News relating the COVID-19 outbreaks has gained a lot of presence online.  As a lot of news sites base their profits on clicks, disinformation and clickbaits are mass-produced to grab any possible attention people have. At the same time, reliable sources try to cover the case of COVID-19 and inform the population about the risk and how to prevent spreading it. Sometimes, it can be difficult for people to differentiate the two types of information. It can lead to misinforming people about the COVID-19 crisis or create anxiety among those people. For the 2020 COVID-19 hackathon we propose a tool to help people know if a news article can be trusted or not. This way, we can solve above problems.

We fine tuned a pretrained version of BERT on a data set containing labeled articles. The data set is available here: You can follow the steps we took to train the model on this notebook: model can be called using an API. We also created two ways of interacting with it: a chrome extension and an integration inside a social media platform.  The chrome extension lets you know if the article on a website you are currently on is considered fake or not. With a simple click, we can extract the article’s text from the website and feed it inside our model. This way, the user can know if the information on the website can be trusted or not.

We also demonstrated how easily our model could be integrated to a social media platform such as Facebook and Twitter by making a social platform where the text inside each post is parsed through our neural network to know if the information can be trusted. A live demo is available here: Note that the server can take a bit of time when loading the results. We parse each text on the website through the neural network and this takes a bit of time.

To test our live version of the extension, please download zip file here: Then extract the content and follow the instruction in this tutorial: When loading the unpacked version, select the folder from the unzipped file. We published the extension on the chrome app store but its currently in the validation process. .

Try It out



javascript, python, pytorch, react, sagemaker

Devpost Software Identifier