Sentiment Analysis
News & social sentiment scoring across prediction markets
Two different things in Predite are called sentiment, and they work nothing alike. Knowing which one you are looking at is the whole point of this page.
What you actually see: article labels
In the News tab of a market page, and in the News feed, every headline carries a positive, negative or neutral tag. That tag comes from a keyword pass over the headline text — roughly a dozen bullish words against a dozen bearish ones, counted and compared. There is no model behind it.
That makes it useful for one job and useless for another. Useful: skimming twenty headlines and seeing at a glance where the tone clusters. Useless: deciding direction. A headline that reads negative for one side of a contract is positive for the other, and a keyword counter has no idea which side you hold. "Rejected" is bearish for the market titled *Will it be approved?* and bullish for the one titled *Will it be blocked?* — same word, opposite trade.
What exists but is not on screen: topic sentiment
Predite also has a topic-level read: given a market question, the AI estimates what social sentiment around that topic probably looks like and returns bullish, bearish, neutral or mixed, plus a score from -1 to +1. It is cached for an hour and limited to 20 requests per hour per user.
Three things to be plain about:
- •It is not wired to any page yet. The endpoint exists; no screen calls it. If you are looking for a sentiment gauge on the market page, it is not there — what is there is the article labels above.
- •It is an estimate, not a measurement. The AI reasons about what the conversation is probably like from what it learned in training. It does not read Twitter or Reddit: there is no scraping and no social API connected, and the response is always marked as an estimate.
- •The scale is -1 to +1, not a percentage and not a hundred-point index.
How to use tone without being used by it
Sentiment is a *timing* hint, never a thesis. The honest workflow:
- Read the headlines in the News tab to know what happened — the labels only order your attention, they do not judge for you.
- Check whether the price already moved. Tone that everyone can see is tone that is already priced.
- Form your own view of the probability, then compare it to the price in the EV Scanner.
The trap this page exists to prevent: reading a wall of red labels as "buy NO". Crowd tone is loudest exactly when it is most priced in, and a keyword counter cannot tell fear from information.