Video learning resource path recommendation method and system based on knowledge graph and attention mechanism

Abstract

The invention provides a video learning resource path recommendation method and system based on a knowledge graph and an attention mechanism, and the method comprises the steps: building a learner interest model, a learner social model and a learner style model according to learner information and interaction information, and carrying out the fusion of the models into a learner model according to the attention mechanism; embedding the learner model into the knowledge graph, and querying all learning paths according to an optimized IDDFS algorithm; constructing a learning resource model according to the learning resource information; calculating a learning resource score according to a Bayesian algorithm; the learning resource scores and the learning paths are added and fused to obtain learning path scores, and the first N learning paths are recommended to the learner; according to the method, on the aspect of learning resource recommendation, feature modeling of learners and learning resources is considered, and personalized recommendation is achieved. In learner modeling, the situation of multi-dimensional learner modeling is considered, and a learner interest model, a learner social model and a learner style model are fused together to form a learner model.

Description

Technical Field

[0001] The present invention relates to the field of recommendation systems in computer technology, and in particular to a method and device for recommending video learning resource paths based on knowledge graphs and attention mechanisms. Background Art

[0002] In the process of learning path recommendation, accurate learner modeling and learning resource modeling are the basis and prerequisite for achieving high-quality recommendations. Traditional learner models usually rely solely on the basic information of learners for modeling. Although this method provides a certain basis for personalized recommendation, it often ignores the importance of comprehensive modeling of learners from multiple dimensions. In fact, factors such as learners' interests, learning styles, and learners' sociality have an important impact on the recommendation of learning paths. Only by fully understanding learners from these different dimensions can we provide more accurate and personalized learning path recommendations, so as to better meet learners' needs and improve learning effects. In addition, traditional learning resource modeling methods mostly focus on popular recommendations and unpopular recommendations, that is, recommendations are made only based on the popularity of resources. This method may cause some learning resources that have potential but have not yet been widely discovered to be neglected. This not only limits learners' access to high-quality learning resources, but also may miss the learning improvement opportunities brought by personalized recommendations. Therefore, in learning resource modeling, a more refined mechanism needs to be introduced to identify and recommend those resources that have potential but have not yet been widely accepted. By comprehensively considering learners' multi-dimensional information and the potential of learning resources, more accurate and personalized recommendations can be provided to promote learners' learning outcomes and learning experience.

[0003] The Chinese patent "Method and system for automatic generation of personalized learning process based on knowledge graph" with publication number CN106777127A focuses on the personalized processing of process models and the detection of changes in resource environment and learners themselves, while the Chinese patent "Method and system for recommending learning resources based on learner preferences and group preferences" with publication number CN114896512A focuses on learners' personal long-term and short-term preferences as well as learners' group preferences. Summary of the invention

[0004] To solve the above problems, the present invention proposes a video learning resource path recommendation method and system based on knowledge graph and attention mechanism. Learners are modeled through self-attention mechanism, combined with knowledge graph technology, and an optimized IDDFS algorithm is used to recommend a learning path that is more suitable for learners. At the same time, learning resources are modeled, and a Bayesian algorithm is used to obtain a learning resource score, thereby obtaining a learning path score, and recommending a learning path that is more suitable for learners, ultimately achieving the purpose of improving the learning quality and speed of learners.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a video learning resource path recommendation method based on knowledge graph and attention mechanism, comprising the following steps: Capture learners’ interests based on learner information and interaction information, and build a learner interest model; identify learners’ social networks based on learner information and interaction information, and build a learner social model; capture learners’ learning styles based on the improved Field Silver learning style model, and build a learner style model; integrate learner interest model, learner social model and learner style model, and combine them with attention mechanism to obtain a learner model; According to the learning resource information, a knowledge graph of learning resources is established based on neo4j, and the learner model is embedded in the knowledge graph. By integrating the learner's learning interests, learning social interactions, and learning styles, a comprehensive learner network knowledge graph is formed; The learner network knowledge graph is searched according to the optimized IDDFS algorithm to obtain all possible learning paths in the learner network knowledge graph; According to the learning resource information, the heat decay model is introduced to screen out potential learning resources, and the long tail theory is introduced to screen out currently unpopular learning resources. Combining the heat decay model and the long tail theory, a comprehensive learning resource model is constructed; The learning resource model is predicted and scored according to the Bayesian algorithm to obtain the score of each learning resource; Add the learning resource score to all possible learning paths to get the score of each learning path. All the obtained learning paths are sorted in descending order according to the scores, and the top N learning paths are recommended to the learners.

