A network learning resource analysis and personalized recommendation method based on knowledge graph
By building an online learning resource analysis and personalized recommendation system based on knowledge graphs, the problems of information overload and learners' differences are solved, personalized learning resource recommendation and intelligent learning path planning are realized, and learning efficiency is improved.
Patent Information
- Application Number
- CN202210633790.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-07
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2042-06-07
AI Technical Summary
With the explosive growth of online education resources, learners are facing information overload problems, and learning goals, knowledge levels, learning paths and learning styles vary greatly, resulting in complex demand and utilization of educational resources, making it difficult to effectively alleviate information overload and personalization. recommend.
Using a knowledge graph-based method, network learning resource knowledge graph is constructed through metadata extraction and transfer learning, and resource analysis is performed using activation diffusion theory and hierarchical deep network. It realizes personalized recommendations in combination with learner portrait analysis, and uses knowledge point centering and Dynamic heterogeneous information networks carry out intelligent learning path planning and resource push.
It realizes personalized recommendation of learning resources and intelligent learning path planning, and accurately push them, alleviates information overload problems, and improves learning efficiency and resource utilization.
Smart Images

Figure CN114861069B_ABST
Abstract
Description
[0001] The present invention belongs to the technical field of learning resource analysis and recommendation methods, and specifically is a network learning resource analysis and personalized recommendation method based on knowledge graph. Background Art
[0002] With the continuous development of Internet technology and the continuous change of application models, the field of online learning has been rapidly developed and widely used. With the rapid development of educational informatization, learners can obtain various forms of online educational resources in a variety of ways, such as MOOCs, Courses, NetEase Cloud Classroom, etc. However, with the explosive growth of online educational resources, the information overload problem faced by learners is becoming increasingly serious. At the same time, there are large differences among learners in learning goals, knowledge levels, learning paths, learning styles, etc., which also makes the demand and utilization of educational resources extremely complicated.
[0003] Therefore, how to effectively alleviate the problems such as information overload brought by massive resources and transform learning activities from the traditional single mode of "people looking for resources" to the intelligent two-way mode of "people looking for resources and resources looking for people" has become an important research topic in the fields of educational informatization and artificial intelligence. In view of this, the present invention proposes a network learning resource analysis and personalized recommendation method based on knowledge graph. Summary of the invention
[0004] The purpose of the present invention is to effectively alleviate the problems such as information overload caused by massive resources, so that learning activities can be transformed from the traditional single mode of "people looking for resources" to an intelligent two-way mode of "people looking for resources and resources looking for people", and a network learning resource analysis and personalized recommendation method based on knowledge graph is proposed.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions:
[0006] A network learning resource analysis and personalized recommendation method based on knowledge graph, specifically comprising the following steps:
[0007] S1. Constructing the knowledge graph of online learning resources: Using metadata extraction methods to study unstructured subject knowledge, solving the problem of insufficient corpus and lack of annotated corpus in the field of education, and using transfer learning methods to construct the knowledge graph of online learning resources based on the existing knowledge graphs of teaching resources in various fields;
[0008] S2. Online learning resource analysis: Based on the online learning resource knowledge graph constructed in S1, the activation diffusion theory is used to model the teaching resources in various fields in the online learning resource knowledge graph, and a vector model that can represent the knowledge points in the corresponding field is obtained. At the same time, feature selection is performed through a hierarchical deep network;
[0009] S3. Learner portrait analysis: Collect learner learning analysis data, build and dynamically update learner portraits based on the collected learning analysis data, and derive learner learning resource needs through analysis of learner portraits;
[0010] S4, personalized recommendation of online learning resources: Combining the contents obtained in S2 and S3, intelligent learning path planning is carried out for learners based on the learning path recommendation method of knowledge point centrality and dynamic heterogeneous information network. At the same time, accurate push and personalized recommendation of online learning resources are realized according to the learner portrait obtained in S3.
