But it is obviously not easy for a machine and it required not only changes at the functional level but also at the machine level, that is, in hardware. How BERT Works Despite the technical details hidden in the whole process, the bottom line is that by applying the BERT model to both ranking and featured Search snippets, Google helps the user find useful information . According to statistics, in America, there is a 10% improvement , that is, 1 in 10 searches give better results. Obviously this concerns English but it will progressively happen in more and more languages.
Especially for larger searches or dialogue-like searches where the use of intents plays a role, the software will be able to understand what the user wants to know . In short, the user will search in a more natural way. Before launching these improvements, many, many, tests were done. Let's look at an example, in Israel phone number list English of course to understand the difference. The example question is whether a visa is necessary for a trip from Brazil to the USA in 2019: “2019 brazil traveler to usa need a visa” The word that determines the meaning of the question in this example is the word 'to'. It's about a Brazilian who wants to go to America and not the other way around. Without the new model, the machine would not understand this difference.
Thanks to BERT this subtle difference of one word is noticeable and the machine returns the correct results. brazil traveler search Let's look at one more example. This time the question is: " if beauticians do work standing up ". (we will see below) is in English. Here the word that determines the output result is " stand ". google search estheticians And here BERT realizes that it is not a question concerning the work of the beautician but a question concerning the nature of the work of the beautician ( standing ). Below we will see some more examples where the " before and after " is clearly visible.google medicine search Here.