Social media platform, WhatsApp bans about two million accounts per month from its messaging platform.
According to the company, three-quarters of problematic accounts are identified without user report or human intervention, 20 percent of them at the time of registration.
At a recent press briefing in New Delhi, company executives said they have built a machine learning system to detect and weed out users who engage in inappropriate behaviour, such as sending bulk messages and creating multiple accounts with the sole purpose of spreading questionable content on the platform.
This machine learning system has reached a level of sophistication that allows it to ban 20 percent of bad accounts at the time of registration.
According to reports by VentureBeat, spam and phishing is a big issue for the company, and that it uses a mix of AI and specific signals to detect them.
“As with any communications platform, sometimes people attempt to exploit our service. Some may want to distribute click-bait links designed to capture personal information, while others want to promote an idea. Regardless of the intent, automated and bulk messaging violates our terms of service, and one of our priorities is to prevent and stop this kind of abuse,” the company said.
Overall, WhatsApp bans about 2 million accounts on its platform each month, a spokesperson said. To address this issue, a machine learning system uses learnings from the company’s past dealings with problematic accounts and from specific scenarios engineers followed when taking down accounts, said Matt Jones, a software engineer at WhatsApp. This machine learning system has reached a level of sophistication that allows it to ban 20 percent of bad accounts at the time of registration, according to the company.
Jones said WhatsApp has identified various ways users abuse the platform, including through special software that allows users to run multiple instances of different WhatsApp accounts on the same computer. The company also found special devices that support dozens of SIM cards, he said.