A Transfer Metric Learning Method for Spammer Detection

Abstract

Microblogs open opportunities for social spammers, who are threatening for microblog services and normal users. Therefore, detecting spammers is an essential task in social network mining. However, existing methods are difficult to achieve desired performance in real applications. The underlying causes are the insufficiency of knowledge learned from limited training examples and the differences between data distributions on training and test examples. To address these, in this paper, we present a transfer metric learning method to extract more informative knowledge underlying training instances by similarity learning and transfer this knowledge to test instances using importance sampling in a unified framework. We evaluate the proposed method on real-world data. Results show that our method outperforms many baselines.