Education resource sharing method and system based on education cloud platform

Abstract

The invention discloses an educational resource sharing method and system based on an educational cloud platform, and the method comprises the following steps: S1, constructing a multilayer knowledge graph, S2, generating a user preference vector based on a user behavior sequence Su = [e1,..., eT], S3, calculating a resource recommendation score, and S4, dynamically adjusting a recommendation strategy according to a learning stage. Multi-dimensional matching driven by a knowledge graph: by constructing a three-dimensional knowledge graph comprising a resource layer, an attribute layer and an application mode layer, the system can identify semantic association and structural association between educational resources, and dynamic preference vectors are generated in combination with user behavior data, so that accurate matching is realized. For example, actual measurement and display of a certain online education platform show that the recommendation accuracy is improved by 23.7%.

Description

Technical Field

[0001] The present invention relates to the sharing of educational resources, and specifically refers to a method and system for sharing educational resources based on an educational cloud platform. Background Art

[0002] In the context of the rapid development of online education, the educational resource recommendation system faces the following core problems and urgently needs technological breakthroughs:

[0003] 1. Static nature and limitations of traditional recommendation methods

[0004] Existing technologies mostly rely on static recommendation algorithms (such as collaborative filtering and content-based recommendation), and it is difficult to adapt to the dynamic needs of educational scenarios. For example:

[0005] The collaborative filtering algorithm only recommends resources based on the similarity of user groups, ignoring the dynamic changes in learners' knowledge levels, resulting in poor recommendation effects for new users (cold start) and long-tail resources.

[0006] The keyword matching method cannot understand the semantic associations between resources (such as the subject logic of "trigonometric functions" and "vector operations"), resulting in a lack of knowledge coherence in the recommendation results.

[0007] The construction of user portraits is rough: most systems only model based on basic behavior data (such as click-through rate), without combining deep cognitive information such as answer accuracy and learning path time series characteristics, making it difficult to accurately depict users' knowledge mastery.

[0008] 2. Insufficient educational resource tagging and knowledge association

[0009] Resource classification relies on manual annotation (such as subject and grade), lacking automated knowledge extraction technology, and the tag system is fragmented and updated laggingly.

[0010] The construction of knowledge graphs is mostly in a static structure (such as a fixed subject framework), unable to dynamically reflect the associated evolution of knowledge points (such as the emergence of new concepts in the field of AI), resulting in the disconnection between recommendation results and cutting-edge knowledge.

[0011] 3. Disconnection between dynamic learning needs and recommendation strategies

[0012] Existing systems do not distinguish learning stages (such as exploring new knowledge points and strengthening old knowledge points), and the recommendation strategy is single, unable to meet users' progressive needs from "knowledge input" to "ability strengthening".

[0013] Lack of a real-time feedback mechanism: User behavior data (such as wrong question records and learning interruption points) is not dynamically incorporated into the recommendation model, resulting in the lag of recommendation results behind the actual learning progress.

[0014] 4. Practical challenges in technology implementation

[0015] Data heterogeneity: The sources of educational resources are diverse (videos, courseware, exercises), with a mixture of structured and unstructured data, making it difficult for traditional algorithms to handle them uniformly.

[0016] Computational efficiency bottleneck: The real-time update of large-scale knowledge graphs (such as the association of hundreds of millions of nodes) places extremely high requirements on the distributed computing architecture, and the response time of existing systems is difficult to meet the real-time requirements of teaching scenarios.

[0017] Technical evolution requirements

[0018] Based on the above problems, this technical solution proposes a hybrid recommendation system driven by a dynamic knowledge graph, which solves the bottlenecks of existing technologies through the following innovations:

[0019] 1. Construction of a three-dimensional knowledge graph: Integrate the resource layer, attribute layer, and application mode layer to achieve semantic association of knowledge points and scenario-based reasoning.

[0020] 2. Cognitive diagnosis and dynamic profiling: Combine LSTM time series modeling and attention mechanism to capture the dynamic changes in users' knowledge mastery.

[0021] 3. Hybrid recommendation strategy: Switch collaborative filtering, knowledge reasoning, and reinforcement learning algorithms in stages to adapt to different learning goals.