[0011] Preferably, the network learning resource knowledge graph mentioned in S1 is represented by a composite structure, and the network learning resource knowledge graph construction process specifically includes the following steps:
[0012] A1. Using metadata extraction methods and combining the knowledge graphs of existing teaching resources in various fields, the learning content in the knowledge graphs of various fields is divided into individual knowledge points;
[0013] A2. Based on the knowledge point data obtained in A1, find out the dependency relationship between knowledge points in the same field and between knowledge points in different fields, and determine the order of learning;
[0014] A3. Guided by the dependency relationship between the knowledge points obtained in A2, based on the transfer learning method, the knowledge points are linked together to construct a complete knowledge graph of online learning resources.
[0015] Preferably, the learner learning analysis data mentioned in S3 includes explicit data and implicit data, and the explicit data specifically includes learner knowledge and ability structure data, learning style analysis data, online learning preference analysis data and learning resource demand analysis data; the implicit data includes online learning process analysis data and online learning result analysis data.
[0016] Preferably, the personalized recommendation of online learning resources mentioned in S4 specifically includes the following steps:
[0017] B1. Based on the constructed network learning resource knowledge graph and knowledge point vector model, form a "network learning resource database";
[0018] B2. Combining the "online learning resource database" obtained in B1 and the explicit data in S3, conducting learning analysis on the learners, specifically including learning style analysis, online learning preference analysis and learning resource demand analysis;
[0019] B3. After the learning analysis in B2 is completed, the content of personalized learning resources is determined and called from the "online learning resource database" according to the analysis results, including selecting resource type, confirming push content, confirming recommendation time and confirming recommendation frequency;
[0020] B4. Automatically push the content determined in B3, and learners conduct personalized learning based on the pushed content, while automatically recording and storing the online learning process;
[0021] B5, analyzing the online learning data recorded and stored in B4, specifically including online learning process analysis and online learning result analysis, and evaluating the online learning based on the obtained analysis data;
[0022] B6. Based on the online learning evaluation results obtained in B5, the push content in B3 is updated and adjusted in real time;
[0023] B7. After the adjustment is completed, intelligent learning path planning, accurate push and personalized recommendation of online learning resources will be automatically provided to learners.
[0024] Preferably, in the learning path recommendation method based on knowledge point centrality mentioned in S4, the knowledge point centrality can be adjusted dynamically. When evaluating the centrality of a knowledge point, the degree value and contribution value of the knowledge point network node are considered at the same time. Based on the degree value and contribution value of the knowledge point, a knowledge point centrality calculation model is proposed, and its calculation formula is:
[0025] C i =AZ i Q (1)
[0026] In the formula, C i represents the centrality of the node; A is the evaluation coefficient matrix, which represents the contribution of the knowledge point itself and its neighboring nodes of each order to the importance of the knowledge point i; Z i is the evaluation index matrix of the degree and contribution value of knowledge point i, including the degree and contribution value of knowledge point i and neighboring nodes of each order; Q is the index weight matrix. For the degree and contribution value centrality calculation method, the calculation method of the contribution of neighboring nodes of the same order to the centrality of node i in the calculation of centrality based on contribution value is retained, that is:
[0027]
[0028] In the formula, is the total contribution of the m-order neighbor node set of node i to the centrality of node i, is the value of node j in the m-order neighbor node set of node i; therefore, the moderate value and contribution value evaluation index matrix Z of formula (1) i It can be expressed as:
[0029]
[0030] In the formula, the first column element is the degree value, and the second column element is the contribution value. Due to the different ranges of degree values and contribution values, the meaning and measurement unit of each indicator are also different, which will lead to different magnitudes of data. i Do normalization processing; for Z i The degree value and contribution value in are normalized using the formula:
[0031]
[0032] In the formula, is the normalized index value, Calculate the normalized degree value and contribution value evaluation index matrix Z′ i ,make C i =AZ i Q, so the centrality of node i can be expressed as C i =AZ′ i Q, that is:
[0033]
[0034] By comprehensively considering the centrality of the knowledge point itself and the m-order neighbor nodes under the constraints of degree value and contribution value, an accurate result can be obtained; in summary, the algorithm of the knowledge point centrality based on degree value and contribution value can be summarized as follows:
[0035] C1. According to the topological structure of the directed acyclic graph, extract the 1 to m order neighbor node set of knowledge point i:
[0036] C2. Calculate the degree and contribution values of the neighbors of each order of knowledge point i, and determine the degree and contribution evaluation index matrix Z of knowledge point i i ;
[0037] C3, targeting Z i The degree and contribution values in are normalized, and the normalized degree and contribution evaluation index matrix Z′ is calculated. i ;
[0038] C4. Calculate the centrality of each node according to formula (5) and output C i .
