Social networks, such as Facebook, Twitter and Youtube, are becoming robust sources that radical groups use to spread the propaganda of extremist group to gather resources for their cause. Therefore, locating radical social media content has become an priority task for counter-terrorism agencies, technology companies and governments. This study has employed a systematic literature review protocol to identify the publicly available corpora that can be used for training machine learning techniques for extremism detection in text. Furthermore, the snowballing approach is employed to identify the studies that have used these corpora for extremism detection. The identified datasets and the studies have been identified to develop an understanding of landscape of extremism detection. Based on the observations we have developed recommendations for processing this important area of research.