Fully Connected Neural Network with Advance Preprocessor to Identify Aggression over Facebook and Twitter

Abstract

Aggression Identification and Hate Speech detection had become an essential part of cyberharassment and cyberbullying and an automatic aggression identification can lead to the interception of such trolling. Following the same idealization, vista.ue team participated in the workshop which included a shared task on ’Aggression Identification’. A dataset of 15,000 aggression-annotated Facebook Posts and Comments written in Hindi (in both Roman and Devanagari script) and English languages were made available and different classification models were designed. This paper presents a model that outperforms Facebook FastText (Joulin et al., 2016a) and deep learning models over this dataset. Especially, the English developed system, when used to classify Twitter text, outperforms all the shared task submitted systems.