Detecting Spam on Twitter via Message-Passing Based on Retweet-Relation - Technologies and Applications of Artificial Intelligence - page 64

Comparison between Different Classification Algorithms. Figure 4 shows
the experiment result of applying different number of clusters of k NN algorithm.
Both k NN(with k=2) and k NN(with k=3) achieve high accuracy. The reason
why we choose k NN(with k=3) is not only because of the lower FP rate but also
the higher accuracy. In addition, when choosing k NN(with k=2) will encounter
a draw. Therefore, we choose the k NN(with k=3) as our best result.
5 Conclusions
In this work, we propose Message-passing Graph Analyzer, a combination of
graph-based features to detect Twitter spam without collusion mask. Our sys-
tem is able to distinguish variant of spam collusion behavior. Stringhini et al. [16]
observed that the average lifetime of a spam account on Twitter is 31 days. But
by using our method, the lifetime of a spam account can be shortened in 1 to
2 hours. Due to the limitations of Twitter we can only construct the proxi-
mate message-passing graph. But there are many researches focused on analyz-
ing retweet behavior and inferred the possible information diffusion direction,
such as Du et al. [5] utilized the user activity time to infer the retweet behav-
ior. Therefore, we may further improve the message-passing graphs to that the
graphs more complete.
Acknowledgments. This research is supported in part by the Ministry of
Science and Technology of Taiwan under grants number MOST 102-2218-E-011-
011 MY3.
References
1. Baltazar, J., Costoya, J., Flores, R.: The real face of koobface: The largest web 2.0
botnet explained. Trend Micro Research (2009)
2. Barabasi, A.L., Oltvai, Z.N.: Network biology: understanding the cell's functional
organization. Nature Reviews Genetics 5(2), 101-113 (2004)
3. Bilge, L., Strufe, T., Balzarotti, D., Kirda, E.: All your contacts are belong to us:
automated identity theft attacks on social networks. In: Proceedings of Interna-
tional Conference on World Wide Web, pp. 551-560 (2009)
4. Boyd, D., Golder, S., Lotan, G.: Tweet, tweet, retweet: Conversational aspects
of retweeting on twitter. In: Proceedings of Hawaii International Conference on
System Sciences, pp. 1-10 (2010)
5. Du, J., Song, D., Liao, L., Li, X., Liu, L., Li, G., Gao, G., Wu, G.: ReadBehavior:
Reading probabilities modeling of tweets via the users' retweeting behaviors. In:
Tseng, V.S., Ho, T.B., Zhou, Z.-H., Chen, A.L.P., Kao, H.-Y. (eds.) PAKDD 2014,
Part I. LNCS, vol. 8443, pp. 114-125. Springer, Heidelberg (2014)
6. Ghosh, R., Surachawala, T., Lerman, K.: Entropy-based classification
of'retweeting'activity on twitter. In: Proceedings of KDD Workshop on So-
cial Network Analysis (2011)