Deep Context-Aware Embedding for Abusive and Hate Speech Detection on Twitter

Violence usually spread online, as it has spread in the past. With the increasing use of social media, the violence attributed to online hate speech has increased worldwide resulting rise in number of attacks on immigrants and other minorities. Analysis of such short text posts (e.g. tweets etc.) is valuable for identification of abusive language and hate speech. In this paper, we present Deep Context-Aware Embedding for the detection of Hate speech and abusive language on twitter. To improve the classification performance, we have enhanced the quality of the tweets by considering polsemy, syntax, semantic, OOV words as well as sentiment knowledge and concatenated to form input vector. We have used BiLSTM with attention modeling to identify tweet with hate speech. Experimental results showed significant improvement in the classification of tweets.

Tags: Abusive language detection, Contextual Word Embedding, Deep learning, NLP