A Comparison of Machine Learning Approaches for Detecting Misogynistic Speech in Urban Dictionary

Recent moves to consider misogyny as a hate crime have refocused efforts for owners of web properties to detect and remove misogynistic speech. This paper considers the use of deep learning techniques for detection of misogyny in Urban Dictionary, a crowdsourced online dictionary for slang words and phrases. We compare the performance of two deep learning techniques, Bi-LSTM and Bi-GRU, to detect misogynistic speech with the performance of more conventional machine learning techniques, logistic regression, Naive-Bayes classification, and Random Forest classification. We find that both deep learning techniques examined have greater accuracy in detecting misogyny in the Urban Dictionary than the other techniques examined.