Social Media Sentiment Analyzer

A personal project about how opinion actually moves through a network.

I built API pipelines that pulled more than 10,000 Mastodon posts and 1,500 profiles, ran Llama3-8B over them for toxicity and sentiment scoring, and then mapped influence with NetworkX graphs: PageRank, degree centrality, and cluster entropy.

The finding that stuck with me is that loudness and influence are nearly uncorrelated. The accounts that moved sentiment were rarely the ones posting most. The same is true in markets, which is partly why this project exists in the same ledger as my trading work.