Assessing Items Reliability for Collaborative Filtering Within the Belief Function Framework

摘要

Item-based collaborative filtering is among the most widely used recommendation approaches. It consists of identifying the most similar items in order to perform recommendations accordingly. However, the reliability of the information provided by these pieces of evidence cannot be fully trusted. Hence, quantifying their reliability seems imperative to form more valuable evidence. This paper contributes to the problem of covering uncertainty in the prediction process using the belief function theory. Our approach tends to take into account the different degrees of reliability of each similar item based on the discounting factor. Then, Dempster's rule of combination is used as an aggregation operator to combine these pieces of evidence. The performance of the new evidential method is validated on a real world data set.