[0006] Furthermore, learner information includes learner name or ID, learner region, learner mastered nodes and learner target nodes; learning resource information includes learning resource title, learning resource duration and learning resource upload time; interaction information includes learner views of learning resources, viewing completion rate, number of comments, comment relevance and learner direct rating of learning resources, among which learning resource rating is explicit feedback and the rest is implicit feedback.

[0007] Furthermore, when constructing a learner model, other learners who have watched the same learning resources as the learner are found, and the learner social model is constructed based on the number of views, viewing completion rate, number of comments, comment relevance, region, learning resource length, and learning resource rating of other learners; The learning style model was constructed based on the improved FSLSM. The learning style model includes eight perspectives: active / reflective, sensing / intuitive, chart / literal, and sequential / global. Combined with the attention mechanism, the learner interest model, learner social model and learner style model are fused into the learner model.

[0008] Furthermore, the IDDFS algorithm is optimized by introducing the pruning strategy into the iterative deepening depth-first algorithm. The iterative deepening depth-first search algorithm is used to search from the mastered nodes layer by layer with increasing depth until the target node is found or the maximum depth limit is reached. During the search process, the repeated state pruning method is used. When searching a certain layer, it is necessary to check whether the current node is a repeated state. If it is a repeated state, the search will not continue downward, and it will go back to the previous layer to continue exploring other branches. In each layer of the iterative deepening depth-first algorithm, repeated state pruning judgments are performed, and the iterative deepening search continues until the target node is found or the maximum depth limit is reached, and finally all possible learning paths are obtained.

[0009] Furthermore, building a comprehensive learning resource model includes: Introducing a heat decay model, through the decay analysis of the heat of resources, the learning resources screened by the heat decay model are regarded as resources with potential value; Introducing the long tail theory, identifying currently unpopular or less well-known learning resources based on interactive information; By introducing the attention mechanism, two different types of learning resources are fused, and learning resources that have not yet received widespread attention but have potential are identified and constructed into a learning resource model.

[0010] Furthermore, the learning resource model is predicted and scored according to the Bayesian algorithm, and the scores of each learning resource include: Establish a user-item rating matrix based on the interaction information; Calculate the probability of learning resource rating without any conditions, that is, the prior probability of learning resource rating, as follows:

[0011] Where r is a possible rating value; Apply Bayes' theorem to calculate the conditional probability that, given a learner U and a prior learning resource I, the score R UI The probability is as follows:

[0012] in, is the joint probability of learner U and learning resource I under rating r, is the prior probability of rating r, is the marginal probability of learner U and learning resource I; Predict the learner's rating of the learning resource based on the posterior probability, and select the rating with the highest probability as the predicted rating. The method is as follows: .

[0013] Furthermore, the loss function is constructed based on the convolutional neural network, and the learner modeling is optimized including: Use the convolutional layer to extract the features of learners and learning resources, construct a loss function and calculate the loss based on the output of the feature extraction layer, use the optimization algorithm to minimize the loss function, and update the weights of the convolutional neural network through the back-propagation algorithm.