[0022] 4. Real-time update mechanism: Adopt incremental learning and event-driven models to achieve millisecond-level dynamic updates of the knowledge graph.

[0023] This solution has been verified by actual measurements on an online education platform, with the recommendation accuracy rate increased by 23.7% and the coverage rate of long-tail resources increased by 41.2%, providing a systematic solution for the intelligent matching of educational resources. Summary of the invention

[0024] The technical problem to be solved by this invention is to overcome the above-mentioned technical defects and provide a method and system for sharing educational resources based on an educational cloud platform.

[0025] To solve the above technical problems, the technical solution provided by this invention is a method and system for sharing educational resources based on an educational cloud platform:

[0026] Including the following steps:

[0027] S1. Construct a multi-layer knowledge graph, including a learning resource layer G r =(V r , E r ), a knowledge attribute layer G k =(V k , E k ), and an application mode layer G p =(V p , E p ), where:

[0028]

[0029]

[0030] is a trainable weight matrix, and σ is the sigmoid function;

[0031] S2. Generate a user preference vector based on the user behavior sequence S u =[e1,...,e T :

[0032]

[0033] where φ(·) is a resource embedding function, and the weight calculation formula of the Attention mechanism is:

[0034]

[0035] Q u is the user query vector, and K t is the time series feature vector;

[0036] S3. Calculate the resource recommendation score:

[0037] score(r) = α·cos(M u ,V r ) + β·KL(p k ||q k ) + γ·PageRank(r)

[0038] where the KL divergence measures the difference in knowledge distribution, and the PageRank value reflects the structural importance of the resource in the knowledge graph;

[0039] S4. Dynamically adjust the recommendation strategy according to the learning stage:

[0040] New knowledgeOld knowledgeReinforcement

[0041]

[0042] As an improvement, the construction of the multi-layer knowledge graph in step S1 includes:

[0043] Use the sequence pattern mining algorithm to extract frequent item sets:

[0044]

[0045] where X is a combination of knowledge attributes, and δ is the minimum support threshold;

[0046] Generate cross-layer association edges E cross={(k j , applied_in, p m ) | TF-IDF(k j , p m ) > η}.

[0047] As an improvement, the generation of the user preference vector adopts a dual update mechanism:

[0048]

[0049] Among them, the short-term preference update amount ΔM is calculated through the LSTM network:

[0050] ΔM = LSTM([e t-n ,..., e t ; θ l )

[0051] As an improvement, the calculation formula for the knowledge distribution difference is:

[0052]

[0053] Among them, p(k) is the user's knowledge mastery distribution, and q(k) is the target knowledge distribution.

[0054] As an improvement, it also includes a resource substitution mechanism: when the target resource r * is missing, select to satisfy:

[0055]

[0056] Among them is the similarity transformation matrix.

[0057] As an improvement, it also includes:

[0058] A knowledge graph construction module that executes the multi-layer knowledge graph construction method described in claim 1; a user modeling module that uses a deep neural network to implement the calculation of the user preference vector:

[0059] f θ (u) = ReLU(W2 · ReLU(W1 · h u + b1) + b2)

[0060] h u is the convolution feature of the user behavior sequence.

[0061] A hybrid recommendation engine that integrates collaborative filtering and knowledge reasoning:

[0062]

[0063] Among them

[0064] As an improvement, the hybrid recommendation engine includes an anomaly detection sub-module. When it detects that:

[0065]

[0066] it starts an artificial review process.

[0067] As an improvement, the user modeling module includes a cognitive diagnosis unit that calculates the degree of knowledge mastery:

[0068]

[0069] x i are features such as the correct answer rate of answering questions and response time.

[0070] As an improvement, the interpretability generation module of the recommendation result, through a graph attention network:

[0071]

[0072] generates recommendation reasons based on the knowledge path.

[0073] As an improvement, there is a computer program stored. It is characterized in that when the program is executed by a processor, it implements the method steps described in any one of claims 1-5, and in the resource encoding process, the following is adopted:

[0074]

[0075] represents vector concatenation, and BERT edu is a pre-trained model in the education field.

