RankBrain: How Google Deals with Confusing or Ambiguous Search Terms

In recent years, Google has evolved the accuracy of their results search, introducing innovations such as the Knowledge Graph. From it, you can have complete information taken from the Internet a number of items, such as information books, celebrities and local. Another example is Google Now, which accepts direct questions and works similarly to a virtual assistant, answering direct questions such as “How tall is Mount Everest?”.


And the Mountain View giant does not stop there. Google has hundreds of algorithms, including PageRank, to organize quickly and accurately search results. One (and one of the most important) is the RankBrain, which deals with searches that are not easily decipherable by a computer (or even by a human).

Unlike RankBrain compared to other algorithms is that most of them are designed by Google engineers based on things they see or experience in the real world. With RankBrain is different: the algorithm uses artificial intelligence to interpret the search terms using “intuition” and learn with time.

But why is this necessary? Greg Corrado, scientist senior Google searches, in an interview with  Bloomberg , said many of the searches done on Google are with ambiguous terms or written in colloquial terms, as if you were talking to a real person.”That’s where the computers usually fail because they fail to understand or never encountered that phrase before,” he explains.

This confusion until it is understandable: it is easy to give results for a search like “primary consumer food chain,” but more complicated for a more elaborate question, with ambiguous terms, such as “What is the consumer title that is at the highest level a chain?”, according to the example of Blooomberg.

Then, the algorithm can, in a way, to generalize the question and assign it to any search term in the past. Thus, the result shown will not be as accurate as possible, but it’s an advance taking into account the type of question. Corrado compares the performance of RankBrain a conversation in a crowded bar. “I can not hear exactly what you’re saying, but I can guess what you mean and have a conversation with you”, he says.

Okay that many people takes over some Google search techniques, how to use quotes in research to make it more preicsa. But also a lot of people who still depend on this algorithm: about 15% of the millions of searches per second using terms that Google never seen before.

But how accurate is this tool anyway? Enough: some search engineers were given the task of analyzing what the research meant in order to compare it with RankBrain. The algorithm hit 80% of the time, while engineers have been successful in only 70% of results. Disabling this algorithm “would be to cause harm to users like forgetting to offer half of the Wikipedia pages,” according to the scientist.