What sentiment scoring actually is
Sentiment analysis reads the tone of what is being written about a ticker and reduces it to a number. That is the whole mechanism. It is genuinely useful and it is routinely oversold, so it is worth seeing exactly how a score becomes a signal — using ours, because we can show the arithmetic rather than describe it.
From tone to a direction
Each event about a ticker — a headline, a post, a filing — carries a raw sentiment value. Those get averaged over the timeframe you are looking at and normalised to a 0–1 scale, where 0.5 is neutral.
A score on its own does not produce a call. Direction requires two conditions to agree: sentiment above 0.65 and a trend that is improving gives bullish; below 0.35 and declining gives bearish. Anything else is neutral, and neutral is the common case by design.
The trend half matters more than the level. A ticker sitting at 0.7 that has been drifting down from 0.85 is a different situation from one at 0.66 climbing off 0.4, and a threshold alone cannot tell them apart. The comparison is literally recent average against older average over the same window.
What the confidence number is made of
Confidence is not a probability that the call is right. It is a blend of two things, weighted:
- Sentiment strength (60%) — how far from neutral the score sits. A reading of 0.9 and a reading of 0.52 are both "positive" and carry completely different weight.
- Data confidence (40%) — how much material the score is built on, rising with event count and topping out around twenty events. Three articles and a strong opinion is not the same evidence as forty.
So a high-confidence call means the signal was strong and there was enough of it. A ticker nobody is writing about cannot produce a confident reading no matter how positive the handful of mentions are, which is the correct behaviour and worth knowing when a thinly covered small cap shows up quiet.
The part most tools hide: where the price target comes from
This is the detail worth taking away. The predicted price is derived from the confidence number, not forecast independently. Conviction maps to a band of roughly 1% to 5% from the current price, in the direction of the call.
That means the target is an expression of how strongly the signal reads — not a model's opinion about where the stock is going to settle. Read it as "the signal is strong enough to be worth about this much move", not as a price prediction. Any tool presenting a precise target without telling you how it was produced is asking for more trust than the method supports, and we would rather show the arithmetic.
Read the reasoning, not the number
Every prediction carries a plain-language explanation of its own inputs: how many events were analysed, how many distinct sources they came from, the average sentiment, and the trend comparison that produced the direction.
That string is the most useful thing on the object. "Bullish, high confidence" is unfalsifiable. "Analysed 34 events from 5 sources, average sentiment positive, trend improving from 0.41 to 0.72" is something you can disagree with — you can look at those events and decide the model read them wrongly. Disagreeing with the premise is far more productive than disagreeing with the conclusion.
What sentiment genuinely cannot do
It cannot see anything nobody has written about yet, which is most of what moves a stock on any given day. It reads loudness as well as substance, so a coordinated push and genuine enthusiasm look similar. And it has no concept of whether the thing being discussed is already in the price.
Treat a sentiment shift as a prompt to go and look, not as a reason to act. The predictions resolve to correct or incorrect once their timeframe closes, so the honest way to judge any of this is to read the record rather than the marketing — ours included.
See the live scores and their reasoning on the predictions page, or create a free account.
Nothing here is financial advice. Sentiment scores are estimates derived from public text and do not indicate future price movement.