Date created: | Last Updated:

Category: Project

Description: Sentiment analysis is widely used to classify emotional tone in text, yet its application to autobiographical narratives—emotionally complex, self-referential memories—is underexplored. Existing sentiment dictionaries were developed using domains such as product reviews and social media posts, raising concerns about their suitability for autobiographical data. In this study, we evaluate the performance of five widely-used sentiment dictionaries (GI, HE, QDAP, VADER, and TextBlob) using a dataset of 272 autobiographical narrations. A supervised machine learning approach employing cost-sensitive learning and conditional inference trees was used to assess dictionary performance against self-reported emotional ratings. Results reveal that no single dictionary consistently outperforms across all metrics; different dictionaries show strengths in accuracy, recall, or specificity depending on the research goal. The findings underscore the need for task-specific evaluation of sentiment tools and caution against relying solely on accuracy. Implications for dictionary selection, limitations, and directions for future research are discussed.