What Is Sentiment Analysis, and Can It Really Measure Diplomacy?
2026-09-03 · by Jordi Fuster
"Sentiment analysis" comes up a lot on this site, usually in one sentence and then never explained. Here's the fuller version: what it actually is, where the technique came from, and — honestly — where it runs into trouble.
The basic idea
Sentiment analysis is a set of techniques for automatically deciding whether a piece of text is positive, negative, or neutral, and roughly how strongly. The simplest versions work the way you'd guess: a dictionary of words scored in advance for how positive or negative they typically read ("catastrophic" scores very differently from "modest"), applied word by word across a passage of text and combined into an overall score. More advanced versions, built on modern language models, try to account for context, negation ("not good" isn't the same as "good"), sarcasm, and intensity — with real, if imperfect, success.
WorldRelat uses VADER (Valence Aware Dictionary and sEntiment Reasoner), a well-established, open-source tool originally built for exactly the kind of short, informal, real-world text you find on social media — but it generalizes reasonably well to the more formal, encyclopedic prose that WorldRelat actually feeds it. VADER's particular strength is that it's tuned to handle punctuation, capitalization and degree words (the difference between "good" and "extremely good") without needing a huge, expensive model behind it, which is exactly the trade-off a site running sentiment analysis across thousands of country pairs actually needs.
Where this technique gets used elsewhere
This isn't a niche idea. Financial firms run sentiment analysis over earnings calls and news flow to gauge market mood before it shows up in price. The GDELT Project applies it at a global scale to monitor how world events are being covered as they unfold, across effectively every language and country, in close to real time. Customer service platforms use it to flag an angry support ticket before a human ever reads it. What all of those share with WorldRelat is the same basic premise: language itself carries a measurable signal about how something is actually going, even before anyone states it directly.
Where it genuinely struggles
It's worth being honest about the limits, because they're real. Sentiment analysis reads tone, not truth or stakes — a short, dry, matter-of-fact account of a serious conflict can score more neutrally than a long, dramatic account of a comparatively minor dispute, simply because the language used is calmer. Sarcasm and understatement — genuinely common in diplomatic and political writing — can flip a sentence's real meaning without changing its literal words enough to fool a dictionary-based model. And sentiment analysis has no concept of context outside the text itself: it can't know that a given phrase means something different this year than it did five years ago, because the words on the page haven't changed even if the world around them has.
That's exactly why WorldRelat treats its score as a starting point rather than a verdict — a consistent, comparable read across every country pair, not a claim to have solved a genuinely hard problem. It's also why the site leaves room for real people to push back: the vote-adjusted "Combined score" on every relation page exists specifically because a language model, however good, doesn't get the final word on how a relationship actually feels to live through.