As part of its ongoing effort to stop the spread of fake news, WhatsApp has started using machine learning to crack down on bulk messages and automated behaviour on its platform. WhatsApp, which has over 1.5 billion monthly active users, claims it removes 2 million such accounts a month, 75 per cent of them even before there is a user report against them.
How does WhatsApp use machine learning?
It gathers features about an action and uses it to train a classifier machine. Often, this data is based on the behaviour of previous offenders. Machines are trained to label behaviour as being abusive, or propagating an agenda.
What are the indicators used?
Some kinds of behaviour raise red flags. For instance, if the device is in one country but is using networks somewhere else. Or, if a new number is used to send out only bulk messages, and has no one-on-one conversations.
Another interesting behaviour is the absence of typing indicator in automated messages. “(The absence of) typing indicator is a giveaway. Those who use emulators forget to add that into their code. If an account has never sent a typing indicator, it can be labelled as an automated account,” Matt Jones of WhatsApp said.
What are the challenges to flagging automated behaviour?
Even regular accounts could get reported if they exhibit what could be mistaken as automated behaviour at times. Like a user getting a new phone number, and sending messages to people who at that time seems like strangers. Therefore, WhatsApp says it has trained its systems to make decisions on the basis of more than just one factor.
What does WhatsApp do once automated behaviour is detected?
Once a user is labelled as being automated, that number is removed from the platform. But while that number cannot be used to log in to WhatsApp again, the user can always use a different number to return. WhatsApp has noticed both software and hardware methods being used to automate bulk messaging on the platform.
What controls do users have when they see automated behaviour?
Users have the option to block and report the user. They also have the option to exit a group. And there are now checks to prevent them from being re-added.