Misogyny Speech Detection Using Long Short-Term Memory and BERT Embeddings

Freedom of speech on social media is often abused to express inappropriate speech against individuals or groups. One form of social media abuse is misogyny utterance. Misogyny is hatred directed at women. The number of misogyny utterances on social media keeps increasing. Detection of misogyny needs to be done to prevent misogyny utterances. This work examines misogyny on Twitter and proposes a method to detect it. The dataset used comes from tweets on social media Twitter related to women. The experiment in this work was done by comparing the effect of BERT Embedding for Logistic Regression, Convolutional Neural Network, and Long Short-Term Memory. The performance of the methods tested was evaluated based on the resulting accuracy. This work showed that BERT with LSTM surpassed other methods with 86.15% accuracy. The model could outperform other methods because it understands more about the context of the data.