Google has updated its machine learning (ML) model to recognize numbers overlaid on contributed images in Maps, the company's navigation and mapping technology, by analyzing specific visual details and the layouts of photos. This is part of a strategy to keep Maps free from fraud with helpful contributed content.
On Friday, the company shared examples of how it combats fake contributions in Maps, as well as an update on enforcement statistics.
Google estimates that more than 1 billion people rely on Google Maps monthly to navigate the world.
Some of the methods used by Google to combat fake content throughout the year include blocking or removing more than 115 million policy-violating reviews, with the vast majority of them caught before they were ever seen.
Google took down more than 20% more fake reviews in 2022 compared with 2021.
ML has long identified abuse patterns more rapidly, but Google recently launched a significant update to its models that help identify abuse trends much faster than in previous years.
For example, automated systems detected a sudden increase in Business Profiles in websites that ended in .design or .top. This would be difficult to spot manually across millions of profiles.
Removing fraudulent imagery creates challenges. Google has blocked or removed more than 200 million photos and 7 million videos that were blurry, low quality, or violated its content policies.
With help from technology, Google stopped 20 million attempts to create fake Business Profiles -- 8 million more than in 2021.
Protections were put in place for 185,000 businesses after Google detected suspicious activity and abuse attempts and took legal action to fight malicious actors who violated its policies.
Google said it also filed a lawsuit that successfully took down a group of fraudsters who were impersonating Google through telemarketing calls and attempting to sell fake reviews online. This built on our previous legal action taken against internet scammers and malware operations.
Scammers began to overlay inaccurate phone numbers on top of contributed photos in hopes of tricking unsuspecting victims into calling the fraudster instead of the actual business.