eXplainable AI with GPT4 for story analysis and generation
Brings explainable AI to narrative, a domain where XAI had been largely confined to image classifiers, and supplies the sentence-level story-trajectory methodology that lets narratives of unequal length be compared on shared coordinates.
- Full title. eXplainable AI with GPT4 for story analysis and generation: A novel framework for diachronic sentiment analysis
- Authors. Jon Chun; Katherine Elkins
- Venue. International Journal of Digital Humanities 5(2-3), 507–532
- Published. 11 October 2023
- Identifier. doi:10.1007/s42803-023-00069-8
Summary
Explainable AI had been developed largely for image classification, where a saliency map can show which pixels drove a decision. Narrative has no equivalent. This paper develops an XAI framework for story analysis and generation with GPT-4, working at the level of the sentence rather than the document. It produces story trajectories that can be inspected and argued with, and it supplies the agglomeration and comparability solutions that place narratives of unequal length on shared coordinates.
What this work contributed
- XAI for narrative. Explainability techniques were brought to a domain where they had been largely confined to image classifiers.
- Sentence-level story trajectories. A methodology for tracing a story's trajectory at sentence granularity.
- Comparability across unequal lengths. Agglomeration and comparability solutions that let narratives of different lengths be compared on shared coordinates.
Relationship to earlier and later work
It builds on the ensemble pipeline and DTW distance introduced in SentimentArcs (2021), and is carried across modalities by MultiSentimentArcs (2024). See Narrative trajectories & DTW for the full lineage and Reception for scholarly uptake.
Cite
Plain text. Chun, Jon, and Katherine Elkins. “eXplainable AI with GPT4 for story analysis and generation: A novel framework for diachronic sentiment analysis.” International Journal of Digital Humanities 5, no. 2-3 (2023): 507–532.
BibTeX.
@article{chun2023xai,
title = {eXplainable AI with GPT4 for story analysis and generation:
A novel framework for diachronic sentiment analysis},
author = {Chun, Jon and Elkins, Katherine},
journal = {International Journal of Digital Humanities},
volume = {5},
number = {2-3},
pages = {507--532},
year = {2023},
doi = {10.1007/s42803-023-00069-8}
}