A hybrid Data-Driven framework for Spam detection in Online Social Network

Abstract

Twitter is one of the most prominent online social networks(OSN) used by celebrities, politicians, ordinary people, and organizations to enhance their popularity and brand value. The popularity of Twitter makes it a prime target for spammers to attack. Earlier, spammers used to focus on email and web-search engines; therefore, extensive research in spam detection for email and web search engines has been done. However, the humongous traction of OSN has taken away the focus of spammers from email and web search engines towards OSN and micro-blogging sites. The study is inspired by the need to verify the legitimacy of a profile on OSN to avoid any fake information or rumors on OSN. This paper modulates a hybrid framework for spam detection on Twitter. The sampling algorithm SMOTE-ENN combines SMOTE and Edited Nearest Neighbors (ENN) to generate balanced data that is further fed to various deep learning classification techniques to identify whether the tweet is spam or ham. The efficacy and performance of various state-of-the-art algorithms are evaluated through simulation and are compared through various performance metrics to determine the best spam detection framework for OSN.