"Exploring To the Lighthouse through Digital and Public Humanities" Text segmentation, Emotional Analysis and Interactive Website in Virginia Woolf's novel to explore the potential of GPT-4

This thesis aims to conduct a comparative analysis of the human and AI capability in understanding Virginia Woolf’s work “To the Lighthouse”, with the goal of assessing the machine’s ability to segment text and analyze emotions, and develop an interactive website for visualizing emotional trends in the novel. Specifically, in the first task, we ask a large language model (ChatGPT-4) and a group of human readers to segment the first chapter of the novel. In the second task, we ask instead to assess the emotional perception conveyed by three different passages of the novel, considering a spectrum of six different emotions. In particular, we found that ChatGPT-4 aligns more with STEM-background persons in terms of segmentation capabilities, while people from humanities adopt a segmentation approach that is more adherent to the editorial structure of the text. In the Emotion Analysis, the results obtained show a high correlation between the two samples, with no substantial differences between the male and female participants; though notable differences arise in its emphasis on subtle emotions, such as sadness and fear, where the model tends to amplify symbolic and descriptive elements compared to human readers. Finally, we include a navigable web resource that showcases visualizations of the emotional analysis conducted by ChatGPT-4. This platform traces the emotional evolution of the characters in the text and identifies key emotional dynamics at specific narrative moments, enabling an in-depth exploration of the potential of AI as an assistant for literary analysis.