Explainable Recommendation Enhancing Review Properties and PPLM
Akihiro Kokubo, Kazunari Sugiyama · 2022
Explainable recommendation, which provides items and explanation why they are recommended, have attracted a lot of attention as it could improve transparency, persuasiveness, effectiveness, reliability, and user satisfaction of recommender systems. A lot of explainable recommender systems have been proposed so far. However, they do not take review properties into account, need to further improve generation of explanation, and also need to analyze specific cases in recommendation.Our work aims at developing an explainable recommender model that improves quality of generated explanations as well as accuracy of recommendation. To achieve this, we propose an end-to-end architecture that leverages properties of review sentences and Plug and Play Language Model (PPLM). Experimental results on publicly available datasets demonstrate that our proposed recommender model improves both accuracy of rating prediction and quality of generated explanations by employing multi-task learning framework.