[0014] Based on the concept of the method, the present invention provides a video learning resource path recommendation system based on knowledge graph and attention mechanism, including a learner model construction module, a learner network construction module, a recommended path acquisition module, a learning resource model construction module, a path score acquisition module and a recommendation module; The learner model building module is used to capture learner interests based on learner information and interaction information and build a learner interest model; identify learner social networks based on learner information and interaction information and build a learner social model; capture learner learning styles based on the improved Field Silver learning style model and build a learner style model; integrate the learner interest model, learner social model and learner style model, and combine them with the attention mechanism to obtain a learner model; The learner network construction module is used to build a knowledge graph of learning resources based on neo4j according to the learning resource information, embed the learner model into the knowledge graph, and form a comprehensive learner network by integrating the learner's learning interests, learning social interactions and learning styles; The recommended path acquisition module is used to retrieve the learner network knowledge graph according to the optimized IDDFS algorithm to obtain all possible learning paths in the learner network knowledge graph; The learning resource model construction module is used to introduce the heat decay model based on the learning resource information to screen out potential learning resources, introduce the long tail theory to screen out currently unpopular learning resources, and combine the heat decay model and the long tail theory to build a comprehensive learning resource model; The path score acquisition module is used to predict and score the learning resource model according to the Bayesian algorithm to obtain the score of each learning resource; Add the learning resource score to all possible learning paths to get the score of each path; The recommendation module is used to sort all the obtained learning paths in descending order according to the scores and recommend the top N learning paths to the learners.

[0015] Compared with the existing recommendation system, the present invention has at least the following advantages: The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the above-mentioned video learning resource path recommendation method based on knowledge graph and attention mechanism.

[0016] At the same time, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the above-mentioned video learning resource path recommendation method based on knowledge graph and attention mechanism can be implemented.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: the present invention no longer focuses on the simple recommendation of learning resources, but puts the focus on the recommendation of learning paths. In the recommendation of learning resources, the present invention takes into account the feature modeling of learners and learning resources, and realizes personalized recommendations. In learner modeling, the multi-dimensional learner modeling is considered, and the learner interest model, learner social model and learner style model are integrated together to form a learner model. At the same time, the learner model is mapped to the knowledge graph, and then all learning paths are obtained according to the optimized IDDFS algorithm, so that the recommendation of learning paths is more accurate. In addition, when modeling learning resources, it is considered to recommend unpopular but potential learning resources to learners, which effectively avoids information overload and recommendation fatigue, and improves the learner's recommendation experience.

[0018] Furthermore, the present invention combines explicit feedback with implicit feedback, which can improve the diversity and coverage of recommendations while improving personalization accuracy, thereby optimizing the learner experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 Flowchart of a video learning resource path recommendation method based on knowledge graph and attention mechanism. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] In the description of the present invention, it should be understood that the terms “include” and “comprises” indicate the presence of described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0022] Figure 1 This is a flow chart of a method for recommending video learning resource paths based on knowledge graph and attention mechanism. The implementation methods of the present invention are now described in detail.

[0023] Step 1, collect learner information, interaction information and learning resource information; the collected learner information includes the learner's name or ID, the learner's region, the learner's mastered nodes and the learner's target nodes; the learning resource information includes the learning resource title, learning resource duration and the upload time of the learning resource; the interaction information includes the number of times the learner has viewed the learning resource, the viewing completion rate, the number of comments, the relevance of the comments and the learner's direct rating of the learning resource. Among them, the learning resource rating is explicit feedback, and the rest is implicit feedback. Combining explicit feedback with implicit feedback can improve the diversity and coverage of recommendations while improving personalization accuracy, and optimize the learner experience; Step 2: Capture learners’ interests based on the collected learner information and interaction information, build a learner interest model, identify learners’ social networks based on learner information and interaction information, build a learner social model, capture learners’ learning styles based on the improved Field-Silver learning style model, build a learner style model, and combine these three models to obtain a comprehensive learner portrait, and finally obtain a complete learner model; Step 2.1: Based on the number of views, viewing completion rate, number of comments, comment relevance, learning resource duration, and learning resource rating, and in order to solve the system cold start problem, the student’s region factor is introduced to construct a learner interest model; here, 0-10 minutes is considered a short video, and more than 10 minutes is considered a long video; Step 2.2, find other learners who have watched the same learning resources as the learner, and build a learner social model based on the number of views, viewing completion rate, number of comments, comment relevance, region, learning resource length, and learning resource rating of other learners; Step 2.3, construct a learner style model based on the improved FSLSM, including eight perspectives: active / reflective, sensing / intuitive, chart / literal, sequential / global; Step 2.4, based on the attention mechanism, the learner interest model, the learner social model and the learner style model are integrated into the learner model; Get the feature vector F of the learner's interest model interest , the feature vector F of the learner social model social and the feature vector F of the learner style model style , each eigenvector is mapped to a common space according to a linear transformation, as follows:

