Incorporating users satisfaction to resolve sparsity in recommendation systems

Abstract

Recommendation systems have shown great potential to help users in order to find interesting and relevant items from within a large information space. Information overload experienced a significant response because of huge amount of internet usage and also has been demonstrated by recommendation systems; that is, providing adapted information services. User preferences play a key role in formulating recommendations to search required information over the web. User feedback, both explicit and implicit, has proven to be vital for recommendation systems, with the similarity between users then able to be computed. In this paper we propose that traditional reliance on user similarity may be overstated. Nevertheless, there are many problems to be faced, specifically; sparseness, cold start, prediction accuracy, as well as scalability which can all result in a challenge of accuracy over the recommendation systems. A sparsity rate of 95% is experienced in CF-based commercial recommendation applications. Furthermore, we discuss the manner in which other factors have significant effect in managing recommendations. Specifically, we propose that the issue of user satisfaction must be considered and incorporated with explicit feedback for improved recommendations. Supported by experimental results, our approach demonstrates better results while incorporating user satisfaction with the feedback.