What is the BERT algorithm introduced in Google search engine?

With the spread of smartphones, searching for information on the Internet has become commonplace. Although the word “Google” existed before that time, searching on a computer was still the norm at that time. Another common method was to enter search words using the keyboard.

However, in recent years, the situation has completely changed with the advent of smart speakers. It is becoming common to ask AI assistants to search by voice. Search is now done from anywhere and by speaking to your device.

Under these circumstances, search giant Google has implemented a new natural language processing technology called “BERT” into its search engine, which has become a hot topic.

This time, I will explain “BERT”, which has an algorithm that is completely different from conventional natural language processing.

What is natural language processing technology “BERT”?

BERT

is a natural language processing technology that Google announced in late 2018. It is an abbreviation for Bidirectional Encoder Representations from Transformers and is pronounced “BART”.

It was introduced in the US version of Google in October 2019, and in more than 70 languages, including Japanese, in December of the same year. At the time, it was widely covered by the media around the world as a revolutionary technology for natural language processing.

Natural language processing technology is a processing technology that allows computers to understand human language, but it has existed since before BERT and is called NLP (Natural Language Processing). In other words, BERT is a type of NLP.

As mentioned above, BERT received a lot of attention in the media around the world because its learning and processing methods were revolutionary.

Features of BERT

In conventional NLP, a large amount of data with labels (marks) is prepared and processed by NLP to perform machine learning. The learning accuracy of NLP increases as the amount of data increases, but preparing this large amount of data has been difficult. In BERT, a large amount of unlabeled data is first trained (pre-learning), and then a small amount of labeled data is given to complete the learning (fine tuning). Today, 15% of the words searched on Google every day are completely new words, making learning much more efficient.

Also, the way in which words are processed is the most distinctive feature of BERT.

For example, if you enter “travel bags not for women” in a search, traditional NLP would result in “travel bags for women” being the top search result. Although the search keyword can be broken down into “women’s”, “janai”, and “travel bag”, it is not possible to determine that “janai” depends on “women’s”. As a result, the clear nouns “women’s” and “travel bag” were recognized as search targets, and “janai”, whose meaning was difficult to understand, was ignored.

BERT, on the other hand, understands the context of search keywords. In the above example, it is determined that “Janai” means “not” and applies to “women’s”, and products registered as men’s or gender-free are brought to the top of the search results. *There are differences depending on the search keyword.

By adopting BERT for natural language processing, it is said that an improvement effect of 10% of Google searches in English-speaking countries was seen. BERT is an algorithm that can efficiently learn new words, judge the context of search keywords, and output search results that are more in line with the user’s intentions.

Voice search requires more advanced natural language processing

There are some estimates regarding the penetration rate of smart speakers. In the United States, the number of smartphones used in the United States will increase from about 13% in 2017 to 55% by 2022, and the number of devices in use will reach more than 20 million. As mentioned at the beginning, smart speakers are already voice search-only, and in the future, voice search functionality will become standard in connected cars (vehicles that are always connected to the internet) and smart homes. I’m sure I’ll go.

Humans don’t just speak using organized words like those typed on a keyboard. Voice search requires more advanced natural language processing that can understand complex contexts. BERT is also an algorithm introduced in order to respond to the coming age of voice search.

Does BERT require special SEO measures?

When you hear that “the search algorithm has changed,” do you have to make any changes to your existing SEO measures? You may be wondering. Google’s search engine mainly consists of three programs:

BERT is introduced as part of the program for “keywords entered on search sites”, so there is no need to change your SEO measures. As always, the best SEO strategy is to create high-quality content that is easily collected by Googlebot and that satisfies users’ search intentions.

summary

◆BERT is a type of natural language processing technology (NLP) whose existence was announced by Google in late 2018.

◆BERT is unique in its learning and processing methods compared to other natural language processing (NLP).

◆BERT first trains a large amount of unlabeled data (pre-learning), and then completes the learning by giving a small amount of labeled data later (fine tuning). This has dramatically increased the efficiency of learning new words that are searched every day.

◆BERT understands the context of search keywords. Because voice search doesn’t just speak well-organized words, it can be said to be the perfect natural language processing in the age of voice search.