[0024] Among them, W interest , W social , W style is the weight matrix obtained through training; The attention scores of the learner interest model, learner social model, and learner style model are calculated according to the dot product method and activation function. The specific methods are as follows:

[0025] Where Q is the query vector and f is the activation function; Each attention score is converted into a weight through the softmax function so that its sum is 1. The specific method is as follows: ; According to the calculated weights, the final features are fused by re-weighting the features. The fused features F final A new learner model was formed, and the specific method is as follows:

[0026] Among them, w interest , w social , w style They are the weights calculated by the softmax function previously; Step 3: Based on the learning resource information, a knowledge graph of learning resources is established based on neo4j, and the learner model is embedded in the knowledge graph. By integrating the learner's learning interests, learning social interactions, and learning styles, a comprehensive visualization of the learner's individual characteristics is achieved to form a comprehensive learner network knowledge graph; Step 4, search the learner network knowledge graph according to the optimized IDDFS algorithm to obtain all possible learning paths in the learner network knowledge graph; optimizing the IDDFS algorithm is to introduce the pruning strategy into the iterative deepening depth first (IDDFS) algorithm. Using the iterative deepening depth first search (IDDFS) algorithm, start from the mastered node and increase the depth layer by layer until the target node is found or the maximum depth limit is reached. During the search process, the repeated state pruning method is used to avoid repeated visits to the state nodes that have been explored. When searching a certain layer, it is necessary to check whether the current node is a repeated state. If it is a repeated state, do not continue to search downward, and go back to the previous layer to continue exploring other branches. Duplicate state pruning must be judged at each layer of IDDFS to ensure the efficiency and accuracy of the search. Continue to iterate and deepen the search until the target node is found or the maximum depth limit is reached, and finally all learning paths are obtained; Step 5: Based on the learning resource information, the heat decay model is introduced to screen out potential learning resources, and the long tail theory is introduced to screen out currently unpopular learning resources. Combining the heat decay model and the long tail theory, a comprehensive learning resource model is constructed; Step 5.1, introduce the heat decay model. By performing a decay analysis on the heat of these resources, we can accurately capture those learning resources that have been widely watched and praised in the past, although their current popularity has declined. These learning resources screened by the heat decay model are regarded as resources with potential value. The method is as follows:

[0027] Among them, A is the initial value, λ is the decay rate; Step 5.2: Introduce the long tail theory to obtain currently unpopular learning resources; In step 5.3, based on the attention mechanism, the two learning resources are integrated to obtain a learning resource that is not popular yet but has great potential, and then constructed into a learning resource model.

