The FakeRank score represents the probability of content to be safe or questionable based on our Machine Learning model – FakeRank. It is calculated using deep neural networks and unique psycho-linguistic cues that are able to capture the psychology and semantic meaning of text to automatically detect the various kinds of disinformation and hate speech.
Data is one of the most important elements of a machine learning task. It is especially challenging in this task, where quality labeled data is scarce and the and ground truth established by training algorithms on human work is neither universal nor permanent. Through our collaboration with fact checking partners, we are able to collect unique data which serves to build better models and attract more partners and more data.
Since models are only as good and unbiased as the data they use, the FakeRank AI is trained on data labeled by fact-checking organizations that comply with the IFCN code of principles of Nonpartisanship, Fairness and Transparency. Moreover, the FakeRank AI is not a black-box. We invested in developing explainable AI that provides a score, but more importantly, insights into the logic that resulted in that scoreWhat are the different types of Fake News FakeRank can identify?
FakeRank distinguishes between 3 main categories of Fake News:
It's important to emphasize that FakeRank is not aiming to be the source of truth. We do not provide judgement. We provide measurement. More specifically, we provide AI to empower humans to address the scale of the problem (as opposed to manual checking).