How a Delhi-based professor is fixing WhatsApp’s fake news problem with machine learning

WhatsApp is under the scanner in India for failing to curb the spread of fake news and rumours. The instant messaging platform has been asked to take more serious efforts to resolve the problem in the wake of several mob lynching incidents across the country.

Even as WhatsApp is taking some initiatives in this regard, a Delhi-based professor and his small team are working on a machine-learning based solution to fix the problem.

Ponnurangam Kumaraguru (PK), who is a professor at the Indraprastha Institute of Information Technology (IIIT), Delhi, says he’s building an application that will make it easier to identify fake news on the platform. The machine learning-based solution is capable of identifying fake videos, texts, images, and even audio recording. PK aims to launch the application in the next couple of months.

PK says that his team has been building similar applications for Facebook and Twitter. He said that his team understands the how-to of the solution, and that only the framework and platform have changed. The researcher added that he was open to work with WhatsApp and provide the solution to fix the problem.

How the fake news identifying app will work?

The application will essentially identify each message sent on the platform and provide a label for authenticity. According to the researcher, the messages may have color codes to distinguish between fake and genuine information. For instance, green label will mean the message is genuine and a red label will mean it’s a fake content. PK describes the application as an extension or wrap around the main WhatsApp application.

Machine learning: Under the hood

PK and his team are currently in process of collecting data sets where they are inviting users to send messages to the number 9354325700. The team will then analyse and process the data to feed the necessary information on the application.

“We will also do annotations of the posts. For example, we already have thousands of messages forwarded to our number… we will build a machine learning model which knows which feature [elements of fake news] is stronger and which isn’t. This app will be on your phone and will keep looking at your messages and analyse it through machine learning to annotate a post about its genuineness. The app will also show you a credibility score to rate a message,” he said.

“One of the features could be the URL app. Other feature could be a negative sentiment. We will collate all these kinds of data to create annotations,” he explained.

Privacy

In order to facilitate the content flagging, the app will have to read users’ messages. It is worth noting that WhatsApp provides end-to-end encryption on its platform which means except for the sender and receiver, no other party can access the messages. PK, however, believes privacy will not be an issue for his application as it works offline and on-device.

“The application works on device. We are not sending any data to our servers or anything like that. The problem would have been if the data is sent to our servers and then the algorithms could have taken a decision on flagging a content,” he explained.