[0039] Preferably, the learning path recommendation method based on a dynamic heterogeneous information network mentioned in S4 specifically comprises the following steps:
[0040] D1. Extract subgraphs from the knowledge graph and construct a heterogeneous information network of learners’ learning resources;
[0041] D2, adopts a random walk strategy guided by symmetric meta-paths to fully explore the complex relationships hidden in the semantic associations in the heterogeneous information network in D1;
[0042] D3. Introduce a fusion strategy based on the attention mechanism to organically integrate the preference features generated by different weighted meta-paths;
[0043] D4, by jointly optimizing the matrix decomposition model and the fusion function, using attribute information to solve the scalability problem;
[0044] D5. Conduct experimental analysis on the processing flow proposed in D1 to D4 on a real large-scale dataset, and then build a learning path recommendation model based on a dynamic heterogeneous information network.
[0045] Compared with the prior art, the present invention provides a method for analyzing network learning resources and personalized recommendation based on knowledge graph, which has the following beneficial effects:
[0046] The present invention integrates the personalized education knowledge graph model of subject knowledge, learning resources, and learning strategies and its construction technology, the concept linking, analysis and evaluation technology of learning resources based on the knowledge graph, and the learner knowledge system evaluation model and learning path intelligent planning of the knowledge graph and learning goals, thereby realizing accurate knowledge push and personalized learning resource and learning strategy recommendation centered on learners' personalized interests and needs. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 This is a flow chart of a method for personalized recommendation of network learning resources based on knowledge graph analysis and personalized recommendation proposed by the present invention. DETAILED DESCRIPTION
[0048] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0049] Embodiment 1:
[0050] See also Figure 1 , a network learning resource analysis and personalized recommendation method based on knowledge graph, specifically comprising the following steps:
[0051] S1. Constructing the knowledge graph of online learning resources: Using metadata extraction methods to study unstructured subject knowledge, solving the problem of insufficient corpus and lack of annotated corpus in the field of education, and using transfer learning methods to construct the knowledge graph of online learning resources based on the existing knowledge graphs of teaching resources in various fields;
[0052] The network learning resource knowledge graph mentioned in S1 is represented by a composite structure. The construction process of the network learning resource knowledge graph specifically includes the following steps:
[0053] A1. Using metadata extraction methods and combining the existing knowledge graphs of teaching resources in various fields, the learning content in the knowledge graphs of various fields is divided into individual knowledge points;
[0054] A2. Based on the knowledge point data obtained in A1, find out the dependency relationship between knowledge points in the same field and between knowledge points in different fields, and determine the order of learning;
[0055] A3, guided by the dependencies between the knowledge points obtained in A2, based on the transfer learning method, link the knowledge points to build a complete knowledge graph of online learning resources;
[0056] S2. Online learning resource analysis: Based on the online learning resource knowledge graph constructed in S1, the activation diffusion theory is used to model the teaching resources in various fields in the online learning resource knowledge graph, and a vector model that can represent the knowledge points in the corresponding field is obtained. At the same time, feature selection is performed through a hierarchical deep network;
[0057] S3. Learner portrait analysis: Collect learner learning analysis data, build and dynamically update learner portraits based on the collected learning analysis data, and derive learner learning resource needs through analysis of learner portraits;
[0058] The learner learning analysis data mentioned in S3 includes explicit data and implicit data. Explicit data specifically includes learner knowledge and ability structure data, learning style analysis data, online learning preference analysis data and learning resource demand analysis data; implicit data includes online learning process analysis data and online learning result analysis data;
[0059] S4, personalized recommendation of online learning resources: Combining the contents obtained in S2 and S3, the learning path recommendation method based on the centrality of knowledge points is used to plan the intelligent learning path for learners. At the same time, accurate push and personalized recommendation of online learning resources are realized based on the learner portrait obtained in S3.