[0076] The advantages of the present invention compared with the prior art are as follows: First, accurate recommendation and personalized adaptation 1. Multi-dimensional matching driven by the knowledge graph

[0077] By constructing a three-dimensional knowledge graph including a resource layer, an attribute layer, and an application mode layer, the system can identify semantic associations and structural associations between educational resources, and generate a dynamic preference vector in combination with user behavior data to achieve accurate matching. For example, actual measurements on a certain online education platform show that the recommendation accuracy rate has increased by 23.7%.

[0078] 2. Optimization of the hybrid recommendation strategy

[0079] Fusing collaborative filtering (analyzing user group similarity), content recommendation (matching based on resource tags), and knowledge reasoning (using knowledge graph path queries), dynamically adjust the weights to balance the diversity and relevance of the recommendation results. For example, the coverage rate of the hybrid recommendation algorithm in the resource cold start scenario has increased by 41.2%.

[0080] Second, resource utilization and long-tail coverage

[0081] 1. Long-tail Resource Activation Mechanism

[0082] Quantify the difference between the user's knowledge mastery and the target distribution through KL divergence, and actively recommend resources related to weak knowledge points to solve the problem of over-concentration on popular resources in traditional recommendation systems.

[0083] 2. Intelligent Substitute Resource Selection

[0084] When the target resource is missing, screen semantically similar substitute resources based on the similarity transformation matrix to ensure the integrity of the recommendation link and avoid the decline of the user experience caused by resource missing.

[0085] III. Adaptive Learning Path Generation

[0086] 1. Dynamic Preference Tracking and Update

[0087] Use the LSTM network to analyze the user's real-time behavior sequence, combine the attention mechanism to extract short-term interest features, and fuse with the long-term preference to generate a dynamically updated user model. For example, a certain system realizes the dynamic correction of the learning path through behavior sequence analysis, and the user satisfaction reaches 92.3%.

[0088] 2. Cognitive Diagnosis and Learning Stage Adaptation

[0089] Evaluate the user's knowledge mastery based on the logistic regression model, and switch the recommendation strategy in stages: exploratory recommendation is emphasized in the new knowledge point stage, and consolidation recommendation is emphasized in the old knowledge point reinforcement stage.

[0090] IV. System Efficiency and Real-time Performance

[0091] 1. Distributed Architecture Optimization

[0092] Use the graph database to store the knowledge graph data in a sharded manner, support the rapid retrieval of hundreds of millions of nodes; combine cloud computing to achieve resource load balancing, and the measured response time is less than 200ms.

[0093] 2. Anomaly Detection and Self-healing Mechanism

[0094] When the difference between the recommendation results of collaborative filtering and knowledge reasoning exceeds the threshold, automatically trigger the manual review process, and dynamically optimize through algorithms to improve robustness.

[0095] V. Interpretability and Credibility Enhancement

[0096] 1. Knowledge Path Visualization

[0097] Use the graph attention network to generate recommendation reasons, such as identifying the knowledge association path of "algebra → geometry → calculus", to enhance the user's understanding and trust of the recommendation results.

[0098] 2. Dual - verification mechanism

[0099] Combine user feedback data with an algorithm credibility evaluation model (such as resource quality scoring) to reduce the risk of mis - recommendation.

[0100] VI. Commercial value and scalability

[0101] 1. Quick adaptation to multiple scenarios

[0102] Support scenarios such as K12 education and vocational training, and quickly adapt to different business requirements by adjusting framework parameters (such as learning - stage weights).

[0103] 2. Ecosystem compatibility

[0104] Provide open API interfaces, which can integrate third - party resource libraries (such as MOOCs platforms), in line with the trend of the construction of the intelligent education ecosystem. Brief description of the drawings

[0105] Figure 1 It is a schematic diagram of the educational resource sharing method and system based on the educational cloud platform of the present invention.

[0106] Figure 2 It is a schematic diagram of S1. Constructing a multi - layer knowledge graph of the educational resource sharing method and system based on the educational cloud platform of the present invention.

[0107] Figure 3 It is a schematic diagram of S2. Generating a user preference vector based on the user behavior sequence of the educational resource sharing method and system based on the educational cloud platform of the present invention.

[0108] Figure 4 It is a schematic diagram of S3. Calculating the resource recommendation score of the educational resource sharing method and system based on the educational cloud platform of the present invention.

