Abstract
Providing personalized explanations for recommendations can help users tounderstand the underlying insight of the recommendation results, which ishelpful to the effectiveness, transparency, persuasiveness and trustworthinessof recommender systems. Current explainable recommendation models mostlygenerate textual explanations based on pre-defined sentence templates. However,the expressiveness power of template-based explanation sentences are limited tothe pre-defined expressions, and manually defining the expressions requiresignificant human efforts. Motivated by this problem, we propose to generatefree-text natural language explanations for personalized recommendation. Inparticular, we propose a hierarchical sequence-to-sequence model (HSS) forpersonalized explanation generation. Different from conventional sentencegeneration in NLP research, a great challenge of explanation generation ine-commerce recommendation is that not all sentences in user reviews are ofexplanation purpose. To solve the problem, we further propose an auto-denoisingmechanism based on topical item feature words for sentence generation.Experiments on various e-commerce product domains show that our approach cannot only improve the recommendation accuracy, but also the explanation qualityin terms of the offline measures and feature words coverage. This research isone of the initial steps to grant intelligent agents with the ability toexplain itself based on natural language sentences.