[0060] The personalized recommendation of online learning resources mentioned in S4 specifically includes the following steps:
[0061] B1. Based on the constructed network learning resource knowledge graph and knowledge point vector model, form a "network learning resource database";
[0062] B2. Combine the "online learning resource database" obtained in B1 and the explicit data in S3 to conduct learning analysis on learners, including learning style analysis, online learning preference analysis and learning resource demand analysis;
[0063] After the learning analysis in B3 and B2 is completed, the content of personalized learning resources is determined and called from the "online learning resource database" according to the analysis results, including selecting resource type, confirming push content, confirming recommendation time and confirming recommendation frequency;
[0064] B4. Automatically push the content determined in B3, and learners conduct personalized learning based on the pushed content. At the same time, the online learning process is automatically recorded and stored;
[0065] B5. Analyze the online learning data recorded and stored in B4, including online learning process analysis and online learning result analysis, and evaluate online learning based on the obtained analysis data;
[0066] B6, based on the online learning evaluation results obtained in B5, the push content in B3 is updated and adjusted in real time;
[0067] B7. After the adjustment is completed, intelligent learning path planning and accurate push and personalized recommendation of online learning resources will be automatically provided to learners;
[0068] The learning path recommendation method based on knowledge point centrality mentioned in S4 considers the degree value and contribution value of the knowledge point network node when evaluating the centrality of the knowledge point. Based on the degree value and contribution value of the knowledge point, a knowledge point centrality calculation model is proposed, and its calculation formula is:
[0069] C i =AZ i Q (1)
[0070] In the formula, C i represents the centrality of the node; A is the evaluation coefficient matrix, which represents the contribution of the knowledge point itself and its neighboring nodes of each order to the importance of the knowledge point i; Z i is the evaluation index matrix of the degree and contribution value of knowledge point i, including the degree and contribution value of knowledge point i and neighboring nodes of each order; Q is the index weight matrix. For the degree and contribution value centrality calculation method, the calculation method of the contribution of neighboring nodes of the same order to the centrality of node i in the calculation of centrality based on contribution value is retained, that is:
[0071]
[0072] In the formula, is the total contribution of the m-order neighbor node set of node i to the centrality of node i, is the value of node j in the m-order neighbor node set of node i; therefore, the moderate value and contribution value evaluation index matrix Z of formula (1) i It can be expressed as:
[0073]
[0074] In the formula, the first column element is the degree value, and the second column element is the contribution value. Due to the different ranges of degree values and contribution values, the meaning and measurement unit of each indicator are also different, which will lead to different magnitudes of data. i Do normalization processing; for Z i The degree value and contribution value in are normalized using the formula:
[0075]
[0076] In the formula, is the normalized index value, Calculate the normalized degree value and contribution value evaluation index matrix Z′ i ,make C i =AZ i Q, so the centrality of node i can be expressed as C i =AZ′ i Q, that is:
[0077]
[0078] By comprehensively considering the centrality of the knowledge point itself and the m-order neighbor nodes under the constraints of degree value and contribution value, an accurate result can be obtained; in summary, the algorithm of the knowledge point centrality based on degree value and contribution value can be summarized as follows:
[0079] C1. According to the topological structure of the directed acyclic graph, extract the 1 to m order neighbor node set of knowledge point i:
[0080] C2. Calculate the degree and contribution values of the neighbors of each order of knowledge point i, and determine the degree and contribution evaluation index matrix Z of knowledge point i i ;
[0081] C3, targeting Z i The degree and contribution values in are normalized, and the normalized degree and contribution evaluation index matrix Z′ is calculated. i ;
[0082] C4. Calculate the centrality of each node according to formula (5) and output C i .
