Classifying Emotional Dynamics In Short Narratives: Techniques and Applications
Elliott, Jeremy
Elliott, Jeremy
Citations
Altmetric:
License
License
Abstract
Emotional dynamics are the conscious choice of language in a text that gives each section of the text a distinct emotional characteristic. In a large text, such as a novel, this may be used to evoke certain emotional states of the reader as they progress through the text. Emotional dynamics have been studied in depth for large texts using per-word emotional scores, showing the emergence of six distinct archetypes (Reagan et al. 2016), but techniques for shorter narratives are less developed. We show that the techniques developed for large texts break down for shorter texts, but the same per-word emotional scores can still be used to classify them by archetype. Using a two-stage classifier, we first predict the perceived sentiment of a passage based on summary statistics of the component per-word happiness scores, and then classify the archetype based on the progression of perceived sentiment in the text. Our corpus consists of letters written to one�s descendants about climate change, each with five sections. The perceived sentiment of each section was human coded and used to train our model, which we then applied to texts from a previous study to mimic human perception of sentiment. In this poster, we present the full pipeline for cross-disciplinary replication and demonstrate how to interpret and apply the results to one�s own work.
Description
Date
1/1/2026
Student Status
Graduate Student
Journal Title
Journal ISSN
Volume Title
Type of presentation
Poster
Collections
Research Projects
Organizational Units
Journal Issue
Citation
DOI
Advisor(s)
Department
Program/Major
Complex Systems and Data Science
College/School
College of Engineering and Mathematical Sciences
Organization
Research Category
Social Science