1. This invention employs the graph attention mechanism (GAT) to dynamically learn weights for the concept relationship graph, the learning resource-knowledge concept association, and the student-learning resource interaction graph. This overcomes the limitations of linear modeling in traditional methods, improves the comprehensiveness and accuracy of diagnostic results, and provides a reliable data foundation for subsequent learning path sequence planning and learning resource recommendation.

[0030] 2. This invention considers learning risk scores and priority ranking to construct a learning path sequence, ensuring that the learning path not only follows the priority ranking of knowledge concept dependence, but also dynamically adjusts the sequence through risk quantification mechanism to adapt to individual cognitive differences.

[0031] 3. This invention uses a multi-dimensional matching function (covering ZPD matching degree, knowledge coverage, and learning path matching degree dimensions) combined with ZPD theory to achieve differentiated recommendations for multiple student types (such as basic learners, benchmark learners, and advanced learners). Attached Figure Description

[0032] Figure 1 This is a flowchart of the present invention.

[0033] Figure 2 This is a sequence diagram of the learning path for student A's 15 knowledge concepts.

[0034] Figure 3 This is a sequence diagram of the learning path for student B's 15 knowledge concepts.

[0035] Figure 4 A ZPD matching visualization of student A's abilities and the difficulty of knowledge concepts.

[0036] Figure 5 A ZPD matching visualization of student B's abilities and the difficulty of knowledge concepts. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other.

[0038] Example 1: As Figures 1-5 As shown, according to a first aspect of the present invention, an adaptive learning resource recommendation method based on a knowledge graph is provided, comprising: constructing a three-layer knowledge graph of students, learning resources, and knowledge concepts from an acquired educational information dataset; obtaining, based on the three-layer knowledge graph, the students' mastery of knowledge concepts and a set of learning resource nodes including difficulty parameters; determining a learning path sequence based on the students' mastery of knowledge concepts; and obtaining a recommended list of learning resources for the students based on the students' learning path sequence and the difficulty parameters of the learning resources.

[0039] Furthermore, the step of obtaining the student's mastery of knowledge concepts and the set of learning resource nodes including difficulty parameters based on the student-learning resource-knowledge concept three-layer knowledge graph includes: obtaining weighted feature vectors, fused feature vectors, and behavioral feature vectors based on each subgraph in the student-learning resource-knowledge concept three-layer knowledge graph; wherein, the subgraphs include concept relationship graphs, learning resource-knowledge concept association graphs, and student-learning resource interaction graphs; using the weighted feature vectors, fused feature vectors, and behavioral feature vectors as inputs to the trained student-learning resource-knowledge concept three-layer GAT network, and outputting a fused knowledge concept vector; using the fused knowledge concept vector as input to the word embedding layer, and outputting a standardized feature vector; inputting the standardized feature vector to the diagnostic result output layer to obtain the student's mastery of knowledge concepts and the set of learning resource nodes including difficulty parameters.

[0040] Furthermore, determining the learning path sequence based on students' mastery of knowledge concepts includes: obtaining a set of students' knowledge weaknesses based on their mastery of knowledge concepts; constructing a directed acyclic graph (DAG) based on the dependencies between knowledge concepts in the concept relationship graph; calculating students' learning risk scores for knowledge concepts based on their set of knowledge weaknesses and the DAG; and determining the learning path sequence based on their set of knowledge weaknesses, the DAG, and the learning risk scores.

[0041] Furthermore, obtaining the set of students' knowledge weaknesses based on their mastery of knowledge concepts includes: calculating knowledge gaps for each knowledge concept in the set of knowledge concepts in the concept relationship diagram based on the students' mastery of knowledge concepts, and taking the knowledge gaps with non-zero values ​​as the first set of knowledge gaps; constructing the set of students' knowledge weaknesses by constructing the knowledge concepts corresponding to the first set of knowledge gaps.

[0042] Furthermore, determining the learning path sequence based on the student's set of knowledge weaknesses, the directed acyclic graph (DAG), and the learning risk score includes: adding all nodes in the DAG with an in-degree of 0 that belong to the set of knowledge weaknesses or are connected by a path to knowledge concepts in the set of knowledge weaknesses to a candidate node set; calculating a priority parameter for each knowledge concept in the candidate node set; using the magnitude of the priority parameter as the student's priority for learning the knowledge concept; selecting the knowledge concept with the highest current priority from the candidate node set, removing it from the candidate node set, and outputting it to the learning path sequence; removing the knowledge concept with the highest current priority from the DAG, and adding the nodes with an in-degree of 0 that are the direct successors of the knowledge concept with the highest current priority to the candidate node set; repeating this process until all knowledge concepts in the set of knowledge weaknesses have been output to the learning path sequence.

