Exploiting Item Representations for Soft Clustering Recommendation

Recommender systems help dealing with the information overload problem since they provide personalized content for users. There are two major paradigms in recommendation: content-based and collaborative filtering. Regardless of the paradigm, there has been a great effort into finding additional information to better describe items and/or users, which in turn helps to increase the personalization power of the system. User's reviews turn out to be a great source of information, since they provide information about the characteristics of the items as well as insights about the opinion of the user towards them. In previous works, we explored some techniques for extracting information from reviews in order to generate items' representations and applied them into an item k-NN algorithm. In this work, we explore the impact that those representations, alongside with rating and genre-based representations, can cause into a soft clustering-based recommender system. We compare our findings with the item k-NN algorithm and observe that they are better in some cases, but the soft clustering recommender has lower computational cost.