Now Machine learning is reshaping online search as at the last time the world of online search was truly reinvented, since 1998 search algorithms have been updated iteratively and gradually. Instead of moving forward with any one new innovation, companies like Google work to refine the features they already have like instead of scrapping content quality as a ranking factor, they simply update their algorithms to be better at determining content quality.
These logical, progressive improvements aren't likely to disappear anytime soon, but a new approach to brainstorming and rolling out those updates could have the power to change how we see, use and take advantage of search rankings and it's all based in machine learning.
The RankBrain Difference
Right now, the only known machine learning-based part of Google's algorithm is RankBrain, an update from 2015, designed to work in conjunction with Hummingbird, it enables Google to understand the meaning behind your queries, rather than dissect the queries themselves.
RankBrain uses machine learning to improve this ability even further, rising quickly to become the third most important signal that contributes to the results of a search query. RankBrain works by carefully monitoring the semantics of user queries and behavior of those users after getting their search results, updating its understanding of user intent along the way. RankBrain also assists Google in tying together multiple related queries. So if a user asks, where is the Washington monument? followed by "how tall is the Washington monument". It might learn the common association between those questions. RankBrain is designed to be self-updating and self-improving, on a constant basis. Since its release, it has been gradually increasing the relevance of search results for users, especially for long tail phrasesl and complex strings of words such as those typically used by users relying on voice search.
Changes in RankBrain
When it was first released, RankBrain was used for about 15 percent of user queries. Today, it's used for all of them, and is in use all the time. Its impact and influence can't be understated, yet you probably haven't noticed the changes it's driven. As a Google search user, you probably haven't seen any drastic improvements to the results you've been getting, and as a search optimizer, you probably haven't noticed your site go through any volatile fluctuations because of it.
The changes have been quite significant, but because they've been rolled out so gradually, and because the machine learning process is constantly moving, search users don't have many opportunities to notice them. This is, in some ways, advantageous for Google. It means the company can keep giving customers better and better results without alienating them with a major change.
There are several advantages to using machine learning updates to keep the algorithm progressing, including:
Accounting for individual user differences. You can use a 'general' pattern to understand the bulk of user search behavior, but there will always be distinct individual differences to account for. Only a machine learning algorithm could be able to dedicate time to these outliers, and account for them with improved functionality.
Speed and scalability. Algorithms with the potential to learn from themselves can be updated faster than those that rely on manual inputs. Similarly, they have tremendous scalability; as long as you have the processing power to devote, a machine learning algorithm could be applied to billions of queries the same way it could be applied to a dozen.
Unlimited application potential. RankBrain is currently the only example we have of a machine learning algorithm in the search world, but it's not the only potential application here. Machine learning processes could be used to improve almost any element of Google's core search algorithm in the future.
Google hasn't yet announced a major machine learning update to come after RankBrain, but RankBrain won't be the last AI-driven update we see from the search engine giant. In few years, a self-updating algorithm designed to evaluate content quality or link quality may arise to join its query-understanding counterpart, or RankBrain may continue to improve our search results all on its own.