On October 25, 2019, Google made us a promise:

We'll understand research better than ever before.

It is through the creation of a new search algorithm , BERT, that Google promises to revolutionize searches once again.

According to the press release posted on their English blog , the search engine explains that they analyze billions of searches every day, and 15% of these queries are ones they've never seen before. Therefore, they've created ways to return results for queries they can't predict.

Official Launch of Bert

Today, December 9th, 2019, Google announced in a tweet that it has released an update to its algorithm for 70 different languages.

Tweet

BERT, our new way for Google Search to better understand languages, is now rolling out to over 70 languages ​​worldwide. It initially launched in October for US English. You can read more about BERT below & a full list of languages ​​is in this thread…. https://t.co/NuKVdg6HYM

— Google SearchLiaison (@searchliaison) December 9, 2019

People like you and me use searches. Every day. But we're not always sure what we're looking for, or how to search for it.

The essence of research is understanding language.

Often we know exactly what we want, but we're not entirely sure of the words that will lead us to the result we need. This is because search engines need questions phrased in their own way to deliver what we want.

That's what BERT aims to change!

search tools have evolved, especially Google's, it has become even more evident how some things still don't work well.

People still feel compelled to search from the algorithm , trying to understand how the search engine will understand what they want to find, instead of asking questions the way one being asks another.

The problem that BERT wants to solve

To solve this problem, the company is making a significant improvement in how they understand queries, representing the biggest leap forward in the last five years and one of the biggest leaps forward in the history of Search.

The latest advances from Google's research team in the science of language understanding, enabled by learning , indicate that future results will be incredible.

Want an example?

Google Translate, with the help of Google Assistant, can now translate speech in real time, directly from your phone. Called Interpreter Mode, this feature is designed to facilitate conversation between people who speak different languages.

Enter BERT

One of the biggest challenges in natural language processing (NLP) is the lack of data to process our language in order to solve the search .

Since NLP is a field with many distinct tasks, most specific datasets contain only a few thousand (or a few hundred thousand) of human-labeled training examples.

Natural language processing at Google BERT

Modern Natural Language Processing models, based on deep learning, demonstrate good results when provided with large amounts of data.

They improve when trained using millions (or billions) of annotated training examples.

Training learning models to improve searches

To help fill this gap with the best data available, researchers have developed a variety of techniques for training general-purpose language representation models using the vast amount of unannotated text on the Web .

This technique is known as pre-training.

The pre-trained model can be adjusted during the NLP learning process with a small sample of data.

This training provides answers to questions and analysis , resulting in substantial improvements in accuracy compared to training on these datasets from scratch.

This technology allows anyone to train their own state-of-the-art question answering system, opening up new business for many companies of all sizes.

Transform, the new language learning model.

Transformer uses a new neural network model to understand the meaning of sentences.

It no longer uses criteria such as word length, but models that process words in relation to all other words in a sentence, instead of one by one in order.

Semantics therefore , is becoming increasingly fundamental to understanding Google's algorithms.

BERT models can therefore consider the full context of a word by observing the words that come before and after it – particularly useful for understanding the intent behind search queries.

Breaking the search code

Current search engines can understand the context of the words in your query, using technologies such as BERT.

You'll be able to search in a way that feels more and more natural to you.

Google itself explained our search, boosted by BERT:

Previously, the word "to" in this English search was given little consideration because it is a very commonly used term. Therefore, the result shown was not accurate for the question: "Does a Brazilian traveling to the USA need a visa in 2019?".

After this implementation, the algorithm understands that the preposition is extremely important, as it is what gives meaning to the question.

In short, critical words will be taken much more into account from this implementation in the algorithm.

Improving search in more languages

By applying BERT to improve Search, Google aims to make search an easier tool to use for people around the world.

A powerful feature of these systems is that they can learn from one language and apply it to others. It is possible to use models that learn from improvements in English (a language in which the vast majority of content ) and apply them to other languages, such as Portuguese, for example.

This helps search engines return better results to users . Increasingly relevant results in the various languages ​​in which Search is offered.

For featured snippets, Google is using a BERT model to improve featured snippets in the two dozen countries where this feature is available.

Analyzing these results, it is already possible to observe significant improvements in languages ​​such as Korean, Hindi, and our Portuguese.

How do users understand search results?

Complex SERPs alter user behavior in relation to searches, and the pinball pattern Kate Moran and Cami Goray .

Complex and dynamic content on search pages (SERPs) receives a lot of attention from search engines. When these features, such as highlighted snippets, are present on the results page, Google researchers found that they received user attention in 74% of cases.

Since search results pages no longer show consistent patterns from query to query, users are often forced to evaluate the page before digging deeper and clicking on anything.

This means that the layout of a SERP can determine which links gain visibility and which ones are clicked.

The inconsistency of SERPs

The inconsistency in SERP layouts means that users are working harder to process information today than they used to in the past.

We can say that search engines are encouraging us to explore the pages with the results of our searches , going beyond the first result. However, people are quite quick in choosing a search result – we found that users spent an average of 5.7 seconds considering the results before making their first selection (with a 95% confidence interval of 4.9 to 6.5 seconds).

You don't always have to be first.

What does this new way of looking mean for digital product teams and content creators?

In 2006, the first result on any search engine results page received 51% of the clicks. In our project , by contrast, we found that the first position on a SERP (defined as the first item listed in the search box) received only 28% of the clicks – almost half, which is a drastic change in user behavior in just about a decade. 59% of the clicks were concentrated in the top three positions, but the lower positions received slightly more clicks than in 2006.

When search engine results pages contain complex and visually appealing elements, users are more likely to be drawn to those elements and distribute their attention across the SERP.

If you manage to reach the top 5 positions in a SERP, you have a good chance (40 to 80%) of gaining valuable insights into your user.

It's still important to appear on the first page of results, as people are unlikely to click through to the second page.

Consider adding some of these non-traditional SERP features to your site if it makes sense for your content. But remember that when results pages are too inconsistent from query to query, users need to re-evaluate the page each time, which increases the cost of interaction.

Source: https://www.nngroup.com/articles/pinball-pattern-search-behavior/

Frequently Asked Questions