Profiling hate speech spreaders on Twitter using BERT pre-trained model and neural network

Abstract

The profiling hate speech readers task is to determine whether the author of a given group of English or Spanish tweets posted on twitter spreads hate speech. At present, the automatic recognition of hate, irony and false speech is the top priority in the field of author profiling, which has practical significance. This paper proposes a deep learning model based on Bert pre-trained model to extract deep text semantic features, and use counting method to obtain stylistic features. Finally, hate speech profiling is regarded as a binary classification task, and full connected neural network is used for classification prediction. The accuracy is 71.5% on the English data set and 78% on the Spanish test set.