Does Google's RankBrain machine learning improve search results for users?

Googles AI technology RankBrain uses guesswork to return better search results to user queries according to a recent study
Written by Eileen Brown, Contributor

Google switched its search algorithm to machine learning technology called RankBrain for its search results in October 2015.

RankBrain uses an Artificial Intelligence learning system to guess at words and phrases it might not have seen before.

This should make it more effective at responding to queries and improving overall search quality.

RankBrain learning system interprets language and queries in a similar way that 'gut feeling' or 'guessability' works in people.

Colloquial terms or ambiguous search terms can now be guessed at by RankBrain which Google said was the third most important factor in its set of ranking signals.

It also does not 'affect crawling, nor indexing or replace anything in ranking' according to Gary Illlyes, Trends analyst at Google

RankBrain might impact search query results based on better classification of pages and better understanding related groups of words.

Earlier this year Boston based consultancy company Stonetemple carried our a study to measure RankBrains impact on search results and to learn more about how it worked.

The company has stored query results from previous studies it carried out on Google's rich answers. It used this dataset of 1.4 million queries to look at differences and highlight queries that Google did not understand before using RankBrain.

The company found 163 queries that fit the following criteria:

  1. The search results shown indicated that Google didn't understand the query in the Baseline Set
  2. There is, in fact, a reasonable set of results that Google should be able to find for the query

It discovered that 89 out of 163 queries improved with RankBrain - that is 89 more queries that were difficult to understand before RankBrain was used. The results improved 54.6 percent of the time.

An interesting example discovered by the company is this example. The phrase 'why are pdfs so weak' now show that Google now understands that the term 'weak' could be used in the context of security encryption. It then elevates that term on its results page.

Does Google's RankBrain machine learning improve search results for users ZDNet

This is just the start of our search experience. The results delivered by RankBrain will get better as it learns what we are trying to ask of it.

It will learn our nuances, language shortcuts that we use, and the true meaning underneath the collection of words in a search query. With 10.8 billion core searches carried out on Google in the US to February 2016, it has a lot of data to learn from.

Related content:

Editorial standards