Validating User Spam Reports in Chat Networks

1. Abstract In this paper, we consider the problem of validating user reports and detecting malicious users in chat networks. Bad behavior in chat networks is usually handled by a rather tedious system where the victim generates a report for the administrators to act on. However, since most of these reports are spurious, administrators have to manually validate each report before taking action. We analyze the effectiveness of Naive Bayes and Support Vector Machines using a couple of different kernels. Our experiments indicate that a SVM based algorithm using a polynomial kernel with the right features achieves an overall accuracy of 98% 99%. The recall for the positive and the negative classes are also over 95%.