News sentiment analysis measures whether coverage of a topic is positive, negative or neutral — and, more usefully, how that changes over time and between countries. Traders use it to watch the mood around companies, communications teams to track reputation, researchers to study how media frames events. This guide explains how it works and how to avoid the classic misreadings.
What "tone" actually means
Tone is a number attached to an article that summarises the emotional colour of its words. "Record growth", "breakthrough" and "agreement" push it up; "crisis", "killed" and "collapse" push it down. It does not measure whether an article is fair, true or biased — only how emotionally loaded its language is.
Three ways to measure it
- Dictionaries (lexicons). Count positive and negative words from a large curated list. Fast, transparent and language-scalable; blind to irony and context. GDELT's tone is of this type.
- Machine-learning classifiers. Models trained on labelled examples (for example, financial headlines marked positive or negative). More accurate in their domain, weaker outside it.
- Large language models. Can judge sentiment towards a specific target ("negative about the company, positive about the regulator") and explain why. Slower and costlier, best for smaller samples or for checking the other methods.
In practice, good systems combine them: a dictionary score over millions of articles for trends, and an LLM on a sample to understand why tone moved.
How Mirada measures tone
Mirada uses GDELT's dictionary-based tone for every article, on a scale from −10 (very negative) to +10. On the site, an average below −1.5 is labelled "negative" and above +1 "positive". Each topic page shows the average tone for 30 days and the tone of each recent article; the methodology is described on the methodology page.
How to read tone scores correctly
- News is negative by default. Conflicts, disasters and crime dominate. An average of −2 is ordinary; what matters is change.
- Compare, don't judge. Tone is meaningful relative to something: last month, another country, another outlet, another company.
- Watch volume and tone together. Rising coverage with falling tone is the classic signature of an emerging problem. Rising coverage with stable tone is often just a scheduled event.
- Mind the topic itself. A topic like earthquakes is always negative; that says nothing about bias.
- Check the headlines behind a shift. A tone drop is a reason to read, not a conclusion.
Where sentiment analysis is useful
- Markets. Tone and volume around companies, commodities and central banks move with news shocks; many funds use news sentiment as one signal among many.
- Reputation and PR. Brands and public bodies track whether coverage turns negative and in which countries.
- Politics and research. Comparing how media in different countries frame the same leader, election or conflict.
- Personal news reading. Tone helps spot topics that are heating up before they hit the front pages.
Limits
Dictionary tone misses sarcasm, negation ("not a disaster") and domain language (in finance "volatile" is not always bad). Machine translation can shift tone for non-English articles. And tone of a headline about a tragedy is negative even when reporting is balanced. Treat sentiment as a thermometer, not a judge.
FAQ
What is news sentiment analysis?
Automatic scoring of how positive or negative news coverage is, usually tracked over time, by topic, outlet or country.
Is news sentiment useful for stocks?
As one input among many. Sudden changes in coverage volume and tone around a company often accompany price moves, but sentiment alone is not a trading strategy.
How accurate is GDELT tone?
Good for averages over many articles and for comparisons; unreliable for judging a single article because dictionaries miss context and irony.
Can I see the tone of coverage for any topic?
Yes. Search the topic on Mirada: the page shows the 30-day average tone and the tone of the latest articles.