Penta-nlp at EXIST 2024 Task 1-3: Sexism Identification, Source Intention, Sexism Categorization In Tweets

Abstract

Social media platforms contains a vast user base and offers ease with which information can be shared. This can adversely facilitate the spread of sexist content which is infeasible for human monitoring and filtering. This paper investigates the automated detection of sexism in tweets using Natural Language Processing (NLP) techniques. Sexist tweets can create a hostile online environment and perpetuate harmful stereotypes. Manual identification is impractical due to the vast amount of data. The research proposes a system utilizing machine learning models to analyze text, identify bias and discriminatory language patterns, and flag tweets for moderation. The fourth edition of EXIST shared task 2024 Tweets Dataset, containing labeled English and Spanish tweets, is used to train and evaluate the models. The system explores various approaches, including TF-IDF with different classifiers (SVM, XGB, RF), Long Short-Term Memory (LSTM) networks with and without attention mechanisms, and pre-trained transformer models (XLM-Roberta, mBERT, BETO). The effectiveness of different preprocessing techniques and the role of attention weights in identifying sexism are also explored. The paper outlines the methodology, experimental setup, and analysis of results, paving the way for further discussion on error analysis and conclusions in subsequent sections.