Detecting Online Harassment in Social Networks

Online Harassment is the process of sending messages over electronic media to cause psychological harm to a victim. In this paper, we propose a pattern-based approach to detect such messages. Since user generated texts contain noisy language, we perform a normalization step first to transform the words into their canonical forms. Additionally, we introduce a person identification module that marks phrases which relate to a person. Our results show that these preprocessing steps increase the classification performance. The pattern-based classifier uses the information provided by the preprocessing steps to detect patterns that connect a person to profane words. This technique achieves a substantial improvement compared to existing approaches. Finally, we discuss the portability of our approach to Social Networks and its possible contribution to tackle the abuse of such applications for the distribution of Online Harassment