Thursday, April 21, 2022
In 2021, the world searched for how to heal and how to come back stronger. To help people find helpful information on questions big and small, we worked to keep spam and malicious content away from Search.
SpamBrain: our most effective solution against spam
We caught 200 times more spam sites in 2021 compared to when we first started nearly two decades ago, thanks, in part, to our AI-based spam-prevention system called SpamBrain.
SpamBrain was launched in 2018 and we’ve been continuously improving its performance. In 2021, SpamBrain identified nearly six times more spam sites than in 2020. This resulted in a major reduction in hacked spam (70%), which was a spam type commonly observed in 2020, and gibberish spam on hosting platforms (75%). Another notable feature of SpamBrain is that it was built to be a robust and evolving platform to address all types of abuse.
With an increasing volume of sophisticated spam being produced every day, SpamBrain’s ability to identify disruptive and malicious behaviors among billions of web pages has allowed us to keep more than 99% of searches spam-free.
Protecting search quality and user safety
We made significant progress in several areas beyond traditional web spam in 2021: most notably in fighting link spam, scams, and online harassment.
Links still help us discover and rank results in meaningful ways, and we made a lot of progress in 2021 to protect this core signal. We launched a link spam update to broadly identify unnatural links and prevent them from affecting search quality.
Scams are a big threat to online user safety. Expanding on our work in 2020, we launched several algorithm updates that resulted in a 40% reduction of scammy results. The improvement in coverage allowed us to protect people against many more scam types beyond the customer support queries that we’ve been fighting for the past few years.
To protect user safety, we extended SpamBrain to address online harassment and, for name queries, reduce the prominence of sites with exploitative removal practices.
Reducing the effects of ranking manipulation
Besides spam, we also work hard to reduce low quality content and ranking manipulations by fighting behaviors that attempt to narrowly avoid violating our quality guidelines, but are still manipulative in nature and degrade the user experience.
For example, one aspect of this initiative was to improve the ranking of product review queries, where content often consisted of rewritten product descriptions, and not genuine, hands-on reviews. We made two substantial updates to how we evaluate product reviews in 2021 that resulted in significant reduction in low quality reviews, while promoting ones with better content and expertise.
We want to make sure there’s nothing getting in the way of people finding the most useful content through Search. If you see manipulative behaviors in search results, you can send us feedback right on the search results page.
Posted by Cody Kwok, Principal Engineer