Leveraging Explainable Artificial Intelligence for Comparative Large Language Models in Body Shaming Detection

This study introduces a novel methodology to detect Body Shaming using Explainable Artificial Intelligence (XAI) on social media content. By combining Fine Tuned and partially tuned Large Language Models (LLMs), along with Natural Language Processing (NLP) techniques, the system aims to enhance the early detection of Body Shaming. To optimize performance, preprocessing steps include stop word removal, tokenization, lemmatization, and emoji substitution. Multiple LLMs (Fine Tuned PubMed BERT, Partially PubMed BERT, Fine Tuned XLNet, Partially XLNet, Fine Tuned Psycho RoBERTa, and Partially Psycho RoBERTa) are trained on annotated data to identify Body Shaming language in Reddit posts. Integrating XAI with Local Interpretable Model-Agnostic Explanation (LIME) enhances interpretability, ensuring transparency in Body Shaming detection. Results indicate high accuracy in identifying Body Shaming tendencies in social media language, with a Fine Tuned Psycho RoBERTa achieving 0.9818678 accuracy and 0.9940621 F1-Score. Overall, the study highlights the efficacy of NLP and LLMs in diagnosing body shame using social media data, showcasing innovative XAI applications in mental health.