Automatic Identification of Misogyny in Social Networks

Abstract

The present research work focuses on the automatic detection of misogyny in unstructured texts, specifically on the Twitter platform, from which two datasets were analyzed: Evalita and HATEVAL, using different supervised learning techniques, Convolutional Neural Networks (CNNs), and a meta-classifier that implemented and combined the models in each dataset. The results showed that the meta-classifier outperformed the base classifiers and the convolutional neural networks, with an accuracy of 95.3% in Evalita and 93.7% in HATEVAL, thus highlighting the importance of data preprocessing for obtaining accurate results.