At INTERFACEing2024 (National Taiwan University), Wanwan Zheng (Group 1), Yasuko Nakamura (Group 1), and Kazunori Hayanagi (Nagasaki University) organized and presented a panel titled “Narrative Sentiment Analysis using Language Models.”
In this panel, the presenters explored the possibilities and challenges of emotion analysis using large language models (LLMs), focusing on the complexity of emotions found in literary works and policy speeches. We aimed to analyze latent and multilayered emotional expressions that cannot be fully captured by a simple “positive/negative” dichotomy. In addition to sentiment analysis, discussions were held on the transition of topics within documents, text understanding, and the interpretation of how expressions of subjectivity can be combined with emotional data.
During the Q&A session, there was a lively exchange of opinions on the relationship between emotion and meaning, models and traditional humanistic interpretation, and many insights were gained on the possibility of connecting interpretability in machine learning with humanistic reading. All of the presentations from this panel were further examined and included in the first volume of the series.
