Description

📘 “Dynamic Topic Modelling for Bitcoin Message Fora” (Bommes, Chen, Teo, Härdle, Linton, 2017)

1️⃣ Bitcoin message boards contain massive unstructured “dark data” with evolving themes.
2️⃣ Dynamic Topic Modelling (DTM) applies Bayesian LDA to track word–topic evolution over time.
3️⃣ Forums (e.g. bitcointalk.org) are parsed into weekly corpora from 2009–2016.
4️⃣ Topics correspond to latent themes such as mining, trading, scams, technology, regulation.
5️⃣ Temporal DTM smoothing reveals topic prominence trajectories linked to major events.
6️⃣ Mt.Gox collapse (2014) shows clear topic transitions: from “trading” to “scams & hacks.”
7️⃣ Topic intensities validate event detection against known hack and scam lists.
8️⃣ The DTM captures collective sentiment shifts—a proxy for herding and market attention.
9️⃣ “Words become numbers”: text mining bridges crypto community chatter and financial indicators.
🔟 DTM offers an interpretable, unsupervised, early-warning framework for cryptocurrency ecosystem risks.