Tsukuba, Japan—Online chat rooms and social networking platforms frequently experience harmful behavior as discussions drift from their intended topics toward personal conflict. Traditional predictive models typically depend on platform-specific data, limiting their applicability and increasing implementation costs.
In this study, the researchers applied a zero-shot prediction method to LLMs to detect conversational derailments. The performance of various untrained LLMs was compared to that of a deep learning model trained on curated datasets. The results showed that untrained LLMs achieved comparable, and in some cases superior, accuracy.
These findings suggest that platform operators can implement effective moderation tools at reduced cost by leveraging general-purpose LLMs, supporting healthier online communities across diverse platforms.
Title of original paper:
Zero-Shot Prediction of Conversational Derailment With Large Language Models
Journal:
IEEE Access
DOI:
10.1109/ACCESS.2025.3554548
Associate Professor YOSHIDA, Mitsuio
Institute of Human Sciences, University of Tsukuba
NONAKA, Kenya
Doctoral Program in Risk and Resilience Engineering, Degree Programs in Systems and Information Engineering, University of Tsukuba
Institute of Business Sciences
Master's / Doctoral Program in Risk and Resilience Engineering
IEEE Access
Zero-Shot Prediction of Conversational Derailment With Large Language Models
25-Mar-2025