Research On Online Course Recommendation Method Based On Knowledge Graph Enhancement

With the development of online education,the scale and application scope of online education in China has gradually climbed to the top in the world.While online courses provide people with convenient and fast learning methods,they are also prone to problems such as cognitive trek or difficulty in choosing courses due to excessive course resources.The use of personalized recommendation methods can help learners to automatically select exclusive courses suitable for their cognitive basis or learning interests from a large number of courses.At present,the application of course recommendation algorithms still has problems such as data sparseness,lack of knowledge and logical relationship,and insufficient interpretability.First,the problem of sparse data.The online education platform provides a large number of courses for the large learning group to choose.However,due to the small number of courses selected by learners,the interactive behavior data available for the recommendation model is very sparse,so the user interests mined are prone to bias Reduce the accuracy and stability of the recommended results.Second,the lack of knowledge and logic.Course recommendation based on the interests of learners does not consider the inherent logical relationship of different courses in the knowledge system.There are problems such as difficulty and disorder of the recommended courses.It is difficult for learners to plan the learning path according to the internal logic between courses.Therefore,it is necessary and meaningful to look for the user's interest path at the knowledge level when considering the semantic relationship between courses when recommending courses.Third,there is insufficient interpretability.In the recommendation scenario,giving proper interpretation of the recommendation results can improve the user's trust and acceptance.Most of the current mainstream recommendation methods fit the functional relationship between the learner and the course selection behavior through training,but because the obtained functional relationship does not have specific actual meanings for the parameters and variable dimensions,the interpretability is insufficient.In view of the above problems,this paper intends to fully exploit the interest of learners in the case of sparse data by integrating knowledge graph technology and deep recommendation models,using knowledge graphs to model the courses at the semantic level,and corresponding course collections to knowledge graphs to solve the logic relationship Lack of problems.Through knowledge graph-based linear feature mining,on the one hand,the knowledge graph information is used to alleviate the data sparsity problem,and on the other hand,the interest features that cannot correspond to specific influencing factors are mapped to specific entities in the knowledge graph,so that entities and relationships are used to visually display The learner's interest path enhances the interpretability of the recommendation.The specific work is as follows:(1)Construct a course knowledge map suitable for online course recommendationFind a knowledge graph data source suitable for online course recommendation from a large amount of open knowledge graph data,combine the course name and course content data crawled from the Internet,perform data processing through entity links,screening and other methods,using bottom-up extraction The subgraph and knowledge graph completion methods construct a course knowledge graph suitable for course recommendation.(2)Recommendation algorithm Ripple_mlp based on knowledge graph enhancementBy organically integrating knowledge graph information and deep collaborative filtering algorithm at the model level,a recommendation algorithm Ripple_mlp for knowledge graph enhancement is proposed.The algorithm uses the feature extraction method of multi-layer relationship diffusion and deep perceptron for dual network learning training,and finally obtains the recommendation probability.Experiments using real data show that the Ripple_mlp algorithm obtains higher accuracy of recommendation results than the benchmark algorithm and improves the interpretability of recommendation results.(3)Knowledge graph recommendation algorithm Ripple_mlp+based on co-actual body networkIn order to further improve the recommendation accuracy of the model,we recommend the special nature of "one-to-many" of the "course name-entity set" of course recommendation,and introduce a co-realistic network Co-net based on the Ripple_mlp algorithm to retroactively capture the user's Explicit interest features,so as to be able to better tap the interests of learners.Experiments using real data show that the Ripple_mlp+algorithm obtains the best accuracy of recommendation results on the basis of ensuring that the recommendation results are interpretable.In summary,this article uses entity linking and other methods to construct the course knowledge graph for recommendation,which depicts the semantic network relationship in the course recommendation field,alleviates the problem of data sparseness,and supplements the missing knowledge logical relationship;proposes a recommendation algorithm with enhanced knowledge graph Ripple_mlp improves the accuracy of recommendation results in the case of sparse data,and solves the problem of lack of interpretability of recommendations;proposes a knowledge graph recommendation algorithm Ripple_mlp+based on a co-actual body network to capture explicit user interest,and the algorithm ensures that the recommendation has interpretability On the basis of the nature,the accuracy of the recommendation results in the case of sparse data is further improved.The research has significantly improved the recommendation effect of the course recommendation algorithm and laid the foundation for the practical application of course recommendation.