Contextualized Early Detection of Online Firestorms: A Sequential LLM-Based Approach
Artificial Intelligence
Summary
The authors developed a system using large language models (LLMs) to detect online firestorms—rapid bursts of negative content—on Reddit. Their system has two ways to work: one looks at entire discussion threads after they happen to decide if a firestorm occurred, and the other monitors ongoing conversations to give early warnings. The early warning mode tracks negativity, how fast negativity grows, and how many users are involved. Their tests show the system can accurately spot firestorms and issue warnings quickly using only a few comments. This suggests LLMs can help monitor social media discussions in real time, not just analyze them after the fact.
Authors
Besim Shala, Peter Mandl, Andreas Humpe, Martin Häusl
Abstract
Online firestorms are rapid collective escalations of highly negative user-generated content and may cause substantial reputational and economic damage. Existing detectors usually work with volume signals, sentiment scores, or predefined linguistic features. Such signals are useful, but they capture contextual meaning shifts in evolving discussion threads only indirectly. This paper proposes an LLM-based detection system with two operating modes. The first mode classifies complete Reddit threads retrospectively by combining local chunk-level assessments into a thread-level judgment. The second mode processes threads sequentially and issues early warnings when a sliding window exceeds calibrated thresholds. In this mode, the language model estimates three firestorm indicators: negativity share, escalation level, and contributor count. On a balanced Reddit dataset, the global mode achieves strong classification performance, while the early warning mode reaches high recall and detects escalating threads after only a small number of comments and distinct contributors. The results indicate that LLMs can be used not only for static judgment tasks, but also as repeated estimators in context-aware monitoring of social media discourse.