Detecting Spam Tweets in Trending Topics Using Graph-Based Approach

In recent years, social media has changed the way people communicate and share information. For example, when some important and noteworthy event occurs, many people like to "tweet" (Twitter) or post information, resulting in the event trending and becoming more popular. Unfortunately, spammers can exploit trending topics to spread spam more quickly and to a wider audience. Recently, researchers have applied various machine learning techniques on accounts and messages to detect spam on Twitter. However, the features of typical tweets can be easily fabricated by the spammers. In this work, we propose a graph-based approach that leverages the relationship between the named entities present in the content of the tweet and the document referenced by the URL mentioned in the tweet for detecting possible spam. It is our hypothesis that by combining multiple, heterogeneous information together into a single graph representation, we can discover unusual patterns in the data that reveal spammer activities - structural features that are difficult for spammers to fabricate. We will demonstrate the usefulness of this approach by collecting tweets and documents referenced by the URL in the tweet related to Twitter trending topics, and running graph-based anomaly detection algorithms on a graph representation of the data, in order to effectively detect anomalies on trending tweets.