A robust ranking method for online rating systems with spammers by interval division

It is crucial to identify spammers from online e-commerce platform to maintain the order of fairness. Existing methods have limitations to detect spammers if the underlying network is extreme sparse. In this paper, a novel method has been proposed to address this challenge from two folds. It is inspired by the idea that a trust-worthy rater will always give a reasonable rating which has been statistically significant and locates in an interval following normal distribution. To deal with low-degree spammers with limited information, rating patterns with preference are involved as well. Such two parts lead to an Interval Division-based Ranking (IDR) method. Experimental study on challenging sparse network Amazon demonstrates that the performance gain of recall is at least 15.4%. Top 50 movies selected by IDR from Douban have a high mean value 9.552 and a low variance 0.036.