TAERT: Triple-Attentional Explainable Recommendation with Temporal Convolutional Network

Explainable Recommendation aims at not only providing the recommended items to users, but also enabling users to be aware of why these items are recommended. To better understand the recommended results, textual reviews have been playing an increasingly important role in the recommender systems. However, how to learn the latent representation of user preferences and item features, and how to model the interactions between them effectively via specific aspects in the reviews are two crucial problems in the explainable recommendation. To this end, we propose a novel Triple-Attentional Explainable Recommendation with Temporal Convolutional Network, named TAERT, which is to jointly generate recommendation results and explanations. Specifically, we first explore a feature learning method based on Temporal Convolutional Network (TCN) to derive word-aware and review-aware vector representations. Then, we introduce three levels of attention networks to model word contribution, review usefulness and importance of latent factors, respectively. Finally, the predicted rating is inferred by the factor-level attention based prediction layer. Furthermore, the attention mechanism is also conducive to identifying the representative item reviews and highlighting the informative words to generate explanations. Compared with the state-of-the-art methods, comprehensive experiments on six real-world datasets are conducted to verify the effectiveness on both recommendation and explanation. (c) 2021 Elsevier Inc. All rights reserved.

Automated summary

This study introduces a novel Triple-Attentional Explainable Recommendation method that jointly generates recommendation results and explanations, and demonstrates its effectiveness in both recommendation and explanation through comprehensive experiments on six real-world datasets.