Towards a Hybrid User and Item-Based Collaborative Filtering Under the Belief Function Theory

Abstract

Collaborative Filtering (CF) approaches enjoy considerable popularity in the field of Recommender Systems (RSs). They exploit the users’ past ratings and provide personalized recommendations on this basis. Commonly, neighborhood-based CF approaches focus on relationships between items (item-based) or, alternatively, between users (user-based). User-based CF predicts new preferences based on the users sharing similar interests. Item-based computes the similarity between items rather than users to perform the final predictions. However, in both approaches, only partial information from the rating matrix is exploited since they rely either on the ratings of similar users or similar items. Besides, the reliability of the information provided by these pieces of evidence as well as the final predictions cannot be fully trusted. To tackle these issues, we propose a new hybrid neighborhood-based CF under the belief function framework. Our approach tends to take advantage of the two kinds of information sources while handling uncertainty pervaded in the predictions. Pieces of evidence from both items and users are combined using Dempster’s rule of combination. The performance of the new recommendation approach is validated on a real-world data set and compared to state of the art CF neighborhood approaches under the belief function theory.