基于知识图谱的MOOC课程推荐

Knowledge Graph Based Course Recommendation

传统推荐方法主要采用用户之间的共同项目计算相似度,忽略课程推荐中隐含较多课程语义信息。为充分利用课程之间的语义信息,发掘用户的个性化兴趣,提出一种基于知识图谱的课程推荐。首先利用课程内容信息和用户行为记录构建知识图谱,通过课程知识图谱计算课程之间的相关性;再将课程实体与邻域信息选择性聚合来增强自身表示,发掘用户的潜在兴趣;最终通过语义信息深度挖掘课程之间的关系,利用图卷积网络来预测用户对课程的评分。结果表明,基于知识图谱的MOOC课程推荐与基线模型相比,在精确率、召回率、f1指标上分别提高3.33%、4.8%、4.68%。因此本文提出的方法在一定程度上能提升推荐效果,同时缓解数据稀疏和冷启动问题。

The traditional recommendation method mainly uses the common items between users to calculate the similarity,which ignores the implicit semantic information of the course in the course recommendation.In order to make full use of semantic information between the courses and explore users′personalized interests,a course recommendation based on the knowledge map was proposed.First,the knowledge map was constructed by using the course content information and user behavior records,and the correlation between courses was calculated by the course knowledge map and then selectively aggregated the course entities and neighborhood information to enhance their own representation and explored the potential interests of users.Finally,the semantic information was used to deeply mine the relationship between the courses,and the graph convolution network was used to predict the user′s grading of the courses.The results showed that compared with the baseline model,the MOOC curriculum recommendation based on knowledge map had improved 3.33%,4.8%and 4.68%in accuracy,recall and f1 indicators respectively.Therefore,the method can improve the recommendation effect to a certain extent,and at the same time alleviate the data sparsity and cold start problems.