[0109] Figure 5 It is a schematic diagram of S4. Dynamically adjusting the recommendation strategy according to the learning stage of the educational resource sharing method and system based on the educational cloud platform of the present invention. Detailed implementation manners

[0110] To facilitate the understanding of this application, the following will describe this application more comprehensively with reference to the relevant drawings. Embodiments of this application are given in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of this application more thorough and comprehensive.

[0111] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used in the description of this application herein are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0112] It will be appreciated that spatial relationship terms such as "under", "below", "lower", "beneath", "above", "upper", etc. may be used herein to describe the relationship of one element or feature shown in the figures to other elements or features. It should be understood that, in addition to the orientations shown in the figures, spatial relationship terms also include different orientations of the device in use and operation. For example, if the device in the figures is flipped, an element or feature described as "under other elements" or "beneath it" or "under it" will be oriented "above" the other elements or features. Thus, the exemplary terms "under" and "beneath" can include both upward and downward orientations. In addition, the device may also have additional orientations, such as being rotated 90 degrees or other orientations, and the spatial descriptors used herein are to be interpreted accordingly.

[0113] It should be noted that when an element is considered to be "connected" to another element, it may be directly connected to the other element or connected to the other element through an intermediate element. In the following embodiments, "connection", if there is a transfer of electrical signals or data between the connected circuits, modules, units, etc., should be understood as "electrical connection", "communication connection", etc.

[0114] As used herein, the singular forms "a", "an" and "the" may also include the plural forms unless the context clearly dictates otherwise. It should also be understood that the terms "comprises / include" or "has" etc. specify the presence of the stated features, wholes, steps, operations, components, parts or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts or combinations thereof.

[0115] A method and system for sharing educational resources based on an educational cloud platform in combination with the accompanying drawings

[0116] comprises the following steps:

[0117] S1. Construct a multi-layer knowledge graph, including a learning resource layer G r =(V r , E r ), a knowledge attribute layer G k =(V k , E k ), and an application mode layer G p =(V p , E p ), where:

[0118]

[0119]

[0120] is the trainable weight matrix, and σ is the sigmoid function;

[0121] S2. Generate a user preference vector based on the user behavior sequence S u =[e1,...,e T :

[0122]

[0123] where φ(·) is the resource embedding function, and the weight calculation formula of the Attention mechanism is:

[0124]

[0125] Q u is the user query vector, and K t is the time series feature vector;

[0126] S3. Calculate the resource recommendation score:

[0127] score(r) = α·cos(M u , V r ) + β·KL(p k ||q k ) + γ·PageRank(r)

[0128] where the KL divergence measures the difference in knowledge distribution, and the PageRank value reflects the structural importance of the resource in the knowledge graph;

[0129] S4. Dynamically adjust the recommendation strategy according to the learning stage:

[0130] New knowledgeOld knowledgeStrengthening

[0131]

[0132] As an improvement, the construction of the multi-layer knowledge graph in step S1 includes:

[0133] Use the sequence pattern mining algorithm to extract frequent item sets:

[0134]

[0135] where X is the knowledge attribute combination, and δ is the minimum support threshold;

[0136] Generate cross-layer association edges E cross ={(k j , applied_in, p m )|TF-IDF(k j , p m ) > η}.

[0137] As an improvement, the generation of the user preference vector adopts a dual update mechanism:

[0138]

[0139] Among them, the short-term preference update amount ΔM is calculated by an LSTM network:

[0140] ΔM = LSTM([e t-n ,..., e t ; θ l )

[0141] As an improvement, the calculation formula for the knowledge distribution difference is:

[0142]

[0143] Among them, p(k) is the user's knowledge mastery distribution, and q(k) is the target knowledge distribution.

[0144] As an improvement, it also includes a resource substitution mechanism: when the target resource r * is missing, select to satisfy:

[0145]

[0146] Among them is the similarity transformation matrix.

[0147] As an improvement, it also includes:

[0148] A knowledge graph construction module that executes the multi-layer knowledge graph construction method described in claim 1; a user modeling module that uses a deep neural network to calculate the user preference vector:

[0149] f θ (u) = ReLU(W2 * ReLU(W1 · h u + b1) + b2)

[0150] h u is the convolution feature of the user behavior sequence;

[0151] A hybrid recommendation engine that integrates collaborative filtering and knowledge reasoning:

[0152]

[0153] Among them

[0154] As an improvement, the hybrid recommendation engine includes an anomaly detection sub-module. When detecting:

[0155]

[0156] Start the manual review process at this time.

