Islamophobic hate speech detection from electronic media

Abstract

Islamophobic hate speech is the indiscriminate negative attitude and behavior towards Muslims and Islam. Due to the rapid growth of the Internet and availability of low cost devices, the number of social and electronic media users has increased tremendously in the recent years. A downside of this growth is the increase in conflicts, hate speech, cyber trolling and bullying. Among these, the focus of our current research lies on hate speech identification and more specifically the Islamophobic hate speech. Islamophobic hate speech in the form of tweets, posts or articles has caused a serious damage to the Muslim community and has resulted in a deshaped perception of Islam for the West, especially after the 9/11 attacks. In this context, it is highly important to develop solutions which can automatically identify the Islamophobic hate speech from electronic media so that corrective measures may be taken. In this research, we present an effective machine learning based solution to identify Islamophobic hate speech in the Tweets. The technique relies on converting raw tweets into word embeddings which allow similar concepts to be treated in the same manner. Effective feature representations are learned using one dimensional convolutions on the embedded tweets while a bi-directional long short-term memory (LSTM) network is employed to model the input tweets and identify potential Islamophobic hate speech content. The technique was evaluated on a subset of a publicly available dataset of tweets and yielded an accuracy of 90.13%.