How We Classify News Sentiment

Every article we show on a stock's page gets run through a sentiment classifier that answers one question: does this article read as positive, negative, or neutral specifically for this company?

The pipeline

1. We pull recent headlines for the ticker from Google News.
2. Each headline and description is sent to a fast primary classifier.
3. If that classifier is unavailable or times out, a backup model picks up the slack automatically.
4. Results are cached, so the same article is never re-analyzed twice.

Why sentiment, not just headlines

A headline alone can be misleading — "Company X sued over patent dispute" sounds bad, but if the suit is minor and the stock barely moves, that context matters. Sentiment analysis distills the tone of the whole article, not just the first ten words.

We're continuing to refine the prompts and models behind this pipeline. If you notice a sentiment call that looks off, we'd love to hear about it.