【摘要】 在课程推荐领域,通常会遇到数据稀疏性和冷启动问题,导致推荐效果不理想。为此,基于端到端深度学习框架,提出一种融合课程知识图谱的深度卷积神经网络(KGCN-CR)。通过聚集课程实体邻域信息增强自身实体表示,获取学生个性化潜在兴趣。以慕课(MOOC)平台为例,通过爬取计算机类和艺术类学生的课程交互数据和课程属性,构建课程知识图谱作为辅助信息增强课程推荐的性能,分别使用18 135条交互数据以及44 600条课程属性进行试验。结果表明,KGCN-CR的ACC以及AUC分别达到了82.3%和78.2%,比SVD提升15%,精确率、召回率以及F1值也最优。因此,知识图谱作为辅助信息能有效提升课程推荐的性能,能较好解决数据稀疏性以及冷启动问题,并具有较好的推荐可解释性。更多还原

【Abstract】 Data sparsity and cold start are the problems in recommendation system,which will lead to bad recommendation effect.Therefore,based on the end-to-end deep learning framework,propose a deep convolutional neural network(Course Knowledge Graph Convolutional Networks,KGCN-CR)that integrates course knowledge graphs(KG),KG enhances entity representation by gathering information about the neighborhood of course entities,which,the individualized potential interests of students are obtained. As an example,students and course interaction data and course attributes of the major in Computer Science and Art are crawled on the MOOC platform. The course attributes data crawled is aimed to be constructed a course KG. The course KG as the auxiliary information enhance the performance of course recommendation,by using 18,135 pieces of interaction data respectively and 44,600 course attribute for experimentation. The experiment results shows that the ACC and AUC of KGCN-CR are reached 82.3% and 78.2%,respectively,which are 15% higher than SVD,and the accuracy rate,recall rate and F1 value are also optimal. Even when the interactions between students and the coursesare extremely sparse,the good performance is maintained. The KG as an auxiliary information can effectively improve the performance of online courses recommendation,can solve the problem of data sparsity and cold start,and has better recommendation interpretability.更多还原