[0028] Step 6: Use the Bayesian algorithm to predict and score the learning resource model to obtain the score of each learning resource. The method is as follows:

[0029] Where P(i) is the prior probability of event i, and P(i|j) is the probability of event i occurring given condition j; Step 6.1, establish a user-item rating matrix based on the interaction information; Step 6.2, based on each learning resource score in the user-item rating matrix, calculate the probability of the learning resource score without any conditions, that is, the prior probability of the learning resource score, as follows:

[0030] Where r is a possible rating value; Step 6.3, based on the ratings of each learner on the learning resource in the user-item rating matrix, apply Bayes’ theorem to calculate the conditional probability, that is, given the learner U and the first learning resource I, the rating R UI The method is as follows:

[0031] in, is the joint probability of learner U and learning resource I under rating r, is the prior probability of rating r, is the marginal probability of learner U and learning resource I; Step 6.4, predict the learner's rating of the learning resource based on the posterior probability, and select the rating with the highest probability as the predicted rating, as follows: ; Step 7: Add the learning resource score to all the obtained learning paths to obtain the score of each learning path; Step 8: sort all the obtained learning paths in descending order according to the scores, and recommend the top N learning paths to the learners; Step 9: Build a loss function based on the convolutional neural network to optimize the recommendation model. First, use the convolutional layer to extract the features of the learner and the learning resource, then build a loss function to calculate the loss based on the output of the feature extraction layer, and finally use the optimization algorithm to minimize the loss function. Update the weights of the convolutional neural network through the back propagation algorithm. The method is as follows:

[0032] Where N is the total number of samples, is the actual value of the ith sample, is the predicted value of the ith sample.

[0033] The present invention also provides a learning path recommendation system, comprising a learner model construction module, a learner network construction module, a recommended path acquisition module, a learning resource model construction module, a path score acquisition module and a recommendation module; The learner model building module is used to capture learner interests based on learner information and interaction information and build a learner interest model; identify learner social networks based on learner information and interaction information and build a learner social model; capture learner learning styles based on the improved Field Silver learning style model and build a learner style model; integrate the learner interest model, learner social model and learner style model, and combine them with the attention mechanism to obtain a learner model; The learner network construction module is used to build a knowledge graph of learning resources based on neo4j according to the learning resource information, embed the learner model into the knowledge graph, and form a comprehensive learner network by integrating the learner's learning interests, learning social interactions and learning styles; The recommended path acquisition module is used to retrieve the learner network knowledge graph according to the optimized IDDFS algorithm to obtain all possible learning paths in the learner network knowledge graph; The learning resource model construction module is used to introduce the heat decay model based on the learning resource information to screen out potential learning resources, introduce the long tail theory to screen out currently unpopular learning resources, and combine the heat decay model and the long tail theory to build a comprehensive learning resource model; The path score acquisition module is used to predict and score the learning resource model according to the Bayesian algorithm to obtain the score of each learning resource; Add the learning resource score to all possible learning paths to get the score of each learning path. The recommendation module is used to sort all the obtained learning paths in descending order according to the scores and recommend the top N learning paths to the learners.

[0034] In summary, the method of the present invention captures learners' interests according to learner information and interaction information, and constructs a learner interest model; identifies learners' social networks according to learner information and interaction information, and constructs a learner social model; captures learners' learning styles according to the improved Field Silver learning style model, and constructs a learner style model; integrates learner interest models, learner social models, and learner style models, and obtains a learner model by combining attention mechanism; establishes a knowledge graph of learning resources based on neo4j according to learning resource information, embeds the learner model into the knowledge graph, and forms a comprehensive learner network knowledge graph by integrating learners' learning interests, learning social interactions, and learning styles; retrieves the learner network knowledge graph according to the optimized IDDFS algorithm, and obtains all possible learning paths in the learner network knowledge graph; and The source information introduces a heat decay model to screen out potential learning resources, and introduces the long tail theory to screen out currently unpopular learning resources. The heat decay model and the long tail theory are combined to construct a comprehensive learning resource model; the learning resource model is predicted and scored according to the Bayesian algorithm to obtain a score for each learning resource; the learning resource score is added and fused with all possible learning paths obtained to obtain a score for each learning path; all the obtained learning paths are arranged in descending order according to the score, and the top N learning paths are recommended to the learner; compared with the prior art, the present invention not only focuses on the learner's interest, social and style models, but also focuses on learning resource modeling for resources that have potential but have not yet been widely accepted, thereby increasing the diversity of recommendations; regional factors are taken into account when modeling learners, and recommendations are made based on learners in the same region when learners first learn, thereby avoiding the problem of cold start of recommendations.