[0083] The learning path recommendation method based on dynamic heterogeneous information network mentioned in S4 specifically includes the following steps:
[0084] D1. Extract subgraphs from the knowledge graph and construct a heterogeneous information network of learners’ learning resources;
[0085] D2, adopts a random walk strategy guided by symmetric meta-paths to fully explore the complex relationships hidden in the semantic associations in the heterogeneous information network in D1;
[0086] D3. Introduce a fusion strategy based on the attention mechanism to organically integrate the preference features generated by different weighted meta-paths;
[0087] D4, by jointly optimizing the matrix decomposition model and the fusion function, using attribute information to solve the scalability problem;
[0088] D5. Conduct experimental analysis on the processing flow proposed in D1 to D4 on a real large-scale dataset, and then build a learning path recommendation model based on a dynamic heterogeneous information network.
[0089] The present invention integrates the personalized education knowledge graph model of subject knowledge, learning resources, and learning strategies and its construction technology, the concept linking, analysis and evaluation technology of learning resources based on the knowledge graph, and the learner knowledge system evaluation model and learning path intelligent planning of the knowledge graph and learning goals, thereby realizing accurate knowledge push and personalized learning resource and learning strategy recommendation centered on learners' personalized interests and needs.
[0090] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A method for analyzing and recommending online learning resources based on knowledge graphs. It is characterized in that The specific steps include: S1. Constructing the knowledge graph of online learning resources: Using metadata extraction methods to study unstructured subject knowledge, solving the problem of insufficient corpus and lack of annotated corpus in the field of education, and using transfer learning methods to construct the knowledge graph of online learning resources based on the existing knowledge graphs of teaching resources in various fields; S2. Online learning resource analysis: Based on the online learning resource knowledge graph constructed in S1, the activation diffusion theory is used to model the teaching resources in various fields in the online learning resource knowledge graph, and a vector model that can represent the knowledge points in the corresponding field is obtained. At the same time, feature selection is performed through a hierarchical deep network; S3. Learner portrait analysis: Collect learner learning analysis data, build and dynamically update learner portraits based on the collected learning analysis data, and derive learner learning resource needs through analysis of learner portraits; S4, personalized recommendation of online learning resources: Combining the contents obtained in S2 and S3, intelligent learning path planning is performed for learners based on the learning path recommendation method of knowledge point centrality and dynamic heterogeneous information network. At the same time, accurate push and personalized recommendation of online learning resources are achieved based on the learner portrait obtained in S3; The learning path recommendation method based on knowledge point centrality described in S4 specifically refers to: the knowledge point centrality can be adjusted dynamically, and when evaluating the knowledge point centrality, the degree value and contribution value of the knowledge point network node are considered at the same time. Based on the knowledge point degree value and contribution value, a knowledge point centrality calculation model is proposed, and its calculation formula is: C i =AZ i Q (1) In the formula, C i represents the centrality of the node; A is the evaluation coefficient matrix, which represents the contribution of the knowledge point itself and its neighboring nodes of each order to the importance of the knowledge point i; Z i is the evaluation index matrix of the degree and contribution value of knowledge point i, including the degree and contribution value of knowledge point i and neighboring nodes of each order; Q is the index weight matrix. For the degree and contribution value centrality calculation method, the calculation method of the contribution of neighboring nodes of the same order to the centrality of node i in the calculation of centrality based on contribution value is retained, that is: In the formula, is the total contribution of the m-order neighbor node set of node i to the centrality of node i, is the value of node j in the m-order neighbor node set of node i; therefore, the moderate value and contribution value evaluation index matrix Z of formula (1) i It can be expressed as: In the formula, the first column element is the degree value, and the second column element is the contribution value. Due to the different ranges of degree values and contribution values, the meaning and measurement unit of each indicator are also different, which will lead to different magnitudes of data. i Do normalization processing; for Z i The degree value and contribution value in are normalized using the formula: In the formula, is the normalized index value, Calculate the normalized degree value and contribution value evaluation index matrix Z′ i , let A = [α, β, β 2 , …, β m ], C i =AZ i Q, so the centrality of node i can be expressed as C i =AZ′ i Q, that is: C i =[a, b, b] 2 ,...,b m ]·Z′ i ·Q (5) By comprehensively considering the centrality of the knowledge point itself and the m-order neighbor nodes under the constraints of degree value and contribution value, an accurate result can be obtained; in summary, the algorithm of the knowledge point centrality based on degree value and contribution value can be summarized as follows: C1. According to the topological structure of the directed acyclic graph, extract the 1 to m order neighbor node set of knowledge point i: C2. Calculate the degree and contribution values of the neighbors of each order of knowledge point i, and determine the degree and contribution evaluation index matrix Z of knowledge point i i ; C3, targeting Z i The degree and contribution values in are normalized, and the normalized degree and contribution evaluation index matrix Z′ is calculated. i ; C4. Calculate the centrality of each node according to formula (5) and output C i ; The learning path recommendation method based on a dynamic heterogeneous information network described in S4 specifically includes the following steps: D1. Extract subgraphs from the knowledge graph and construct a heterogeneous information network of learners’ learning resources; D2, adopts a random walk strategy guided by symmetric meta-paths to fully explore the complex relationships hidden in the semantic associations in the heterogeneous information network in D1; D3. Introduce a fusion strategy based on the attention mechanism to organically integrate the preference features generated by different weighted meta-paths; D4, by jointly optimizing the matrix decomposition model and the fusion function, using attribute information to solve the scalability problem; D5. Conduct experimental analysis on the processing flow proposed in D1 to D4 on a real large-scale dataset, and then build a learning path recommendation model based on a dynamic heterogeneous information network.
2. According to the method for analyzing and recommending network learning resources based on knowledge graph according to claim 1, It is characterized in that The network learning resource knowledge graph mentioned in S1 is represented by a composite structure, and the network learning resource knowledge graph construction process specifically includes the following steps: A1. Using metadata extraction methods and combining the knowledge graphs of existing teaching resources in various fields, the learning content in the knowledge graphs of various fields is divided into individual knowledge points; A2. Based on the knowledge point data obtained in A1, find out the dependency relationship between knowledge points in the same field and between knowledge points in different fields, and determine the order of learning; A3. Guided by the dependency relationship between the knowledge points obtained in A2, based on the transfer learning method, the knowledge points are linked together to construct a complete knowledge graph of online learning resources.
3. According to the method for analyzing network learning resources and personalized recommendation based on knowledge graph according to claim 1, It is characterized in that The learner learning analysis data mentioned in S3 includes explicit data and implicit data. The explicit data specifically includes learner knowledge and ability structure data, learning style analysis data, online learning preference analysis data and learning resource demand analysis data; the implicit data includes online learning process analysis data and online learning result analysis data.
4. A method for analyzing and recommending network learning resources based on knowledge graph according to claim 1 or 3, It is characterized in that The personalized recommendation of online learning resources mentioned in S4 specifically includes the following steps: B1. Based on the constructed network learning resource knowledge graph and knowledge point vector model, a "network learning resource database" is formed; B2. Combining the "online learning resource database" obtained in B1 and the explicit data in S3, conducting learning analysis on the learners, specifically including learning style analysis, online learning preference analysis and learning resource demand analysis; B3. After the learning analysis in B2 is completed, the content of personalized learning resources is determined and called from the "network learning resource database" according to the obtained analysis results, including selecting resource type, confirming push content, confirming recommendation time and confirming recommendation frequency; B4. Automatically push the content determined in B3, and learners conduct personalized learning based on the pushed content, while automatically recording and storing the online learning process; B5, analyzing the online learning data recorded and stored in B4, specifically including online learning process analysis and online learning result analysis, and evaluating the online learning based on the obtained analysis data; B6. Based on the online learning evaluation results obtained in B5, the push content in B3 is updated and adjusted in real time; B7. After the adjustment is completed, intelligent learning path planning, accurate push and personalized recommendation of online learning resources will be automatically provided to learners.
Citation Information
Patent Citations
Dynamic heterogeneous information network embedding method and device and electronic equipment
CN109992631A
Recommendation method and system based on commodity knowledge graph feature learning
CN111369318A
