Hoax Analyzer, give me a hoax
This experiment aims to select the best algorithm for classifying hoax and non-hoax news in Indonesian and English using text-mining and machine learning methods.
Text Mining
Filtering, stopword removal, and stemming to clean the text before classification.
Machine Learning
Comparing algorithms to pick the best hoax / non-hoax classifier on ID & EN data.
Fact-Check Archive
Every check is matched against thousands of published fact-checks, so a claim that was already debunked is recognized at once, with the evidence.
11,141 fact-checks, 2000–2022
Collected from 13 fact-checking sources: 9,202 in Indonesian and 1,939 in English. Every entry keeps its verdict and a link to the original article.
- Hoax 8,224
- Misleading 1,405
- Credible 1,512
Sources
12,937 labeled examples
The original detector learned from pages collected by its crawler, each labeled with the class the fact-checkers gave it. Another 4,189 collected pages were never labeled.
- 1. Your text: pasted, read from a link on the server, or read from a screenshot on your device.
- 2. The archive: the rare words in your text are matched against every fact-check, so a claim that was debunked before is found even when it is worded differently.
- 3. The verdict: comes from the trained detection model when it is connected. Until then, the closest matching fact-check decides, and a simple keyword check covers texts with no match; the result says which one answered.
- •It reads text. A screenshot is turned into text on your device first, so blurry or stylized images can be misread; check the text before analyzing.
- •It compares your text with fact-checks published up to 2022. A brand-new claim may not be in the archive yet.
- •A verdict is a prediction, not proof, and cannot serve as legal evidence. Always read the fact-check it points to.
- •Several source sites have closed or moved since their articles were collected. Their links open the Internet Archive's copy instead.
- •Indonesian and English are supported. Other languages are not analyzed reliably.
Hoax Detection for Indonesian and English Text
The ongoing research focuses on detecting Indonesian hoax news by combining a classifier trained on thousands of published fact-checks with a search of the archive itself, so every verdict comes with references to verify it.