[0043] Furthermore, the step of combining the student's learning path sequence and obtaining the student's learning resource recommendation list based on the difficulty parameters of the learning resources includes: constructing a ZPD interval expression based on the student's mastery of knowledge concepts; obtaining the student's ZPD interval based on the ZPD interval expression; defining a comprehensive matching score function based on the student's learning path sequence, the ZPD interval, and the difficulty parameters of the learning resources; obtaining the comprehensive matching score between the student and the learning resources based on the comprehensive matching score function; obtaining the strategic score between the student and each learning resource based on the comprehensive matching score between the student and each learning resource; and obtaining the student's learning resource recommendation list based on the strategic score between the student and each learning resource.

[0044] Furthermore, the ZPD interval expression is:

[0045] ;

[0046] in, Students ZPD interval, Students knowledge concepts The degree of mastery; For students The personalized zone of proximal development radius, whose value is determined by the base radius parameter. With adjustment coefficient The product is determined, and the adjustment coefficient is adjusted. The learning process is dynamically adjusted based on student type to suit the learning needs of students with different cognitive levels.

[0047] .

[0048] Furthermore, the comprehensive matching degree scoring function is:

[0049] ;

[0050] in, Indicates learning resources Difficulty parameters, Represents students and difficulty parameters Corresponding learning resources The overall matching score, express and ZPD matching degree express and Knowledge coverage express and The degree of matching of the learning path; ZPD matching degree weight coefficient, For knowledge coverage weighting coefficient, Let be the learning path matching degree weight coefficient, and satisfy . .

[0051] Furthermore, the step of obtaining a strategic score for the student and each learning resource based on the overall matching score between the student and each learning resource, and obtaining a recommended list of learning resources for the student based on the strategic score, includes: defining a strategy optimization function according to the student type; applying the corresponding strategy optimization function to adjust the overall matching score between the student and each learning resource to obtain the strategic score between the student and each learning resource; sorting the strategic scores of each student and each learning resource in descending order, and selecting the top... Each learning resource is used to create a recommended list of learning resources for students; among them... Indicates the student type.

[0052] According to a second aspect of the present invention, an adaptive learning resource recommendation system based on knowledge graphs is provided, comprising modules of any of the methods described above.

[0053] Example 2: As Figures 1-5 As shown, an adaptive learning resource recommendation method based on knowledge graphs includes:

[0054] Step 1: Construct a three-layer knowledge graph of students, learning resources, and knowledge concepts based on the acquired educational information dataset.

[0055] The construction of a three-layer knowledge graph of students, learning resources, and knowledge concepts based on the acquired educational information dataset includes:

[0056] Step 1.1: Define the knowledge graph as a directed graph structure. ;in, It represents a set, including knowledge concepts, learning resources, and students; Represents a set of relations, encoding the educational semantic relationships between them; Store knowledge facts in the form of triples, that is... .

[0057] Step 1.2: Construct a concept relationship diagram The conceptual relationship diagram Encapsulate knowledge concepts and their dependencies, and formalize them as follows: ,in, Represents a set of knowledge concepts. This represents a set of knowledge concept dependencies, categorized into prerequisite relations, remedial relations, similarity relations, and collaborative relations. These relations are generated based on conditional probability calculations of student answer sequences, ensuring that the relations originate from real data. For example, the values ​​for prerequisite, remedial, similarity, and collaborative relations range from 0 to 1, with values ​​closer to 1 indicating a stronger relationship.

[0058] Step 1.3: Construct a learning resource-knowledge concept association diagram The learning resource-knowledge concept association graph modeling represents a many-to-many mapping between learning resources and knowledge concepts, formalized as follows: ,in, For the collection of learning resources, A collection of knowledge concepts For the set of mapping relationships between learning resources and knowledge concepts, when When, it indicates learning resources Includes knowledge concepts ;when When, it indicates learning resources Does not contain knowledge concepts , , .

