RankBrain: Artificial Intelligence in Web Search
After years of research in artificial intelligence, Google has announced it’s new search feature: RankBrain. Idealized to deal with queries the search engine has never seen before (about 15% of all queries), RankBrain has already become the third most important signal contributing to the result of a search query. The signal, using vectors, has the ability to guess words or phrases with similar meanings to ones it isn’t familiar with, being specially useful for queries that contain colloquial terms or that are ambiguous.
Unlike the other hundreds of signals, that were based in engineers’ experiments or previous knowledge, RankBrain is a learning system; it uses AI to interpret queries as people would with gut-feeling and guessability. It becomes more efficient every time it is used, which is already a very large fraction of the millions of queries that Google receives every second.
When put to the test against Google’s own engineers, the new signal proved it’s value: it guessed correctly which pages the search engine would put on top 80% of the time, beating the 70% success rate of the engineers. In experiments with users, it has been determined that turning off this feature would be incredibly damaging.
In order to help RankBrain deal with new concepts, Google carefully monitores the system and updates it periodically by feeding it a load of new data. Besides that, now that the search engine knows the value of machine learning to web search, investments in research in the AI field will certainly continue, creating a promising future not only for RankBrain but for other learning systems to come.