Multi-Pointer Co-Attention Networks for Recommendation

Abstract

Many recent state-of-the-art recommender systems such as D-ATT, TransNet and DeepCoNN exploit reviews for representation learning. This paper proposes a new neural architecture for recommendation with reviews. Our model operates on a multi-hierarchical paradigm and is based on the intuition that not all reviews are created equal, i.e., only a selected few are important. The importance, however, should be dynamically inferred depending on the current target. To this end, we propose a review-by-review pointerbased learning scheme that extracts important reviews from user and item reviews and subsequently matches them in a word-byword fashion. This enables not only the most informative reviews to be utilized for prediction but also a deeper word-level interaction. Our pointer-based method operates with a gumbel-softmax based pointer mechanism that enables the incorporation of discrete vectors within differentiable neural architectures. Our pointer mechanism is co-attentive in nature, learning pointers which are co-dependent on user-item relationships. Finally, we propose a multi-pointer learning scheme that learns to combine multiple views of user-item interactions. We demonstrate the effectiveness of our proposed model via extensive experiments on 24 benchmark datasets from Amazon and Yelp. Empirical results show that our approach significantly outperforms existing state-of-the-art models, with up to 19% and 71% relative improvement when compared to TransNet and DeepCoNN respectively. We study the behavior of our multi-pointer learning mechanism, shedding light on 'evidence aggregation' patterns in review-based recommender systems.

Are we really making much progress? A worrying analysis of recent neural recommendation approaches

“…The fraction of papers that were reproducible according to our relatively strict criteria per conference series are shown in Table 1. Non-reproducible: KDD: [43], RecSys: [41], [6], [38], [44], [21], [45], SIGIR: [32], [7], WWW: [42], [11] Overall, we could reproduce only about one third of the works, which confirms previous discussions about limited reproducibility, see, e.g., [3]. The sample size is too small to make reliable conclusions regarding the difference between conference series.…”

A Capsule Network for Recommendation and Explaining What You Like and Dislike

“…Baselines. Here, we compare the proposed CARP against the conventional baseline and recently proposed state-of-the-art rating prediction methods: (a) probabilistic matrix factorization that leverages only rating scores, PMF [23]; (b) latent topic and shallow embedding learning models with reviews, RBLT [26] and CMLE [37]; (c) deep learning based solutions with reviews, DeepCoNN [39], D-Attn [24], TransNet [3], TARMF [19], MPCN [27] and ANR [7]. Among these methods, D-Attn, TARMF, MPCN and ANR all identify important words for rating prediction.…”

“…Recent years, capitalizing user reviews to enhance the precision and the interpretability of recommendation have been investigated and verified by many works [1,3,4,6,8,15,18,20,24,26,27,29,37,39]. In earlier days, many efforts are made to extract semantic features from reviews with the topic modeling techniques [2,14].…”

“…Likewise, MPCN exploits a pointer-based co-attention schema to enable a multi-hierarchical information selection. Both important reviews and their important words can be identified towards better rating prediction in MPCN [27].…”

“…Specifically, empowered by continuous real-valued vector representations and semantic composition over contextual information, uptodate review-based deep learning models significantly push the frontier of the state-of-the-art further. Examples include DeepCoNN [39], D-Attn [24], TransNet [3], ANR [7] and MPCN [27].…”