A³R: Argumentative explanations for recommendations

Existing recommender systems often apply factorization-based models, which have been proved to be efficient in rating prediction. However, the explicit semantics of the learned latent factors are not clear, which makes it difficult to explain the recommendations returned. In another line of research, argumentation-based methods have become an important tool in explainable artificial intelligence. In this work, we propose an Attribute-Aware Argumentative Recommender (A3R) that combines factorization-based methods and argumentation. With the help of argumentation framework, each step of A3R is endowed with explicit semantics, enabling A3R to generate easily understandable explanations for recommendations. Experiments on five datasets from three different domains (movie, music, and book) show that A3R can achieve competitive rating prediction when compared with factorization-based methods; A3R can largely improve prediction accuracy when compared with state-of-art argumentative recommendation methods.