I’m excited to share the final poster from my COSMOS summer project, A Comparative Analysis of NLP Models for Hate Speech Identification. In this project, we explored how Natural Language Processing (NLP) can be applied to detect online hate speech and compared several machine learning models to determine which one is most effective for this task.

We focused on the rise of targeted online harassment across various platforms, especially since Elon Musk’s acquisition of X (formerly Twitter), where instances of slurs have significantly increased. Using techniques like TF-IDF for tokenization and testing models like Random Forest, Logistic Regression, BERT, etc. we evaluated their performance in identifying hate speech related to categories such as race, gender, and religion.

Our analysis dives into the accuracy, precision, recall, and F1 scores of a wide variety models and offers insights into the challenges of dealing with false negatives in this sensitive area. We also explored how these models can be applied to automatically flag harmful content on social platforms in the future.

Check out the full poster below for a deeper look into our research!