CoFiSet: Collaborative Filtering via Learning Pairwise Preferences over Item-sets

Abstract

Collaborative filtering aims to make use of users' feedbacks to improve the recommendation performance, which has been deployed in various industry recommender systems. Some recent works have switched from exploiting explicit feedbacks of numerical ratings to implicit feedbacks like browsing and shopping records, since such data are more abundant and easier to collect. One fundamental challenge of leveraging implicit feedbacks is the lack of negative feedbacks, because there are only some observed relatively "positive" feedbacks, making it difficult to learn a prediction model. Previous works address this challenge via proposing some pointwise or pairwise preference assumptions on items. However, such assumptions with respect to items may not always hold, for example, a user may dislike a bought item or like an item not bought yet. In this paper, we propose a new and relaxed assumption of pairwise preferences over item-sets, which defines a user's preference on a set of items (item-set) instead of on a single item. The relaxed assumption can give us more accurate pairwise preference relationships. With this assumption, we further develop a general algorithm called CoFiSet (collaborative filtering via learning pairwise preferences over item-sets). Experimental results show that CoFiSet performs better than several stateof-the-art methods on various ranking-oriented evaluation metrics on two real-world data sets. Furthermore, CoFiSet is very efficient as shown by both the time complexity and CPU time.

CroRank: Cross Domain Personalized Transfer Ranking for Collaborative Filtering

“…The most widely used ones are P re@k [25], NDGG@k [14] and Rec@k [15]. And we can derive F 1 score [15] based on P re@k and Rec@k. For each of the evaluation metric, we obtain the corresponding performance for each user u in the test data and then average the performance over all users to attain the ultimate results.…”

“…Pan et al [15] consider the group pref erence information to alter basic assumptions of BPR. In [14], the authors try to aggregate related items to form item-sets, and model user's preference level over the item-sets which is maybe more reasonable the original assumptions of BPR. [1] adds a social regularization term to BPR to promote better performance.…”

Item group based pairwise preference learning for personalized ranking

“…7. We omit the detailed iteration algorithm due to the space limit, and the similar updating method is described in [6,7].…”

“…Two popular baseline methods are used for empirical comparison, which is PopRank [6] and BPR [7]. (1)PopRank is a basic algorithm for the problem of CF with implicit feedbacks, which makes the recommendation to users in terms of global popularity of items.…”

Recommending user generated item lists

“…Another work which proposes to model implicit dataset through ranking is CLiMF [30], which instead of optimizing AUC as in BPR, considers Mean Reciprocal Rank as the optimization criterion. In [24], the authors argue that reasoning about preferences between individual items might be too fine a granularity, and propose to lift item preferences to preferences over sets of items. The focus of this work is still on item recommendation, and when aggregating item preferences, the proposed model does not consider the position of an item, and the defined item sets are hidden from users thus there is no need to model how users might view items within a set.…”