When it comes to finding your way around, Google Maps is one of the world’s best apps to do it with. Barring Apple Maps, it’s really got no competition. Its prominent positioning on Android devices as well as constant feature additions have played a big part of Maps’ longevity. But what keeps the data on the app as accurate as it is are the user contributions that come in all the time — be it reviews or photos or speed trap spotters. But, as with any field of user-submitted content, it can be a scammers’ playground full of shady links, non-existent businesses, or worse. The Google Maps team is now offering some details on what goes on behind the scenes to detect scammy content, usually before it even goes up.
With all the data flowing into Google’s servers, it only makes sense that machine learning models are at the heart of Maps’ defense against scammers. The search giant said it introduced a new update in 2022 to these models which ultimately helped them catch on to scams faster than before.
Citing one example of identifying scams, Google said in a blog post that its automated systems found a surge in Business Profiles with .design or .top domains. Analysts then jumped in to confirm that these were fake websites, leading the team to quickly disable the Google accounts associated with the domains. Google further mentioned that it halted 20 million attempts to generate a fraudulent Business Profile last year, an increase of 8 million from 2021, while implementing new protections for more than 185,000 businesses that were targets of scams or abuse.
In total, Maps removed or blocked more than 115 million reviews deemed to be in violation of its policies. This accounts for a 20% increase in the takedown of fraudulent reviews in 2022 as compared to the year prior.
Another area that the Maps team touches on is the removal of modified or fraudulent imagery. The company found that some scammers pasted fake phone numbers on pictures of storefronts to trick and attract unsuspecting Maps users.
Google deployed another novel machine learning model to counteract these newer attack vectors with the ability to identify numbers placed on top of an existing image “by analyzing specific visual details and the layouts of photos,” the company said. These early detection methods supposedly helped Google block a bulk of the modified imagery before the scammers had a chance to publish them. A total of 200 million photos and 7 million videos were removed by leveraging these revised ML models. However, the number also includes content that was either deemed low quality or blurry in addition to the ones attributed to scammers.
The company also promoted a victory against a network of scammers who were impersonating telemarketers at Google and even trying to sell reviews online. Google also said it is sharing relevant data with the Federal Trade Commission (FTC) and other government agencies outside the US to help eradicate the widespread practice of fake reviews.
Machine-powered algorithms have helped Google wrangle petabytes of incoming and outgoing data through its many various services, but they aren’t always perfect as recent Play Store policy enforcements show. The company is also ramping up its efforts in the field of generative AI, as evidenced by Bard’s recent baby steps, to compete with OpenAI’s splashy ChatGPT. Google was recently accused of using ChatGPT’s data to train its own chatbot, an allegation that the company has denied.