[0059] Step 1.4: Constructing the Student-Learning Resource Interaction Graph The student-learning resource interaction graph captures the interaction behavior between students and learning resources, and formalizes it as follows: ,in, Gathering for students For the interaction between students and learning resources; when At that time, it indicates that the student With learning resources There is interaction; when At that time, it indicates that the student With learning resources There is no interaction. Based on the student-learning resource interaction graph, a student-learning resource interaction heatmap can be output for visualization.

[0060] Step 1.5: Create a concept relationship diagram Learning Resources - Knowledge Concept Association Diagram Interactive graph of learning resources with students Integrate into a three-layer knowledge graph of students, learning resources, and knowledge concepts, and formalize it as follows: .

[0061] The three-layer knowledge graph obtained above has a hierarchical structure. The educational information dataset is processed into a JSON format output of a three-layer knowledge graph of students, learning resources, and knowledge concepts, providing a data foundation for subsequent processing.

[0062] Step 2: Using the three-layer knowledge graph of students, learning resources, and knowledge concepts as input, independently deploy a GAT network for each subgraph in the student-learning resource-knowledge concept knowledge graph to calculate the weights between entities and obtain weighted feature vectors, fused feature vectors, and behavioral feature vectors; where subgraphs include concept relationship graphs. Learning Resources - Knowledge Concept Association Diagram Interactive graph of learning resources with students .

[0063] Step 2 includes:

[0064] Step 2.1: Create a concept relationship diagram Construct the first GAT network; use the concept relationship graph as the input of the first GAT network, calculate the weights between nodes corresponding to knowledge concepts through the first GAT network, and output a weighted feature vector; for example, the weights of prerequisite relationship edges are dynamically adjusted through a normalized threshold.

[0065] Step 2.2: Create a learning resource-knowledge concept association diagram Construct a second GAT network; use the learning resource-knowledge concept association graph as the input of the second GAT network, calculate the weights of the mapping between learning resources and knowledge concepts through the second GAT network, and output the fused feature vector.

[0066] Step 2.3: Student-Learning Resource Interaction Graph Construct a third GAT network; integrate student-learning resource interaction graphs. As input to the third GAT network, the weights of student-learning resource interactions are calculated through the third GAT network, and the behavioral feature vector is output.

[0067] In this step, the first, second, and third GAT networks have the same structure (i.e., graph attention networks), and each independently processes the corresponding subgraph to generate node-level dynamic weights. This process ensures specific learning of different types of relationships and avoids the limitations of linear modeling.

[0068] Step 3: Use the weighted feature vector, fused feature vector, and behavioral feature vector as input to the trained student-learning resource-knowledge concept three-layer GAT network, and output the fused knowledge concept vector; use the fused knowledge concept vector as input to the word embedding layer, and output the standardized feature vector.

[0069] Step 3 includes:

[0070] Step 3.1: Assign weight parameters η, γ, and ω (e.g., η=0.4, γ=0.3, ω=0.3) to the concept relationship graph, the learning resource-knowledge concept association graph, and the student-learning resource interaction graph, respectively. Use the three-layer knowledge graph of student-learning resource-knowledge concept as input for training and optimization to obtain a trained three-layer GAT network of student-learning resource-knowledge concept. The fourth GAT network has the same structure as the first, second, and third GAT networks.

[0071] Step 3.2: Use the weighted feature vector, fused feature vector, and behavioral feature vector output from Step 2 as input to the trained student-learning resource-knowledge concept three-layer GAT network, and output the fused knowledge concept vector. The expression of the fused knowledge concept vector is as follows:

[0072] ;

[0073] in, The fused knowledge concept vector, This indicates the training of the fourth GAT network. This refers to all knowledge concepts in the concept relationship diagram. These are weighted feature vectors, fused feature vectors, and behavioral feature vectors, respectively.

[0074] Step 3.3: Use the fused knowledge concept vector as input to the word embedding layer and output the standardized feature vector.

[0075] Step 4: Input the standardized feature vector into the diagnostic result output layer. The diagnostic result output layer uses an activation function (such as Sigmoid) to obtain the student's mastery of the knowledge concept and a set of learning resource nodes including difficulty parameters. The mastery level ranges from 0 to 1, with the closer to 1 indicating higher mastery. For example, student A's mastery level of "functions and limits" is 0.75. The difficulty parameter is represented by a decimal between 0 and 1, with the closer to 1 indicating greater difficulty of the learning resource node.

