Fight fake news with the new LetsCheck platform
The LetsCheck website showing the results of a user’s search.
In an age where information can be disseminated quickly through social media, it is easy to communicate with our loved ones, and share stories that inspire. Yet, it is through these same channels that fake news and falsehoods can also spread, especially during the current uncertain times where we are bombarded by a multitude of reports aboutCOVID-19.
To help people verify claims surrounding COVID-19, the NUS Centre for Trusted Internet and Community (CTIC) has developed LetsCheck, an online platform that lets users check claims circulating on Twitter about the coronavirus against reputable scientific sources or news media outlets.
LetsCheck was launched on 31 August by NUS President Professor Tan Eng Chye during a webinar jointly organised by CTIC and the NUS Institute of Data Science.
Prof Tan said, “Misinformation not only hampers an effective public health response, but can also create fear, confusion and distrust among people.
“The launch of NUS CTIC’s LetsCheck platform is a timely one. It is difficult for individuals to assess the credibility of information and to discern between fake and real news. The platform tries to make this more accessible to the public by using AI to help retrieve evidence from independent and authoritative sources to help individuals assess whether a piece of information is true or not.”
The LetsCheck platform contains a clean and intuitive interface that taps on three different sources of information to do fact-checking.
Users can perform a search for claims about COVID-19 that are making rounds on Twitter. The platform generates results for influential tweets – i.e. those with a high number of likes and retweets – containing the search term requested by the users, as well as results from news stories and scientific articles.
A semantic dense matching model enables the platform to retrieve large amounts of data from a wide variety of trusted sources, such as scientific articles. This model allows the team to extract relevant passages from millions of articles on the internet, even from articles which did not directly overlap with the search query. Once the relevant passages have been obtained, language inference will then be performed on them to decide whether the passage supports – or denies – the claim the user has searched for.
For news articles, the team uses a news crawler to go through reputable news websites. The results are cleaned up, put through a semantic encoder to check how closely they match the user’s query, and then ranked according to their relevance.
“Fake news is quite hard to spot as it is usually a mix of fact and fiction, making them very believable. Stopping the spread of fake news is even harder. Once a post goes viral, it is difficult to contain it, and millions of people will have been affected by this fake news before anything can be done,” said Professor Wynne Hsu, Director of the Institute of Data Science, who is the Principal Investigator of the LetsCheck project.
“It is very important for us to be able to quickly and efficiently mitigate the spread of fake news on social media,” she added.
Try out the LetsCheck platform here.