[0035] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads part or all of the computer executable program from the memory and executes it, and when the processor executes part or all of the computer executable program, it can implement the research on the video learning resource path recommendation method based on knowledge graph and attention mechanism described in the present invention.

[0036] On the other hand, the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can implement the video learning resource path recommendation method based on knowledge graph and attention mechanism described in the present invention.

[0037] The computer device may be a laptop computer, a desktop computer or a workstation.

[0038] The processor of the present invention may be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC) or a readily available field programmable gate array (FPGA).

[0039] The memory described in the present invention may be an internal storage unit of a notebook computer, a desktop computer or a workstation, such as a memory or a hard disk; or an external storage unit, such as a mobile hard disk or a flash memory card.

[0040] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0041] The above description is only a preferred specific implementation manner 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 video learning resource path recommendation method based on knowledge graph and attention mechanism, characterized in that: The following steps are involved: Capture learners' interests based on learner information and interaction information, and build a learner interest model; identify learners' social networks based on learner information and interaction information, and build a learner social model; capture learners' learning styles based on the improved Field Silver learning style model, and build a learner style model; integrate learner interest models, learner social models, and learner style models, and combine them with attention mechanisms to obtain a learner model; According to the learning resource information, a knowledge graph of learning resources is established based on neo4j, and the learner model is embedded in the knowledge graph. By integrating the learner's learning interests, learning social interactions, and learning styles, a comprehensive learner network knowledge graph is formed; The learner network knowledge graph is retrieved according to the optimized IDDFS algorithm to obtain all possible learning paths in the learner network knowledge graph; According to the learning resource information, the heat decay model is introduced to screen out potential learning resources, and the long tail theory is introduced to screen out currently unpopular learning resources. Combining the heat decay model and the long tail theory, a comprehensive learning resource model is constructed; The learning resource model is predicted and scored according to the Bayesian algorithm to obtain the score of each learning resource; Add the learning resource score to all possible learning paths to get the score of each learning path. All the obtained learning paths are sorted in descending order according to the scores, and the top N learning paths are recommended to the learners.

2. The video learning resource path recommendation method based on knowledge graph and attention mechanism according to claim 1 is characterized in that: Learner information includes the learner's name or ID, the learner's region, the learner's mastered nodes and the learner's target nodes; learning resource information includes the learning resource title, learning resource duration and the upload time of the learning resource; interaction information includes the number of times the learner has viewed the learning resource, the viewing completion rate, the number of comments, the relevance of the comments and the learner's direct rating of the learning resource, among which the learning resource rating is explicit feedback and the rest is implicit feedback.

3. The video learning resource path recommendation method based on knowledge graph and attention mechanism according to claim 1 is characterized in that: When constructing the learner interest model, the factors of the students' regions are introduced to construct the learner interest model based on the number of views, viewing completion rate, number of comments, comment relevance, learning resource length and learning resource rating; When building a learner social model, find other learners who have watched the same learning resources as the learner, and build a learner social model based on the number of views, viewing completion rate, number of comments, comment relevance, region, learning resource length, and learning resource rating of other learners; The learner style model is constructed based on the improved FSLSM. The learning style model includes eight perspectives: active / reflective, sensing / intuitive, chart / literal, sequential / global. Combined with the attention mechanism, the learner interest model, learner social model and learner style model are fused into the learner model.

