Sentiment Analysis of Nakba Oral Histories: A Critical Study of Large Language Models

Abstract

This study explores the use of Large Language Models (LLMs), specifically ChatGPT, for sentiment analysis of Nakba oral histories, which document the experiences of Palestinian refugees. The study compares sentiment analysis results from full testimonies (average 2500 words) and their summarized versions (300 words). The findings reveal that summarization increased positive sentiment and decreased negative sentiment, suggesting that the process may highlight more hopeful themes while oversimplifying emotional complexities. The study highlights both the potential and limitations of using LLMs for analyzing sensitive, trauma-based narratives and calls for further research to improve sentiment analysis in such contexts.