[0076] The diagnostic results are the student's mastery of knowledge concepts and the set of learning resource nodes including difficulty parameters output in Step 4 above.

[0077] Step 5: Based on the students' mastery of knowledge concepts output in Step 4, obtain a set of students' knowledge weaknesses.

[0078] Step 5 specifically includes:

[0079] Step 5.1: Based on students' mastery of the knowledge concepts, analyze each knowledge concept in the knowledge concept set of the concept relationship diagram. Calculate knowledge gaps The knowledge gaps with non-zero values ​​are taken as the first group of knowledge gaps.

[0080] The knowledge gap Calculate using the following formula:

[0081] ;

[0082] in, The preset mastery threshold is (typically 0.6 to 0.7, and 0.65 in this embodiment of the invention). The student's understanding of knowledge concepts as output in Step 4 The knowledge gap calculation formula ensures that the knowledge gap is positive only when the knowledge level is below a threshold.

[0083] Step 5.2: Construct a set of students' knowledge weaknesses W by combining the knowledge concepts corresponding to the first set of knowledge gaps.

[0084] Step 6: Based on the dependencies between knowledge concepts in the concept relationship graph, construct a directed acyclic graph.

[0085] Step 6.1: Extract the dependencies between knowledge concepts in the concept relationship graph to form the first directed graph.

[0086] Step 6.2: Perform cycle detection on the first directed graph to construct a directed acyclic graph. Specifically, if a cycle exists during the detection process, it is resolved by adjusting dependencies or ignoring minor circular dependencies. The resulting directed acyclic graph is denoted as […]. This is to ensure the feasibility of topological sorting.

[0087] Step 7: Based on the student's set of knowledge weaknesses and the directed acyclic graph, calculate the student's learning risk score for the knowledge concepts.

[0088] Step 7.1: For each knowledge concept in the knowledge weakness set W In a directed acyclic graph, find all the knowledge concepts that have direct prerequisite relations, and denote the set as . .

[0089] Step 7.2: For each knowledge concept in the knowledge weakness set W Calculate learning risk score The formula is as follows:

[0090] ;

[0091] in, It is the student's prior knowledge concepts Learning risk score, prerequisite knowledge concepts This represents the knowledge concepts that students currently need to process. exist The corresponding knowledge concepts. The learning risk score reflects the learning of knowledge concepts due to insufficient mastery of prerequisite knowledge. The higher the value, the greater the risk.

[0092] Step 8: Determine the learning path sequence based on the student's set of knowledge weaknesses, the directed acyclic graph, and the learning risk score.

[0093] Step 8.1: Transform the directed acyclic graph All nodes with an in-degree of 0 that belong to the knowledge weakness set W or have a path connection with knowledge concepts in the knowledge weakness set W are added to the candidate node set. .

[0094] Step 8.2: For the candidate node set Each knowledge concept in Calculate priority parameters The priority parameter is used to determine the student's priority in learning knowledge concepts (a larger value indicates a higher priority).

[0095] ;

[0096] in, and Configurable weighting coefficients (e.g.) , This is used to balance the strategy of "prioritizing learning low-risk concepts" and "prioritizing learning concepts that have been partially mastered"; For students' understanding of knowledge concepts Learning risk score; For students' understanding of knowledge concepts The degree of mastery.

[0097] Step 8.3: Select the knowledge concept with the highest current priority from the candidate node set. Remove it from the candidate node set Remove from the list and output to the learning path sequence; select the knowledge concept with the highest current priority. From a directed acyclic graph Remove from the list and prioritize the highest-priority knowledge concept. Nodes whose in-degree becomes 0 in the direct successor knowledge concept nodes are added to the candidate node set. .

[0098] Step 8.4: Repeat Steps 8.2 to 8.3 until all knowledge concepts in the knowledge weakness set W have been output into the learning path sequence, i.e., the candidate node set. Empty.

[0099] The personalized learning path sequence P generated by the above steps combines the rigor of topological sorting with the adaptability of risk perception, ensuring the scientific nature and personalization of the learning path sequence.

[0100] Step 9: Construct the ZPD interval expression based on the students' mastery of knowledge concepts; obtain the students' ZPD intervals based on the ZPD interval expression.

[0101] The ZPD interval expression is:

[0102] ;

[0103] in, Students ZPD interval, Students knowledge concepts The degree of mastery, For students The personalized zone of proximal development radius, whose value is determined by the base radius parameter. With adjustment coefficient The product is determined, and the adjustment coefficient is adjusted. The learning process is dynamically adjusted based on student type to suit the learning needs of students with different cognitive levels.

