🎲 Demo checks every click brings a NEW example, live from the internet
Try a held-out test example:
Fake news detector
Check a story before you share it
Give Verity any piece of news. Eight trained models read it, our proposed HyFuse model combines their opinions, and you get a verdict with the proof behind it.
✍️ Typed text📄 PDF / Word
🖼️ Screenshot🔗 Web link✉️ Email (.eml)📰 Live fact check
🎯Try it with a real test example
These come from the held-out test sets, which the models never saw during training. After the check you can see whether the model got it right.
⚙️How Verity decides, in 5 steps
- Read the inputText is taken out of whatever you give: typed text, a PDF or Word file, a screenshot (read with OCR) or a web page.
- Pick the right expert teamA long article (70+ words) goes to the models trained on full articles (ISOT). A short sentence goes to the political-claim models (LIAR). A headline goes to the headline models (FakeNewsNet).
- Eight models vote on their own TF-IDF counts tell-tale words, Word embeddings look at word meaning, Style checks how it is written, and BERT reads every word in the context of the whole sentence.
- Live fact check with trusted sourcesVerity splits the text into facts and searches Wikipedia, Google News and fact-checking sites. A fact-checking model reads each source sentence and decides whether it confirms or contradicts the fact.
- HyFuse gives the final answerOur proposed model has learned which voter to trust in which situation, especially when they disagree. It combines the votes into one score from 0% to 100% fake, and strong evidence from trusted sources then moves that score up or down. With no evidence, the models decide alone.
🧭How to read the result
Likely realCan't tellLikely fake
- Fake score
- How likely the text is fake, from 0% to 100%.
- Confidence
- How far the score is from the 50% middle. 0% means a coin toss, 100% means certain.
- Agreement
- How many of the models vote the same way as the final answer.
- Proof
- The words, sentences and writing habits that pushed the score up or down.
🔎What counts as evidence
Signs of fake news
- ALL-CAPS "shouting" and lots of !!! or ???
- Sensational words: shocking, exposed, destroyed
- Vague claims: reportedly, allegedly, some say
- Opinion voice: I, you, we; emotional tone
- Words that were common in fake training news
Signs of real news
- Names its sources: said, according to
- Calm, neutral reporter-style tone
- Specific facts: dates, numbers, places
- Few exclamation marks or capitals
- Words that were common in real training news
The models judge how a story is written (learned from over 70,000 labelled news items). The live fact check adds what trusted sources say. Together they cover both new rumours and checkable facts.