DEVELOPING MACHINE LEARNING METHODS FOR IDENTIFYING TWITTER USERS PROFILE AND BEHAVIOURS

Abstract

The phenomenon of Cyberbullying and cyber aggression has become a negative and worrying effect, especially with its continuous increase on users of social networks of young people and teenagers. This is also a result of the continuous expansion in the use of social networks. Studies have shown that the percentage of those exposed to this negative phenomenon reached more than half of young users all over the world. The problem of online bullying and violence is due to the existence of psychological issues for the person responsible for these acts due to his exposure to psychological problems. Victims also will be vulnerable to be faced with a variety of emotions, with negative effects such as awkwardness, sadness, and friendlessness with other community members who are at risk of causing further critical consequences like trying to kill themselves. In our work, we first studied the phenomena of cyberbullying and cyber aggression in Twitter which is conceder currently on of the biggest social media platforms. We used a dataset around 20,000 to 30,000 rows back to the period from 2013 till 2015 to train and test the model with tweet profiling and body analysis of the Tweet itself for sentiment analysis to identify a positive, negative, or neutral Tweet. We divided the dataset in different ratios several times for training and test purposes. We used 8 Machine learning algorithms to get the best results. In demonstrating that sentiment analysis in this work can draw on the same set of tweets and create additional labels (positive, negative, or neutral) from the basic tweet labels. We added strong accuracy to differentiate Tweets from a user’s Tweets. We have reached an average accuracy of around 98% in the developed model