[0104] ;

[0105] For example, the base radius parameter Set between 0.1 and 0.3, this value provides the baseline interval width without considering individual differences, serving as a reference for personalized adjustments; adjustment coefficient. For students A type of personalized adjustment factor used for the base radius parameter. Scaling is applied (0.8 for basic learners, 1.0 for baseline learners, and 1.2 for advanced learners) to achieve differentiated intervention. This design results in a narrower ZPD range for basic learners to solidify their foundation, and a wider ZPD range for advanced learners to encourage exploration. In summary, the output of this step is... Defined student The student type is determined as follows: based on the average level of students' mastery of knowledge concepts output in Step 4, those with an average of 0-0.4 are classified as basic learners, those with an average of 0.4-0.8 are classified as benchmark learners, and those with an average of 0.8-1.0 are classified as advanced learners.

[0106] Step 10: Combine the student's learning path sequence with the ZPD interval and the difficulty parameters of the learning resources to define a comprehensive matching score function; obtain the comprehensive matching score between the student and the learning resources based on the comprehensive matching score function.

[0107] The comprehensive matching score function is:

[0108] ;

[0109] in, Indicates learning resources Difficulty parameters, Represents students and difficulty parameters Corresponding learning resources The overall matching score, express and ZPD matching degree express and Knowledge coverage express and The degree of matching of the learning path; ZPD matching degree weight coefficient, For knowledge coverage weighting coefficient, Let be the learning path matching degree weight coefficient, and satisfy . According to the comprehensive scoring function, for a given student... and learning resources Its overall matching score It is a weighted sum of the scores from the three dimensions. For example... , , It emphasizes the importance of matching difficulty levels and filling knowledge gaps.

[0110] The ZPD matching degree expression is:

[0111] ;

[0112] in, Indicates difficulty parameter to ZPD range The shortest distance to the boundary; Students knowledge concepts The degree of mastery, For students The personalized zone of proximal development radius. The ZPD matching degree expression can be used to ensure that learning resources within the ZPD range receive high scores and to penalize learning resources that are far from the ZPD.

[0113] The knowledge coverage expression is:

[0114] ;

[0115] Among them, the function Used to calculate students Learning knowledge concepts The revenue is defined as follows: ;in, For students The set of knowledge weaknesses constituted by the sequence of learning paths; The base score (for example, 0.1). A collection of knowledge concepts; Students knowledge concepts The degree of mastery. The knowledge coverage mentioned aims to strengthen students' weak knowledge points.

[0116] The learning path matching degree expression is:

[0117] ;

[0118] in, The sequence of student learning paths output by Step 8; It is for learning paths The search function, if Belongs to the learning path sequence Then return In the learning path sequence The parameters in, if Not part of the learning path sequence If the condition is met, then 0 is returned. For learning resources With students Learning path matching score, The specific formula is:

[0119] ;

[0120] In the formula, Representative learning resources For the student-learning resource interaction diagram in Step 1.4 The knowledge concepts in the corresponding knowledge concept set C ,and Represents knowledge concepts Existing in students learning path sequence In: If it exists and Output knowledge concepts The priority parameter (if a learning resource corresponds to multiple knowledge concepts, the priority parameter of the knowledge concept with the highest priority is output) , (The highest priority knowledge concept); if the knowledge concept is not in the learning path sequence, then the learning resource should not be recommended, and the output should be 0.

[0121] Step 11: Based on the overall matching score between the student and each learning resource, obtain the strategic score between the student and each learning resource; based on the strategic score between the student and each learning resource, obtain the recommended list of learning resources for the student.

[0122] Step 11 includes:

[0123] Step 11.1: Define a strategy optimization function based on the student type; apply the corresponding strategy optimization function to adjust the overall matching score between the student and each learning resource to obtain the strategy-based score between the student and each learning resource.

[0124] The strategic scoring The expression is:

[0125] ;

[0126] in, For student type The strategy optimization function optimizes the overall matching score based on different student types. Weighted adjustments are made to ensure that score optimization conforms to the partitioning principle of ZPD theory, where type∈{A1,A2,A3}, A1 represents basic learners, A2 represents baseline learners, and A3 represents advanced learners.

