for that query. In addition to better understanding the concepts of a web page, Rank rain also helps to better understand the association between several queries, such as: "Where is the Eiffel tower?" Followed by: "How tall is he?" How does Rank rain learn? Examples of Rank rain in action Essentially, RankBrain can use sets of "training" data created by humans to help establish a baseline, and then can apply machine learning to determine the best search results based on various factors over time.
Google has confirmed in the Bloomberg article and in this Search Engine Land article that they periodically update the system giving it new data to better reason with new concepts. At SMX West 2016, some presenters shared examples of RankBrain in action. A study showed how RankBrain better interpreted relationships between words. This could include the use of company employee list stop words in a search query (“the”, “without”, etc.) — words that were historically ignored by Google but are sometimes critically important to understanding the intent behind a request. For example, take the TV series “The Office”.
This is an example of research that would be taken out of context without the all-important “the”. Here's another sample query from an interview with Googler Gary Illyes: "Can you score 100% on Super Mario without taking the step?" Ignoring "without" would potentially return search results about getting a 100 percent score on Super Mario with a tour...so the opposite of the results someone was trying to get. There are other theories about how RankBrain might use data to figure out what the best results are for a search query. It's possible that searchers' engagement with search results could be