Social Media Spam Detection Using Fuzzy String Matching

Abstract

One of the most popular internet activities around the world is visiting online social networks. The number of users and time spend by users on these networks is increasing at a high rate. Moreover users tend to rely on the trustworthiness of the data present on the networks. But in wrong hands this trustworthiness can easily be exploited and be used for spreading spam. Users are harassed by spam messages which waste time and make users to click on malicious links. Spams effect many type of electronic communication including instant messaging, email and social networks but due to open nature, reliance on users for data and because of huge user base social networks are worst hit because of spams. To detect spams from the social networks it is desirable to find new unsupervised techniques which can save the training cost of supervised techniques.

In this thesis we present an unsupervised, distributed and decentralized technique to detect and remove spams from the social networks. We present a new technique which uses fuzzy based method to detect spams which is different from existing techniques and which can catch spam even from single message stream. We have parallelized our work with multi-core and multi-thread to give improved results. To handle huge data of networks we have implemented our technique on MapReduce platform which gave very promising results.