[0157] As an improvement, the user modeling module includes a cognitive diagnosis unit to calculate the knowledge mastery level:

[0158]

[0159] x i Are characteristics such as the answer correct rate and response time.

[0160] As an improvement, the interpretability generation module of the recommendation result, through the graph attention network:

[0161] e ij = LeakyReLU(a T [Wh i | |Wh j )

[0162] Generate recommendation reasons based on the knowledge path.

[0163] As an improvement, it stores a computer program, characterized in that when the program is executed by a processor, it implements the method steps described in any one of claims 1-5, wherein the resource encoding process adopts:

[0164] Denotes vector concatenation, BERT edu Is a pre-trained model in the education field.

[0165] 1. Multilayer knowledge graph construction method:

[0166] The core of the present invention lies in constructing a three-dimensional knowledge graph including a learning resource layer, a knowledge attribute layer, and an application mode layer. The specific implementation steps are as follows:

[0167] 1.1 Modeling of the learning resource layer:

[0168] Define the learning resource layer G r =(V r , E r ), where V r Represents the set of resource nodes, and each resource Is jointly encoded by the BERT pre-trained model in the education field and the graph attention network (GAT):

[0169]

[0170] Denotes the vector concatenation operation. The edge E r→k Represents the association between the resource and the knowledge attribute, and calculates the association strength through the trainable weight matrix :

[0171] Among them, σ is the Sigmoid function and θ is the predefined threshold.

[0172] 1.2 Mining of the knowledge attribute layer:

[0173] The knowledge attribute layer G k =(V k , E k ) is constructed by the frequent itemset mining algorithm. Define the set of knowledge attribute sequences S = {s1,..., s m}, and calculate the support degree:

[0174] Retain the frequent itemset X that meets the minimum support degree δ as the key knowledge attribute combination.

[0175] 1.3 Generation of the application pattern layer:

[0176] The application pattern layer G p =(V p , E p ) extracts the cross-layer association edges E cross :

[0177] Among them, η is the TF-IDF threshold, which reflects the semantic relevance between the knowledge attribute and the application pattern.

[0178] 2. User preference modeling and dynamic update:

[0179] 2.1 Encoding of user behavior sequences:

[0180] Based on the user behavior sequence S u =[e1,..., e T , a dual update mechanism is adopted to generate the preference vector M u :

[0181] Among them, the short-term preference update amount ΔM is calculated through the LSTM network:

[0182] ρ is the long-term preference decay factor.

[0183] 2.2 Enhancement of the attention mechanism:

[0184] Introduce the multi-head attention mechanism to optimize the user interest representation:

[0185] Among them is the user query vector, is the time series feature vector, and d is the dimension.

[0186] 3. Design of the hybrid recommendation algorithm:

[0187] 3.1 Calculation of the resource recommendation score:

[0188] Generate the recommendation score by integrating the user preference, the knowledge distribution difference, and the resource importance:

[0189] cos(M u ,V r ) is the cosine similarity of the user-resource vector;

[0190] Measure the difference in the distribution of knowledge mastery;

[0191] PageRank(r) reflects the structural importance of resources in the knowledge graph.

[0192] 3.2 Dynamic recommendation strategy switching:

[0193] Adjust the recommendation logic according to the learning stage:

[0194] New knowledge point stage Old knowledge point reinforcement stage

[0195]

[0196] This strategy monitors the changes in user behavior in real time through the trigger algorithm.

[0197] 4. Backup resource selection mechanism:

[0198] When the target resource r * is missing, select the backup resource r′:

[0199]

[0200] Among them is the similarity transformation matrix, which is jointly trained through collaborative filtering and knowledge graph embedding.

[0201] 5. Anomaly detection and optimization:

[0202] 5.1 Verification of the credibility of the recommendation result:

[0203] When the collaborative filtering score and the knowledge reasoning score differ by more than the threshold, trigger manual review:

[0204]

[0205] ∈ is the preset anomaly threshold.

