How Do Google's Algorithmic Updates Really Work?

In light of the recent search algorithm updates such as the Panda 4.0 and Payday 2.0, we got to thinking whether it's ever going to stop. Search algorithms are always evolving, as they have to keep up with more and more complex user requirements, and so updates like these will never stop. But how exactly does Google determine whether one iteration of its search algorithm delivers better quality than the others?

Quality raters

Determining whether one algorithm gives out better results than others is known as search evaluation, and there's a whole science to it. So whenever a search engineer is evaluating a search quality change in order to determine whether it's an improvement over the old one, he uses what are known as quality raters.

Quality raters rate URLs as good, bad, spam, etc. These are pre-computed numbers, stored in a large data bank for millions or URLs on the web. Just like Google evaluates the PageRank for each website, so do other quality raters evaluate domains and webpages. Google uses hundreds of such quality raters to determine the quality of search results.

For example, suppose there's a new algorithm Google wants to roll out, called the Panda 5.0. Before it is released, search results for some example queries will be compared for the Panda 5.0 and the older Panda 4.0. Those pages that have moved up, and those that have moved down are looked at. For pages that have moved up, their average score is calculated based on the hundreds of different quality raters, and the update is justified if the overall quality of the pages that have moved up is greater than those that have moved down.

Live experiments

Often times, before an update is launched completely, it is experimentally launched. Results from the two algorithms, the new one and the old one, are taken and interleaved. Then, if there are more clicks on the results generated by the new algorithm, it could imply better search results.

However, that isn't always the case, because spam content also generates a lot of clicks (because it is designed to generate clicks!). In such cases, a host of other metrics are used to ascertain whether a certain result set is better or worse.

It is not surprising if users continue to click on spam after an algorithmic update. It does not matter because Google uses expert raters that can recognize spam. These algorithms also pay special attention to countries where Google knows there is more spam, such as Turkey. The Payday algorithm is estimated to have impacted a total of ~0.5% English queries, whereas the percentage is as high as 4% for Turkish queries.

All in all, it's a pretty good system that works relatively well in evaluating certain search algorithms. Of course, nothing beats the judgement of the launch committee, whose job it is to evaluate the algorithms using their best judgement and intuition, and declare it fit for release. So that is the final stage of any algorithm update. The whole system works relatively well to keep the algorithms updated and working.

SHARES

Share

Pin it

Tweet

Share

Share

Print

Buffer

If you don't want to get yourself into Serious Technical Trouble while editing your Blog Template then just sit back and relax and let us do the Job for you at a fairly reasonable cost. Submit your order details by Clicking Here »

Unknown 5/25/2014