Sentiment Score vs Headline Count: Why Volume Is Not Tone

The short answer: a headline count measures attention and a sentiment score measures tone. A stock can be in the news constantly with neutral coverage, or barely covered with sharply negative coverage, and the two measures will disagree in both cases. Neither is a substitute for the other, and a spike in one says nothing on its own about the other.

The confusion is understandable, because most people encounter both as a single vague notion of a stock being talked about.

What a headline count is

It is the number of articles mentioning a company over some window. It is cheap to compute, easy to verify, and hard to argue with. What it measures is attention, which is a real and useful quantity: attention tends to precede volume, and unusual attention on a normally quiet name is genuinely informative that something happened.

What it cannot tell you is what happened. A product launch, a lawsuit, an index inclusion and an accounting restatement all raise the count identically. A count of forty articles is the same forty whether they are congratulatory or damning.

It also has a persistent scaling problem. The largest listed companies generate a heavy baseline of coverage every single day, most of it routine, so their raw counts stay high permanently and carry almost no information. Comparing a mega cap's count to a mid cap's count compares audience size, not news.

What a sentiment score is

A sentiment score reads the text and places it on a scale from negative to positive. Modern implementations use language models trained on financial text specifically, because general purpose sentiment tools mishandle the vocabulary badly. In ordinary English "the company reported a loss" is negative and "shares slumped" is negative, but "the company cut guidance less than feared" is a positive development described almost entirely in negative words, and a general model reads it upside down.

What sentiment scoring measures is tone, aggregated across the coverage. That is closer to the thing most people actually want when they ask how the news looks, and it is also considerably more fragile than a count, for reasons worth being explicit about.

Where sentiment scoring goes wrong

• Sarcasm, hedging and conditionals. Financial writing is full of "if", "may", and "risks that". Classifiers handle these unevenly.
• Recycled coverage. One wire story republished by thirty outlets is one event and thirty documents. Without deduplication, a single article's tone gets counted thirty times and reads as a consensus that does not exist.
• Positive news about a bad situation. A company emerging from bankruptcy protection generates genuinely positive coverage about a genuinely poor position.
• Thin coverage. A score built from three articles is a score built from three writers. Its precision is illusory.

The four combinations

Because they are independent, the interesting reading comes from looking at both at once.

• High count, neutral tone. Routine heavy coverage of a large company, or an event being reported factually before anyone has drawn conclusions. Usually the least informative quadrant.
• High count, strongly negative tone. Something specific and bad has happened and is being reported as such. The clearest case, and the one both measures agree on.
• Low count, strongly negative tone. Something is wrong at a company few people are watching. Structurally the most overlooked quadrant, because raw count based alerting never surfaces it.
• Rising count, tone unchanged. Attention is building without the story having turned yet. Worth knowing, and invisible to either measure alone.

Why both belong in an aggregate

News sentiment is one of the dimensions EquityBias reads for each covered stock, alongside analyst activity, price structure and fundamentals. It is deliberately one input rather than the reading, because tone is the noisiest of the four and the most prone to the failures listed above.

The value shows up in how it interacts with the others. Negative tone alongside improving fundamentals is a different situation from negative tone alongside deteriorating fundamentals, and that gap between dimensions is measured directly as divergence rather than smoothed away in an average. A composite that quietly averages a confident sentiment reading with a contradicting fundamental one hides the only interesting fact in the set.

Background on the measurement itself is in market sentiment: what it is and how to measure it. How attention and price behaviour relate is covered in volume in trading. The parent comparison for this cluster is bias score vs analyst rating.

The practical version

When a stock shows up on a news spike, the useful question is not how many articles there are. It is whether the tone moved with the count, and whether the other dimensions moved at all. A count spike with flat tone and unchanged fundamentals is usually a story about attention. A count spike with tone collapsing and fundamentals unchanged is a story about a specific event whose consequences are not yet in the reported numbers.

EquityBias is a market data research tool. Nothing here is financial advice. Sentiment readings describe current coverage and are not predictions of price movement.