🚀 TopicWatchdog – Week 3: Stable Topics with BERTopic

KMeans worked, but cluster IDs kept jumping across retrains. This week I added a Python BERTopic stage with a BigQuery registry → stable topic IDs!

🟢 UMAP + HDBSCAN
🟢 Stable IDs via registry
🟢 Auto-labels with Gemini
🟢 Looker Studio dashboards

📊 3,802 topics → 2,472 mapped, top clusters: migration, economy, climate, politics.

👉 Blog: https://dracoblue.net/dev/topicwatchdog-stable-topics-with-bertopic/

#TopicWatchdog #BERTopic #BigQuery
#Clustering
#MachineLearning
#FediScience

TopicWatchdog (Week 2)

Move beyond string matching: cluster political topics from German shorts with embeddings + KMeans in BigQuery ML.

642 videos → 620 topics → 20 clusters (AfD positions, debates, labor market, climate/economy, corruption …).

Blogpost with SQL & charts:
🔗 https://dracoblue.net/dev/topicwatchdog-embeddings-and-kmeans-clustering-of-topics-claims/

#TopicWatchdog #BigQuery #Embeddings #Clustering #MachineLearning #FediScience

Kickoff: TopicWatchdog (Week 1)

I’m building a reproducible pipeline that collects German political shorts, transcribes them, extracts topics & claims with timestamps, and stores everything in BigQuery for transparent analysis.
First finding: topic names are unstable → next up: embeddings + clustering in BigQuery ML.

Full write-up:

https://dracoblue.net/dev/kickoff-topicwatchdog-extracting-topics-and-claims-from-german-politics-videos/

#TopicWatchdog #BigQuery #MachineLearning #LLM #OpenData #FediScience