Top-N Recommendation with Multi-Channel Positive Feedback using Factorization Machines

Abstract

User interactions can be considered to constitute different feedback channels, for example, view, click, like or follow, that provide implicit information on users’ preferences. Each implicit feedback channel typically carries a unary, positive-only signal that can be exploited by collaborative filtering models to generate lists of personalized recommendations. This article investigates how a learning-to-rank recommender system can best take advantage of implicit feedback signals from multiple channels. We focus on Factorization Machines (FMs) with Bayesian Personalized Ranking (BPR), a pairwise learning-to-rank method, that allows us to experiment with different forms of exploitation. We perform extensive experiments on three datasets with multiple types of feedback to arrive at a series of insights. We compare conventional, direct integration of feedback types with our proposed method, which exploits multiple feedback channels during the sampling process of training. We refer to our method as multi-channel sampling. Our results show that multi-channel sampling outperforms conventional integration, and that sampling with the relative “level” of feedback is always superior to a level-blind sampling approach. We evaluate our method experimentally on three datasets in different domains and observe that with our multi-channel sampler the accuracy of recommendations can be improved considerably compared to the state-of-the-art models. Further experiments reveal that the appropriate sampling method depends on particular properties of datasets such as popularity skewness.

A Survey on Heterogeneous One-class Collaborative Filtering

“…It is worth mentioning that the observation in experiments implies that the combination of an interacted item i and an un-interacted item j is a better candidate pair for model training, which means that an un-interacted item has a higher probability to be a negative item. Consequently, the extended method of MF-BPR [Loni et al 2019] improves the sampling strategy, in which negative items are only picked up from the un-interacted items, meaning P − is always the un-interacted level. The authors introduce three different ways for sampling negative items, where the first one is the uniform sampler like the vanilla BPR, the second one over-samples the popular items due to the popularity skewness problem, and the last one is a multi-channel sampler considering both the popularity of the items and the level of the feedback.…”

“…Before diving into the details of the comparisons, we first make a brief introduction of the evaluation protocols and metrics in the reviewed works. As for the evaluation protocols, hold-out [Guo et al 2017a;, k-fold cross validation [Loni et al 2019;Qiu et al 2018], and leave one out [Chen et al 2020;Zhou et al 2019] are three commonly used protocols. As for the evaluation metrics, they are similar to those in most works on information retrieval and recommender systems, including hit ratio, precision, recall, F1, NDCG, 1-call, mean reciprocal rank (MRR), mean average precision (MAP), and area under the curve (AUC) [Ricci et al 2015].…”

“…Pick up some appropriate positive and negative samples in training [Lerche and Jannach 2014;Loni et al 2016Loni et al , 2019.…”

“…There are 107K users and 190K items with 327K replies, 1.3M clicks, 138K likes, and 253K deletes. The benchmark is used in Guo et al [2017a], Loni et al [2019] and Qiu et al [2018]. • BeiBei 9 is the largest e-commerce platform for maternal and infant products in China.…”

Deep Interest-Shifting Network with Meta-Embeddings for Fresh Item Recommendation

“…Another group of studies exploits tensor factorization techniques for modeling useritem-context relations [43,44]. Recently, factorization machines [42,45,46] and deep learning [47,48] based on CARS become increasingly popular, which directly model nonlinear interactions between features. Some studies also use representation learning techniques, e.g., CARS 2 [49] and COT [37], which provide not only a latent vector but also context-aware representations.…”