The second possibility would indicate that click data wasn't vetted well. "Brave Search is a user-first machine learning system."Īs for Bing, Eich said Microsoft "got that hlybbprqag result in their index either by Googlers clicking on the fake result link or else by Bing scraping unclicked results blindly." The first is "akin to search click fraud," he said, where people try to manipulate search results by clicking results they want to see rank highly. Machine learning systems do not merely copy, they aggregate and optimize," Eich said. "Rather than copying, we prefer to say learning, as we believe it's more precise. And he says it applies a lot of machine learning technology that goes well beyond just copying what comes out of Google's search engine. Bing, in some cases, then started recommending the same pages that were Google's search results.Įich, like Microsoft, argues that there's nothing wrong with using users' clickstream data in this way.
In 2011, Google manually wired its search results to show particular pages for nonsense searches like "hiybbprqag." Google employees searched for those terms into computers using Microsoft's Internet Explorer browser running the Bing toolbar extension. That "clickstream" data is anonymized so it can't be tracked to individual users, he said.Ĭhecking clickstream data is similar to an approach Microsoft used in Bing - one that led to Google charging that Bing copied Google search results.
Instead, the startup crowdsources the work with help from Brave users who, if they opt into data sharing, can supply Brave with data about what they search for and what search results they click on, Eich said. Brave doesn't build its search index alone, though. It requires immense resources to scour the entire web for information, build an index of that information, then evaluate the best results for a given search query.
0 kommentar(er)
