Ideological clustering on Bluesky
A study of engagement across 2,900+ users and 77 feeds to understand ideological clustering, overlap, and filter-bubble behavior.
2,900+ users across 77 feeds
Research question
How strongly do users cluster around ideologically aligned custom feeds, and where do communities still overlap?
Approach
I collected engagement signals across more than 2,900 users and 77 Bluesky feeds. The analysis combined feed metadata, likes, replies, reposts, recent creator content, LLM-assisted political-orientation classification, embeddings, and cosine similarity.
Engineering work
- Built batch pipelines for feed likes and cross-feed user reactions.
- Collected creator account details and recent public posts.
- Normalized feed and user data into analysis-ready datasets.
- Used embeddings to measure semantic relationships between feeds and communities.
- Identified clustering patterns, thematic overlap, and possible filter bubbles.
Outcome
The project connected a social-science question to a reproducible data pipeline, making both the research method and its limitations inspectable.