Abstract
Twitter has become an open space for the users' interactions and discussions on diverse trending topics. One of the issues raised on social media platforms is the misunderstanding of 'freedom of speech', which in turn, leads us to a new social and behavioral attack: cyberbullying. Cyberbully affects both individuals and societies. Despite tough sanctions globally and locally, cyberbullying is still a serious issue, which needs further consideration. Thus, this research aims to address this issue by proposing a framework, based on sentiment analysis, to detect cyberbullying in the tweets stream. The proposed framework in this paper extracts the tweets from Twitter. Then, the preprocessing through tweets tokenization was applied to remove noise from tweets and also symbols and phrases such as http, emoji faces, hash tag symbols, mention symbols and retweet. After data tokenization, the proposed system classifies the tweets, based on extracted keywords from both experts and potential victims, using deep-learning classification algorithm with 70% of dataset samples used for the training purpose, and 30% of the dataset samples used for the testing purpose. The experimental results show the ability of proposed methodology to detect cyberbullying effectively with accuracy 81%.