WordRecommender - An Explainable Content-Based Algorithm based on Sentiment Analysis and Semantic Similarity.

Regarding recommender systems, there has been a historically dominance on the literature in favor of collaborative algorithms over content-based ones. However, the latter can work better with users on applications and be more transparent. Therefore, in this paper, we propose WordRecommender, an explainable content-based algorithm that calculates similarity by semantic proximity. On its preprocessing step, reviews in the movie context are analyzed to obtain aspects, defined as words with high sentimental value. After that, the recommendations are generated with a neighborhood algorithm that calculates the films' similarity, based on the semantic proximity of the aspects, ordered by their emotional score. Finally, to produce textual explanations, the algorithm can also consider the semantic comparison of metadata by utilizing the most related aspects from the recommended movie and an enjoyed one by the user. As a result, it was found that the algorithm has competitive accuracy when compared with other baseline neighborhood methods, in conclusion that the semantic data of items can be the source of both representative information and reasoning on recommender systems.