Tackling Sexism in Social Media: Multilingual AI Solutions

Abstract

Ron Keinan

In this paper, we describe our submission to the EXIST-2024 contest. We participated in Task 1: "Sexism Identification in Tweets" in both English and Spanish. To classify the tweets for sexist content, we developed various models by altering the machine learning classifier, feature type (word/character n-grams), feature quantity, and text preprocessing steps. We then vectorized the text using the TF-IDF embedding technique. After training these configurations on the training dataset, we selected the best models based on accuracy and F1-score on the development set and used them to predict the test labels. Our top-performing model achieved an F1 score of 72.23, securing 39th place out of 70 participants.

PDF