4. The video learning resource path recommendation method based on knowledge graph and attention mechanism according to claim 1 is characterized in that: The optimization of the IDDFS algorithm is to introduce the pruning strategy into the iterative deepening depth-first algorithm. The iterative deepening depth-first search algorithm is used to search from the mastered node layer by layer with increasing depth until the target node is found or the maximum depth limit is reached. During the search process, the repeated state pruning method is used. When searching a certain layer, it is necessary to check whether the current node is a repeated state. If it is a repeated state, the search will no longer continue downward, and it will go back to the previous layer to continue exploring other branches. At each layer of the iterative deepening depth-first algorithm, repeated state pruning judgment must be performed, and the iterative deepening search will continue until the target node is found or the maximum depth limit is reached, and finally all possible learning paths are obtained.

5. The video learning resource path recommendation method based on knowledge graph and attention mechanism according to claim 1 is characterized in that: Building a comprehensive learning resource model includes: Introducing a heat decay model, through the decay analysis of the heat of resources, the learning resources screened by the heat decay model are regarded as resources with potential value; Introducing the long tail theory, identifying currently unpopular or less well-known learning resources based on interactive information; By introducing the attention mechanism, two different types of learning resources are fused, and potential learning resources that have not yet received widespread attention are identified and constructed into a learning resource model.

6. The video learning resource path recommendation method based on knowledge graph and attention mechanism according to claim 1 is characterized in that: The learning resource model is predicted and scored according to the Bayesian algorithm, and the scores of each learning resource include: Establish a user-item rating matrix based on the interaction information; According to each learning resource score in the user-item rating matrix, the probability of the learning resource score without any conditions is calculated, that is, the prior probability of the learning resource score, as follows: Where r is a possible rating value; Apply Bayes' theorem to calculate the conditional probability that, given a learner U and a prior learning resource I, the score R UI The probability is as follows: in, is the joint probability of learner U and learning resource I under rating r, is the prior probability of rating r, is the marginal probability of learner U and learning resource I; Predict the learner's rating of the learning resource based on the posterior probability, and select the rating with the highest probability as the predicted rating. The method is as follows: 。 7. The video learning resource path recommendation method based on knowledge graph and attention mechanism according to claim 1 is characterized in that: According to the convolutional neural network, the loss function is constructed and the learner model is optimized including: Use the convolutional layer to extract the features of learners and learning resources, construct a loss function and calculate the loss based on the output of the feature extraction layer, use the optimization algorithm to minimize the loss function, and update the weights of the convolutional neural network through the back-propagation algorithm.

8. A learning path recommendation system, characterized in that: It includes a learner model building module, a learner network building module, a recommended path acquisition module, a learning resource model building module, a path score acquisition module and a recommendation module; The learner model building module is used to capture learner interests based on learner information and interaction information and build a learner interest model; identify learner social networks based on learner information and interaction information and build a learner social model; capture learner learning styles based on the improved Field Silver learning style model and build a learner style model; integrate the learner interest model, learner social model and learner style model, and combine them with the attention mechanism to obtain a learner model; The learner network construction module is used to build a knowledge graph of learning resources based on neo4j according to the learning resource information, embed the learner model into the knowledge graph, and form a comprehensive learner network by integrating the learner's learning interests, learning social interactions and learning styles; The recommended path acquisition module is used to retrieve the learner network knowledge graph according to the optimized IDDFS algorithm to obtain all possible learning paths in the learner network knowledge graph; The learning resource model construction module is used to introduce the heat decay model based on the learning resource information to screen out potential learning resources, introduce the long tail theory to screen out currently unpopular learning resources, and combine the heat decay model and the long tail theory to build a comprehensive learning resource model; The path score acquisition module is used to predict and score the learning resource model according to the Bayesian algorithm to obtain the score of each learning resource; Add the learning resource score to all possible learning paths to get the score of each learning path. The recommendation module is used to sort all the obtained learning paths in descending order according to the scores and recommend the top N learning paths to the learners.

9. A computer device, characterized in that: It includes a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, it can implement the video learning resource path recommendation method based on knowledge graph and attention mechanism as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A computer program is stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the video learning resource path recommendation method based on knowledge graph and attention mechanism as described in any one of claims 1 to 7.

Images