Social personalized ranking with both the explicit and implicit influence of user trust and of item ratings

Due to the inherent deficiency of social collaborative filtering algorithms based on rating prediction, social personalized ranking algorithms based on ranking prediction have recently received much more attention in recommendation communities due to their close relationship with real industry problem settings. However, most existing social personalized ranking algorithms focus on either explicit feedback data or implicit feedback data rather than making full use of the information in the dataset. Until now, no studies have been done on social personalized ranking algorithms by exploiting both the explicit and implicit influence of user trust and of item ratings. In order to overcome the defects of prior researches and to further solve the problems of data sparsity and cold start of collaborative filtering, a new social personalized ranking model (SPR_SVD++) based on the newest xCLiMF model and TrustSVD model was proposed, which exploited both the explicit and implicit influence of user trust and of item ratings simultaneously and optimized the well-known evaluation metric Expected Reciprocal Rank (ERR) Experimental results on practical datasets showed that our proposed model outperformed existing state-of-the-art collaborative filtering approaches over two different evaluation metrics NDCG and ERR, and that the running time of SPR_SVD++ showed a linear correlation with the number of users in the data collection and the number of observations in the rating and trust matrices. Due to its high precision and good expansibility, SPR_SVD++ is suitable for processing big data and has wide application prospects in the field of internet information recommendation.