Cyberbullying Detection, Based on the FastText and Word Similarity Schemes

Abstract

With recent developments in online social networks (OSNs), these services are widely applied in daily lives. On the other hand, cyberbullying, which is a relatively new type of harassment through the internet-based electronic devices, is rising in online social networks. Accordingly, scholars are attracted to investigating cyberbullying behaviors. Studies show that cyberbullying has a devastating effect on mental health, especially for teenagers. In order to reduce or even stop cyberbullying, different machine learning techniques are applied and numerous studies have been conducted so far. However, conventional detection schemes still have challenges, such as low accuracy. Therefore, it is of significant importance to find an efficient detection solution in the natural language processing and machine learning communities. In the present study, characteristics of cyberbullying are initially analyzed from vocabulary and syntax points of view. Then a new detection algorithm is proposed based on FastText and word similarity schemes. Finally, experiments are carried out to evaluate the effectiveness and performance of the proposed method. Obtained results show that the proposed algorithm can effectively improve the detection accuracy and recall rate of cyberbullying detection.