Sorting Facts from Fiction

Misinformation. Disinformation. Hoaxes. Fake news can generate a disproportionate amount of confusion and discord in society. To tackle this problem, we collaborated with researchers from the Massachusetts Institute of Technology (MIT) to find new ways to separate facts from fiction.

Digital-InfoPC-Sorting Facts from Fiction 01

As part of the project, our technologists developed an automated fact-checking prototype. It compares dubious claims against a multitude of data sources such as news outlets and government websites to determine a confidence score that shows whether the claims could be true or false. The prototype even backs up its assessment with an explanation of which parts of the claim could be true, and which parts are likely to be false.

The artificial intelligence (AI) algorithm developed together with MIT researchers was trained extensively to ensure that it could perform a multitude of tasks well to achieve more accurate results. Its tasks included selecting relevant data sources, comparing the claims with the facts, and aggregating evidence to arrive at a final verdict. Things were made more complicated by the diverse range of information features that might require advanced fact checks, such as complete fabrications, half-truths, or simply information being used out of context.

Speaking about this challenging process, Director (Tech Devt & Collaboration) Ong Khoon Kiat said: “Algorithms which performed well for one type of fake news didn’t perform as well for others. We worked together with our MIT partners and tried a novel approach where we analysed claims from multiple perspectives for their truthfulness. For example, is the location where an incident supposedly happened correct? Is the date mentioned correct? Through such analysis, we were able to discover the different types of false claims that could be made.”

Digital-InfoPC-Sorting Facts from Fiction 02
A sample analysis of a claim’s truthfulness by the algorithm.

Another issue the team faced was the lack of data to train and test the AI-driven algorithms. While fact-checking datasets were readily available online, most were geared towards the social context of other countries. This meant that our engineers needed to create the first Singapore-centric fact-checking dataset based on real instances of fake news circulated on social media. This allowed the team to validate that the algorithm could work on real-world data and in Singapore’s context. 

Head (Tech Trawling) Josiah Tan added: “Fake news is nothing new, but the proliferation of the internet has made it even easier to propagate fake news rapidly and on a massive scale. We hope that this joint project with MIT will set the foundation for further exploration and development of more ideas on how to tackle fake news in today’s information age.”