Item group based pairwise preference learning for personalized ranking

Abstract

Collaborative filtering with implicit feedbacks has been steadily receiving more attention, since the abundant implicit feedbacks are more easily collected while explicit feedbacks are not necessarily always available. Several recent work address this problem well utilizing pairwise ranking method with a fundamental assumption that a user prefers items with positive feedbacks to the items without observed feedbacks, which also implies that the items without observed feedbacks are treated equally without distinction. However, users have their own preference on different items with different degrees which can be modeled into a ranking relationship. In this paper, we exploit this prior information of a user's preference from the nearest neighbor set by the neighbors' implicit feedbacks, which can split items into different item groups with specific ranking relations. We propose a novel PRIGP(Personalized Ranking with Item Group based Pairwise preference learning) algorithm to integrate item based pairwise preference and item group based pairwise preference into the same framework. Experimental results on three real-world datasets demonstrate the proposed method outperforms the competitive baselines on several ranking-oriented evaluation metrics.

Spectrum-enhanced Pairwise Learning to Rank

“…It is assumed that all users are independent in BPR, Pan and Chen [30] tried to relax this constraint and proposed a method called group preference-based Bayesian personalized ranking (GBPR), which modeled the preference of user groups. Qiu et al [33] constructed the preference chain of item groups for each user. Liu et al [24] utilized collaborative information mined from the interactions between users and items.…”

“…Like many existing works [30,33,49], we also use the connections of users and items to enhance the pairwise learning. However, in previous works, only one-order connections are utilized while in our SPLR model, we leverage connections with all orders.…”

BayDNN

“…For each user, by predicting the likelihood of friendship between friends larger than non-friends, ranking methods are proved to be e ective combating the imbalance issue [18,30,34]. Widely used in item recommendation tasks, one popular Bayesian personalized ranking (BPR) [30,34] model was integrated with matrix factorization [18], called Bayesian Personalized Ranking Matrix Factorization (BPRMF). ey de ned a Bayesian pairwise ranking relation, i.e., observed data to be rated before unobserved data by user-item linkage probability, which is calculated by matrix factorization.…”

“…Compared to other options such as classi cation or rating problems that su er from the imbalance problem, the ranking formulation can more or less avoid it [18,30,34]. Our model aims to maximize the ranking likelihood probability as follows:…”

“…Regarding friend recommendation as a learning to rank task -i.e., for each user, by ranking the probability of friendship for the users connected to it larger than that of users not connected to itthe ranking method has been proved to be e ective combating the imbalance issue. Among them, the Bayesian personalized ranking (BPR) model [18,30,34] de ned Bayesian pairwise ranking relation between users, i.e., the friendship probability of observed friends should be larger than the unobserved users pairs, was validated to achieve quite good performance.…”