Uncovering collusive spammers in Chinese review websites

摘要: As the rapid development of China's e-commerce in recent years and underlying evolution adversarial spamming tactics, more sophisticated activities may carry out Chinese review websites. Empirical analysis, on recently crawled product reviews from a popular website, reveals failure many state-of-the-art spam indicators detecting collusive spammers. Two novel methods are then proposed: 1) KNN-based method that considers pairwise similarity two reviewers based their group-level relational information selects k most similar for voting; 2) general graph-based classification jointly classifies set transaction correlations. Experimental results show both our promisingly outperform indicator-only classifiers various settings.