BERT (Bidirectional Encoder Representations from Transformers), Bert is one of the Google algorithm update roll-outs that has been rolled out since the Rank Brain algorithm update in the last 5 years.
This is Natural Language Processing (NLP) pre-training with Google, a natural network-base technique, abbreviated as BERT.
Through BART, Google will better understand 10 of 1 search queries, as one person understands. From now on, Google will treat each word of a search query in a different way, making sense, so that we can see the most relevant results on the search engine result page.
Bert rolling out and its effects
Google Bert rolls out this week (October 25-27), and it’s said that the algorithm update will go live in its entirety.
It is currently working for English search queries but it is expected that it will work for queries in other languages at a later time. These updates to Google will also affect feature snippets.
According to Google, the BART algorithm affected search queries by as much as 10%, and that is very high.
Example: Hopefully with this example you can easily understand Bart’s type of work. The following example is from the Google Blog.
In the past, when someone searched for “2019 Japan traveler to USA need a visa”, Google did not consider the word “to” here and its relation to the words after it. As a result, the search results showed Google showing visa information to people whom want to travel from the United States to Japan. Which is the complete opposite information from what is being searched.
But Bert (BERT), here, understands that the word “to” is very important here and its relation to its later words. As a result, Google will understand that there is a Japan here who wants to go to the United States and is seeking visa information. From now on, the results of the search page will be more relevant.
BERT Algorithm Effect
Here is another example where Bert is helping us to understand the small and precise of the language, where the computer is having a lot of trouble understanding it and showing non-relevant results.
Some important issues related to BERT