Bayesian Personalized Ranking with Multi-Channel User Feedback

Abstract

Pairwise learning-to-rank algorithms have been shown to allow recommender systems to leverage unary user feedback. We propose Multi-feedback Bayesian Personalized Ranking (MF-BPR), a pairwise method that exploits different types of feedback with an extended sampling method. The feedback types are drawn from different "channels", in which users interact with items (e.g., clicks, likes, listens, follows, and purchases). We build on the insight that different kinds of feedback, e.g., a click versus a like, reflect different levels of commitment or preference. Our approach differs from previous work in that it exploits multiple sources of feedback simultaneously during the training process. The novelty of MF-BPR is an extended sampling method that equates feedback sources with "levels" that reflect the expected contribution of the signal. We demonstrate the effectiveness of our approach with a series of experiments carried out on three datasets containing multiple types of feedback. Our experimental results demonstrate that with a right sampling method, MF-BPR outperforms BPR in terms of accuracy. We find that the advantage of MF-BPR lies in its ability to leverage level information when sampling negative items.

A Personalised Ranking Framework with Multiple Sampling Criteria for Venue Recommendation

“…[17,29]) are important factors to improve the e ectiveness of venue recommendation systems. In particular, several approaches have been proposed to extend the BPR model to leverage additional information to enhance the e ectiveness of the BPR model [14,21,25,30]. However, these approaches can only incorporate one type of additional information and are not su ciently exible to incorporate other additional information.…”

“…In addition, these approaches did not use either geographical in uence or social correlation as explored in the previous studies mentioned above (Further discussed in Section 3.3-3.4). Recently, Loni et al [14] proposed a pairwise ranking framework that extends the BPR model to leverage multiple types of implicit feedback (e.g. click and likes) for item recommendation.…”

“…For each dataset, we measure the quality of the ranked venue recommendations in terms of Mean Average Precision (MAP), Normalised Discounted Cumulative Gain (NDCG) and Mean Reciprocal Rank (MRR), which are widely used in the recent recommendation literature [14,21,25,30]. In particular, MAP and MRR consider the ranking nature of the task, by taking into account the rank(s) of the venues that each user has previously visited/rated in the produced ranking, while NDCG goes further by considering the check-in frequency/rating value of the user as the graded relevance label.…”

“…Bayesian Personalised Ranking (BPR) [19] is a pairwise ranking-based model that is widely implemented and extended to leverage implicit feedback to generate the top-K venue recommendations (e.g. [14,21,25,30]). …”

“…However, a pairwise venue recommendation framework that seamlessly incorporates multiple types of additional information has not been previously proposed. Moreover, the various extended sampling criteria for BPR previously proposed in the literature [14,21,25,30] are based on pre-de ned assumptions and not on motivated by characteristics of users' movement and social interactions in LBSNs that have been observed in previous check-in studies [3,28,29] ( is is further discussed in Section 3). Furthermore, such sampling criteria are not su ciently exible to incorporate additional sources of information.…”