Towards the Importance of the Type of Deep Neural Network and Employment of Pre-trained Word Vectors for Toxicity Detection: An Experimental Study

As a natural consequence of offering many advantages to their users, social media platforms have become a part of their daily lives. Recent studies emphasize the necessity of an automated way of detecting offensive posts in social media since these 'toxic' posts have become pervasive. To this end, a novel toxic post detection approach based on Deep Neural Networks was proposed within this study. Given that several word embedding methods exist, we shed light on which word embedding method produces better results when employed with the five most common types of deep neural networks, namely, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), GRU (Gated Recurrent Unit), Bidirectional Long ShortTerm Memory (BiLSTM), and a combination of CNN and BiLSTM. To this end, the word vectors for the given comments were obtained through four different methods, namely, (i) Word2vec, (ii) fastT ext, (iii) GloVe, and (iv) the Embedding layer of deep neural networks. Eventually, a total of twenty benchmark models were proposed and both trained and evaluated on a gold standard dataset which consists of 16K tweets. According to the experimental result, the best F1 - score, 84.844%, was obtained on the proposed CNN model without employing pre-trained word vectors which outperformed the state-of-the-art works and implies the effective embedding ability of CNNs. Other key findings obtained through the conducted experiments are that the models, that constructed word embeddings through the Embedding layers, obtained higher F1 - scores and converged much faster than the models that utilized pre-trained word vectors.