Identifying and quantifying literary intimacy with ChatGPT: a computational reading of Great Expectations

As intimacy has garnered increasing attention in sociology and psychology, its systematic study in literary scholarship remains predominantly rooted in qualitative close reading and interpretive analysis. To address this gap, this study introduces a novel computational framework for identifying and quantifying intimacy dynamics in literary texts. Leveraging GPT-4, we construct the first multi-layered intimacy corpus comprising over 12,000 verbal and nonverbal interaction segments, calibrated against established psychological intimacy scales and validated through zero-shot regression and cross-model transfer tests. Applied to Charles Dickens’s Great Expectations, the framework generates chapter-level heatmaps and dyadic trajectory analyses that both reaffirm canonical literary interpretations and reveal underlying emotional and social tensions, and its outputs are further validated through a human-annotation study. This study demonstrates how large language models can complement traditional literary criticism by offering empirically grounded, multi-scale insights into the emotional architecture of narrative fiction.