Model-based collaborative filtering to handle data reliability and ordinal data scale

Abstract

Accompanying with the Internet growth explosion, recommender systems arise to facilitate the searching and comprehending ability of the users who suffer from the information overload problem in acquiring useful information online. Collaborative filtering (CF) that makes recommendations by comparing novel information with common interests shared by a group of people becomes popular among such systems. Particularly, model-based CF receives much attention recently due to its computational efficiency and superior performance. Two issues on model-based CF, however, should be addressed in applications. First, data quality of the rating matrix input can affect the prediction performance. Second, most current models treat the measurement scale of data classes as a nominal one instead of ordinal in ratings. The objective of this research is therefore to propose a model-based CF algorithm that considers both issues. Two experiments are conducted accordingly, and the results show our proposed method outperforms its counterparts especially under data of mild sparsity degree and of a large scale. The feasibility of our proposed approach is thus justified.