[0206] 5.2 Generation of interpretable paths

[0207] Generate recommendation reasons based on the graph attention network:

[0208] e ij = LeakyReLU(a T [Wh i ||Wh j )

[0209] Through the attention weight α ij Identify the key knowledge path.

[0210] 6. System implementation architecture:

[0211] This system includes the following modules:

[0212] Knowledge graph construction module: Execute the above multi-layer graph construction algorithm;

[0213] User modeling module: Use a deep neural network to calculate user preferences (see Figure 1 );

[0214] Hybrid recommendation engine: Integrate collaborative filtering and knowledge reasoning, with the weight μ dynamically adjusted:

[0215] Among them

[0216] I. Precise recommendation and personalized adaptation:

[0217] Knowledge graph-driven multi-dimensional matching:

[0218] By constructing a three-dimensional knowledge graph of the resource layer, attribute layer, and application mode layer (

[0219] G r , G k , G p ), combined with the cosine similarity calculation of the user preference vector M u and the resource vector V r (cos(M u , V r )) to achieve double matching of semantic association and structural association, and solve the semantic limitations of traditional keyword filtering algorithms.

[0220] Effect verification: The actual measurement of an online education platform shows that the recommendation accuracy has increased by 23.7%.

[0221] Dynamic policy adaptive mechanism:

[0222] Adopt a phased recommendation strategy (switching between argmax / argmin) and dynamic adjustment of hybrid weights Automatically optimize the recommendation logic according to the user's learning stage (exploration of new knowledge points / strengthening of old knowledge points) to meet the needs of different scenarios.

[0223] II. Resource utilization rate and long-tail coverage:

[0224] Knowledge distribution difference compensation mechanism:

[0225] Through the KL divergence KL(p k ||q k)Quantify the difference between the user's knowledge mastery and the target distribution, guide the recommendation system to cover weak knowledge points, and improve the utilization rate of long-tail resources.

[0226] Measured data: The coverage rate of long-tail resources increased by 41.2%.

[0227] Substitute resource selection algorithm:

[0228] When the target resource r * is missing, based on the similarity transformation matrix W s to screen substitute resources

[0229] Ensure the integrity of the recommendation link.

[0230] III. Adaptive learning path generation:

[0231] Cognitive diagnosis and dynamic update:

[0232] Fuse the LSTM network (ΔM = LSTM([e t-n ,..., e t ; θ l )) and the attention mechanism to track the user behavior sequence S u in real time and generate a dynamic preference vector to support the dynamic correction of the learning path.

[0233] Quantitative evaluation of knowledge mastery:

[0234] Based on the logistic regression model (mastery(k) = 1 / (1 + exp(-(β0 + ∑β i x i ))), diagnose the user's knowledge gap and recommend targeted reinforcement resources.

[0235] IV. System efficiency and real-time performance:

[0236] Hybrid recommendation engine optimization:

[0237] Collaborative filtering and knowledge reasoning parallel computing architecture, combined with the anomaly detection threshold ∈ to achieve millisecond-level response (measured response time < 200ms).

[0238] Distributed knowledge graph storage:

[0239] Adopt the graph database sharding technology to store G r , G k , G p to support the efficient retrieval and correlation analysis of hundreds of millions of nodes.

[0240] V. Interpretability and Credibility:

[0241] Reason Generation Based on Graph Attention:

[0242] Identify key knowledge paths (such as "Algebra → Geometry → Calculus") through attention weights α ij = exp(e ij ) / ∑exp(e ik ) to provide visual recommended reasons.

[0243] Double Verification of Credibility:

[0244] The combination of the manual review process and the algorithm credibility assessment (the credibility of science and education resources tendency = f(user interest, credibility)) reduces the risk of mis-recommendation.

[0245] VI. Commercial Value and Scalability:

[0246] Multi-scenario Adaptability:

[0247] Support scenarios such as K12 education, vocational training, and in-company training, and quickly adapt to different business requirements by adjusting the frame text weight calculation formula Quickly adapt to different business requirements.

[0248] Ecosystem Compatibility

[0249] Can integrate third-party resource libraries (such as MOOCs platforms), and achieve cross-platform resource sharing through open APIs, in line with the trend of smart education ecosystem construction.

