Learning Peer Recommendation Based On Heterogeneous Information Networks On Online Learning Platforms

With the development of online education,problems such as insufficient course completion rates and students’ poor learning effects on online learning platforms have become increasingly prominent.In order to better guide and motivate students to complete online courses and improve the quality of learning,this thesis proposes two learning peer recommendation algorithms based on the behavior data and exercises test data collected in the learning process.The main tasks are as follows:1.In view of the fact that the existing learning peer recommendation algorithm ignores the heterogeneity of various types of objects(students,teachers,videos,exercises,and knowledge points)and the relationships between objects that appear in the online learning process,learning peer recommendation based on heterogeneous information network representation learning and deep learning is proposed.The algorithm first uses heterogeneous information networks to integrate the five different types of objects and the relationship information between them,and then retains comprehensive semantic and structural information.Secondly,in order to mine the correlation between auxiliary information and students’ preferences in the network,a network-based representation learning and deep learning model are proposed,which combine multi-layer perceptron and network representation learning,and use MLP to perform representation and matching function learning of target students and candidate students to solve the problem of the limited expression ability of the dot product of the matrix factorization model and the insufficient ability to capture the low-rank relationship.This paper compares DMF,Metapath,and MF algorithms on real-world datasets.The experimental results show that our method has the best learning peer recommendation effect.2.Considering that the above learning peer recommendation algorithm needs to manually specify the meta-path and ignores the influence of the attribute value on the link on the recommendation result,we propose a learning peer recommendation based on weighted heterogeneous information networks on online learning platforms.The algorithm first constructs a weighted heterogeneous information network on an online learning platform to integrate different types of objects,relationship information and attribute value information on links,and retain semantic,structural,and link attribute value information more comprehensively.Secondly,according to the network schema of the constructed information network,a method for automatically generating meaningful meta-paths is proposed,which is flexible and effective for automatically generating and identifying all meaningful meta-paths for learning peer recommendation.Finally,the personalized weight of the learning meta-path in the Bayesian optimization framework is studied,and calculate the recommendation score.According to the recommendation score,the student can be recommended learning peers that meet their own needs.Experimental results show that our method outperforms PRHN,Metapath and MF algorithms in terms of precision and recall.