AI-based Topic Modelling for Financial Disruption Analysis: Herd Behaviour and Sentiment Shift in the NFT Market Crash on X (Twitter)

This study examines herd behaviour and fandom dynamics in the NFT market during the 2021 crash by analyzing 184,257 X posts across pre-crash, peak-decline, and post-crash phases. Using LDA, NMF and BERTopic alongside RoBERTa sentiment analysis, we capture evolving topics and emotional trajectories in slang-rich social media discourse. We found that sentiment shifted from strong pre-crash optimism (82.8%) to polarised reactions during the crash, followed by predominantly neutral, risk-averse tones (54.9%) post-crash. Topics reveal transitions from investment enthusiasm and project hype to security concerns, fund recovery, and scam awareness, reflecting changing community priorities. The findings show how fandom and herd behaviour jointly shape sentiment in volatile digital asset markets, highlighting the role of social identity and collective behavioural mechanisms in emotional contagion and instability. Managerially, the results highlight the need for transparency, fraud prevention, and proactive sentiment monitoring to support trust, early risk detection, and more informed engagement.