TweetScore
Yihe Zhang, Hao Zhang, Xu Yuan, Nian-Feng Tzeng · 2019
The spammers have been grossly detrimental since the inception of Twitter social networks and keep polluting social environments by hiding themselves among a large amount of normal users. In this paper, we aim to address two challenges existing in the spammer detection problem: 1) monitoring tweets that have a higher probability of including spam messages; 2) providing an accurate solution for spam classification. To address these two challenges, we first propose a pseudo-honeypot framework for efficient tweets monitoring and collection. By taking advantage of users' diversity and selecting normal users as the parasitic body, the pseudo-honeypot can harness normal users with features having much more potentials of attracting spammers. This lets the pseudo-honeypot collect tweets that are far more likely to include spam messages. Furthermore, we design a novel spam classification solution called TweetScore by exploring both the intrinsic attributes' and users' relationships in social networks. TweetScore quantifies such relationships into a vector of numerical values to represent each tweet's score, reflecting the associated user's behaviors. The neural network is then employed to take these vectors as input to classify spams and spammers. Through extensive experiments, we demonstrate the efficiency of the pseudo-honeypot system on spam monitoring and the accuracy of TweetScore on spam classification. Specifically, the spam and spammer ratios collected by our pseudo-honeypot system are four times as much as those of a non pseudo-honeypot counterpart while the TweetScore can achieve, on an average, 93.5% accuracy, 93.71% precision, and 1.52% false positive in online spam classification.