(Georgia Institute of Technology, 2010)
Balasubramaniyan, Vijay A.; Maheswaran, Arjun; Mahalingam, Viswanathan; Ahamad, Mustaque; Venkateswaran, H.
The growing popularity of Twitter and its ability to enable near instantaneous sharing of information has made
it a target of attacks by malicious entities who use it to
spam and provide links to malware. There is evidence that these entities are using increasingly sophisticated techniques
that mimic the behavior of reputed sources to avoid detection.
We use novel mechanisms that utilize the true social
network of users, the quality of information produced by
them and their tweeting behavior to identify such entities.
A scheme based on these mechanisms is even able to detect
malicious entities that collude to establish dense social
networks. Using actual data from a representative sample
of 278, 758 Twitter users, we demonstrate the effectiveness
of this approach by showing that (1) we identified 5334 accounts
that had links to unsafe websites, and (2) over a
period of 31 days, 181 accounts that our algorithm identified
as potentially malicious were subsequently suspended
by Twitter. We believe our algorithm is one of the first to
automatically deal with a broad range of malicious entities
present in Twitter.