[0127] The policy optimization function includes:

[0128] Basic learner optimization function = ;

[0129] Benchmark learner optimization function and ;

[0130] Advanced learner optimization function = .

[0131] Step 11.2: For each student Sort the scores of each learning resource in descending order and select the top ones. There are (type∈{A1,A2,A3}, among which, basic learners) Benchmark learners Advanced learners Learning resources to form students Recommended list of learning resources .

[0132] Based on the publicly available Junyi Academy dataset, this invention's method is used to recommend learning resource lists. This embodiment uses the publicly available Junyi Academy dataset as an example, preprocessing the dataset (the preprocessing includes cleaning invalid data, such as records with answer duration <1 second); extracting 21,460,249 learner historical interaction records (item ID, answer correctness, timestamp) from the cleaned educational information dataset as the educational information dataset (the educational information dataset contains 525,061 students, 835 knowledge concepts (such as "functions and limits"), and 835 learning resources (such as learning resources including exercises and lectures), verifying the effectiveness of the knowledge graph-based adaptive learning resource recommendation method proposed in this invention.

[0133] To facilitate a more intuitive understanding of the experimental results, the following analysis was performed: two baseline learners (students A and B) were randomly selected, along with 15 knowledge concepts that interacted with the two students, and corresponding learning resources (including 12 lectures and 15 exercises). The results were then presented in charts and visualizations.

[0134] The results of Steps 1-4 show the students A and B's mastery of the knowledge concepts, as shown in Table 1.

[0135] Table 1. Students A and B's mastery of 15 knowledge concepts.

[0136]

[0137] The learning risk scores of students A and B for learning 15 knowledge concepts after Step 5-Step 7 are shown in Table 2.

[0138] Table 2. Learning Risk Scores of Students A and B for 15 Knowledge Concepts

[0139]

[0140] After Step 8, the personalized learning path sequence P constructed for students A and B is as follows: Figure 2 , Figure 3 As shown in the figure, the sequence number above the node is the sequence number of the knowledge concept node on the learning path (starting from 1), which is the learning order.

[0141] After Step 9, the ZPD intervals of students A and B are visualized as follows: Figure 4 , Figure 5 As shown in the figure, the current ability is the center of the ZPD interval, the gray interval represents the student's ZPD interval, the circular nodes represent the learning resources corresponding to the knowledge concepts with difficulty parameters within the ZPD interval, and the square nodes represent the learning resources corresponding to the knowledge concepts with difficulty parameters outside the ZPD interval.

[0142] After Step 10-Step 11, the strategic scores of students A and B regarding the two types of learning resources, lectures and exercises, are shown in Tables 3 and 4. Furthermore, based on the strategic scores, a learning resource recommendation list for the students can be generated (it should be noted that students A and B in the example are both baseline learners, and the top 15 strategic scores are selected. In this example, all scores are less than 15, so all scores are sorted to generate the learning resource recommendation list).

[0143] Table 3. Details of Recommended Lecture Learning Resources

[0144]

[0145] Table 4. Details of Recommended Exercise Learning Resources

[0146]

[0147] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A knowledge graph-based adaptive learning resource recommendation method, characterized in that, include: A three-layer knowledge graph consisting of students, learning resources, and knowledge concepts was constructed based on the acquired educational information dataset. Based on the three-layer knowledge graph of students, learning resources, and knowledge concepts, we obtain the students' mastery of knowledge concepts and the set of learning resource nodes including difficulty parameters. Based on students' mastery of knowledge concepts, determine the sequence of learning paths; Based on the student's learning path sequence and the difficulty parameters of the learning resources, a recommended list of learning resources for the student is obtained. The method, based on a three-layer knowledge graph of students, learning resources, and knowledge concepts, obtains the students' mastery of knowledge concepts and a set of learning resource nodes including difficulty parameters, including: Based on the subgraphs in the three-layer knowledge graph of students, learning resources, and knowledge concepts, weighted feature vectors, fusion feature vectors, and behavioral feature vectors are obtained; among them, the subgraphs include concept relationship graphs, learning resource-knowledge concept association graphs, and student-learning resource interaction graphs. The weighted feature vector, fused feature vector, and behavioral feature vector are used as inputs to the trained student-learning resource-knowledge concept three-layer GAT network, and the output is the fused knowledge concept vector; the fused knowledge concept vector is used as input to the word embedding layer, and the output is the standardized feature vector. The standardized feature vector is input into the diagnostic result output layer to obtain the student's mastery of knowledge concepts and a set of learning resource nodes including difficulty parameters; The process of determining the learning path sequence based on students' mastery of knowledge concepts includes: Based on students' mastery of knowledge concepts, a set of students' knowledge weaknesses is obtained; Based on the dependencies between knowledge concepts in the concept graph, construct a directed acyclic graph; Based on the students' set of knowledge weaknesses and the directed acyclic graph, calculate the students' learning risk score for knowledge concepts; Based on the students' set of knowledge weaknesses, the directed acyclic graph, and the learning risk score, a sequence of learning paths is determined; The process of combining the student's learning path sequence and based on the difficulty parameters of the learning resources to obtain a recommended list of learning resources for the student includes: Based on students' mastery of knowledge concepts, construct a ZPD interval expression; based on the ZPD interval expression, obtain the students' ZPD intervals; Based on the student's learning path sequence, and according to the ZPD interval and the difficulty parameters of the learning resources, a comprehensive matching score function is defined; the comprehensive matching score between the student and the learning resources is obtained based on the comprehensive matching score function. Based on the overall matching score between students and each learning resource, a strategic score for students and each learning resource is obtained; based on the strategic score between students and each learning resource, a recommended list of learning resources for students is obtained.

