A Prediction-Based Approach for Computing Robust Rating Scores

Abstract

Assessing quality of products, especially when purchased online, is always a challenge. One of the widely used approaches for addressing this challenge is to rely on the scores computed by online rating systems, based on the feedback received from other users. For several reasons, like gaining benefits, personal interests or collusion, rating systems have always been facing with challenge of dishonest feedback. Although many techniques have been proposed for collusion detection, there are still issues that need more investigations. One of these issues is dealing with the sparsity problem, i.e., small number of votes per product, which makes it easier to manipulate scores. In this paper we propose a novel technique for calculating robust rank scores which relies on feedback prediction. In our model, we improve quality of computed scores by predicting feedback, for the people who have not assessed a product. This will result in decreasing sparsity. Then, we propose an iterative technique to calculate product rating scores based on the real and predicted feedbacks. We have implemented our method and compared its performance with three well-known related works. The result of comparison shows the superiority of our model.