User Preference-Based Probability Spreading for Tag-Aware Content Recommendation
Jan O. Friedrich, Christoph Lindemann, Michael Petrifke · 2017
In this paper, we show how to integrate user-item scoring into a graph-based tag-aware item recommender system. Building upon the ProbS and PLIERS methods, we introduce refined formulas for affinity and similarity scoring taking into account user-item preference in the mass diffusion of recommender systems. Additionally, we propose a two-step similarity score that recommends items based on a repeated mass diffusion on the item-tag graph. We denote the proposed method as User Preference-based Probability Spreading for content recommendation, UPPS. UPPS relies on the notion that the influence of current user items on the recommendation process should depend on the user item preference. To evaluate the proposed approach, we employ the well-known MovieLens dataset. In comparison to ProbS and PLIERS, UPPS yields an improvement in both the NDCG@10 and P@10 measures by more than 25% over ProbS and more than 100% over PLIERS.