2. The adaptive learning resource recommendation method based on knowledge graphs according to claim 1, characterized in that, The process of obtaining a set of students' knowledge weaknesses based on their mastery of knowledge concepts includes: calculating knowledge gaps for each knowledge concept in the knowledge concept set of the concept relationship diagram based on the students' mastery of knowledge concepts, and taking the knowledge gaps with non-zero values ​​as the first set of knowledge gaps; and constructing a set of students' knowledge weaknesses based on the knowledge concepts corresponding to the first set of knowledge gaps.

3. The adaptive learning resource recommendation method based on knowledge graphs according to claim 1, characterized in that, The process of determining the learning path sequence based on the student's knowledge weakness set, the directed acyclic graph (DAG), and the learning risk score includes: adding all nodes in the DAG with an in-degree of 0 that belong to the knowledge weakness set or are connected by a path to knowledge concepts in the knowledge weakness set to a candidate node set; calculating a priority parameter for each knowledge concept in the candidate node set; using the magnitude of the priority parameter as the student's priority for learning the knowledge concept; selecting the knowledge concept with the highest current priority from the candidate node set, removing it from the candidate node set, and outputting it to the learning path sequence; removing the knowledge concept with the highest current priority from the DAG, and adding the nodes with an in-degree of 0 that are the direct successors of the knowledge concept with the highest current priority to the candidate node set; repeating this process until all knowledge concepts in the knowledge weakness set have been output to the learning path sequence.

4. The adaptive learning resource recommendation method based on knowledge graphs according to claim 1, characterized in that, The ZPD interval expression is: ； in, Students ZPD interval, Students knowledge concepts The degree of mastery; For students The personalized zone of proximal development radius, whose value is determined by the base radius parameter. With adjustment coefficient The product is determined, and the adjustment coefficient is set. The learning process is dynamically adjusted based on student type to suit the learning needs of students with different cognitive levels. 。 5. The adaptive learning resource recommendation method based on knowledge graphs according to claim 1, characterized in that, The comprehensive matching score function is: ； in, Indicates learning resources Difficulty parameters, Represents students and difficulty parameters Corresponding learning resources The overall matching score, express and ZPD matching degree express and Knowledge coverage express and The degree of matching of the learning path; ZPD matching degree weight coefficient, For knowledge coverage weighting coefficient, Let be the learning path matching degree weight coefficient, and satisfy . .

6. The adaptive learning resource recommendation method based on knowledge graphs according to claim 1, characterized in that, The process involves: obtaining a strategic score for each student and each learning resource based on the overall matching score; and generating a recommended list of learning resources for each student based on their strategic scores. This includes: defining a strategy optimization function based on student type; applying the corresponding strategy optimization function to adjust the overall matching score between the student and each learning resource to obtain the strategic score; and sorting the strategic scores of each student and each learning resource in descending order, selecting the top... Each learning resource is used to create a recommended list of learning resources for students; among them... express The number of learning resources selected by students of different student types.

7. An adaptive learning resource recommendation system based on knowledge graphs, characterized in that, The module includes the method described in any one of claims 1-6.

Images