Toward efficient and effective bullying detection in online social network

With the advances of Information Communication Technology (ICT) and the popularity of intelligent terminals, Online Social Network, which is characterized by powerful functions of information publishing, dissemination, acquisition and sharing, has attracted a huge number of users and become one of the most popular internet application services currently. However, the growth of Online Social Network has also led to the emergence of cyberbullying issues. Information spreads extremely fast via Online Social Network, making the harm caused by cyberbullying grow exponentially with time. As a result, it becomes critical to detect the cyberbullying in a quick and efficient way. In this paper, in order to solve this challenge, we propose an improved TF-IDF based fastText (ITFT) model for effective cyberbullying detection. Specifically, in our proposed scheme, we improve the TF-IDF algorithm by adding the position weight, keywords are extracted by the improved algorithm and used as input to achieve the purpose of filtering noise data to improve the accuracy. We use the fastText to construct a binary classifier to categorize the input data. Extensive experiments are conducted, and the results demonstrate that our proposed scheme can achieve better efficiency and accuracy in cyberbullying detection as compared with baselines.