On How Networks Stabilize User Interest Based Methods and Vice Versa

摘要

This paper presents the evaluation of two graph-based recommendation methods compared to collaborative filtering as the baseline. The evaluation is primarily based on the investigation of the Average Receiver operating Characteristic curve on the MovieLens dataset. The presented methods operate on the knowledge graph, which information representation technique is also discussed in this paper. The evaluation results show that combining a network based and a user interest based method leads to a more stable performance and an increase in the recommendation quality.