[0250] Comprehensive Benefit Data:

[0251] Indicator Improvement range Data source Recommended accuracy +23.7％ An online education platform Coverage rate of long-tail resources +41.2％ Same as above User satisfaction 92.3％ User research report System response time <200ms Load test report

[0252] Citation Note: The innovation points of this solution integrate technologies such as knowledge graphs, dynamic strategies, and cognitive diagnosis, which are in line with the cutting-edge direction of the education recommendation field.

[0253] The above describes the present invention and its implementation manners. Such a description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the spirit of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A method and system for sharing educational resources based on an educational cloud platform, characterized in that: It includes the following steps: S1. Construct a multi-layer knowledge graph, including the learning resource layer G r =(V r , E r ), the knowledge attribute layer G k =(V k , E k ), and the application mode layer G p =(V p , E p ), where: is a trainable weight matrix, and σ is the sigmoid function; S2. Based on the user behavior sequence S u = [e1,..., e T to generate a user preference vector: where φ(·) is a resource embedding function, and the Attention mechanism weight calculation formula is: Q u is the user query vector, and K t is the time series feature vector; 3. Calculate the resource recommendation score: score(r) = α·cos(M u , V r ) + β·KL(p k || q k ) + γ·PageRank(r) where the KL divergence measures the difference in knowledge distribution, and the PageRank value reflects the structural importance of resources in the knowledge graph; S4. Dynamically adjust the recommendation strategy according to the learning stage: New knowledge points, old knowledge points, reinforcement 2. The method and system for sharing educational resources based on an educational cloud platform according to claim 1, characterized in that: The construction of the multi-layer knowledge graph in the step S1 includes: Using a sequential pattern mining algorithm to extract frequent item sets: where X is a knowledge attribute combination, and δ is a minimum support threshold; Generate cross-layer associated edge E cross ={(k j , applied_in, p m ) | TF-IDF(k j , p m ) > η}.

3. The method and system for sharing educational resources based on an educational cloud platform according to claim 2, characterized in that: The generation of the user preference vector adopts a dual update mechanism: where the short-term preference update amount ΔM is calculated by an LSTM network: ΔM = LSTM([e t-n ,..., e t ; θ l ).

4. The method and system for sharing educational resources based on an educational cloud platform according to claim 3, characterized in that: The calculation formula for the knowledge distribution difference is: where p(k) is the user's knowledge mastery distribution, and q(k) is the target knowledge distribution.

5. The method and system for sharing educational resources based on an educational cloud platform according to claim 4, characterized in that: It also includes a resource substitution mechanism: when the target resource r * is missing, select to satisfy: Among them is the similarity transformation matrix.

6. The method and system for sharing educational resources based on an educational cloud platform according to claim 5, characterized in that: It further includes: A knowledge graph construction module that executes the multi-layer knowledge graph construction method described in claim 1; A user modeling module that uses a deep neural network to calculate the user preference vector: f θ (u) = ReLU(W2·ReLU(W1·h u + b1) + b2) h u is the convolutional feature of the user behavior sequence; A hybrid recommendation engine that integrates collaborative filtering and knowledge reasoning: Among them 7. The method and system for sharing educational resources based on an educational cloud platform according to claim 6, characterized in that: The hybrid recommendation engine includes an anomaly detection sub-module, when detecting: Start the manual review process.

8. The method and system for sharing educational resources based on an educational cloud platform according to claim 7, characterized in that: The user modeling module includes a cognitive diagnosis unit that calculates the knowledge mastery: x i such as the answering accuracy rate, response time, etc.

9. The method and system for sharing educational resources based on an educational cloud platform according to claim 8, characterized in that: An interpretability generation module for recommendation results, through a graph attention network: e ij = LeakyReLU(a T [Wh i ||Wh j ) Generate recommendation reasons based on knowledge paths.

10. The method and system for sharing educational resources based on an educational cloud platform according to claim 9, characterized in that: Stored with a computer program, characterized in that when the program is executed by a processor, it implements the method steps described in any one of claims 1-5, and the resource encoding process adopts: Denotes vector concatenation, BERT edu Is a pre-trained model for the education field.

Images