Google's BERT Algorithm – A Revolution Is Coming

BERT is the latest technology developed by Google one and a half years ago. Its purpose is to improve and enhance the processing of natural language, enabling the search engine to better understand user queries.

Bidirectional Encoder Representations from Transformers, abbreviated as BERT, is the latest technology developed by Google one and a half years ago. Its purpose is to improve and refine natural language processing, enabling the search engine to better understand user queries.

How does the BERT algorithm work?

For every query entered into the search engine, there are hundreds of thousands of pages with useful information. Algorithms consider over 200 different factors that help evaluate the relevance of a given page and ultimately display the most appropriate results. Google’s new technology is based on the use of neural networks. BERT understands the words in a query, their context, order, grammatical case, and treats them as a coherent whole. Special attention is given to correctly interpreting negations, pronouns, conjunctions, and prepositions (so-called stop words), which were previously overlooked. Additionally, BERT assesses which of the searched words are most important for a given query and selects the best results accordingly. This technology is particularly effective for complex phrases. About 15% of the queries that reach Google every day have never been seen before, which is why BERT is an algorithm designed for unpredictable results. Furthermore, according to Google, this technology is so powerful that the necessary resources forced the company to use Cloud TPUs for the first time and deploy BERT in the cloud.

How will BERT affect search?

Google experts claim that the new algorithm impacts 1 in 10 searches, whereas previous updates affected about 3–4% of results. BERT matches queries with content written in natural language found on websites. However, BERT does not replace RankBrain — Google will use the new algorithm only when it determines that a query will be better interpreted by BERT. One definite change will be in the use of keywords. Previously, phrase selection was based on the most important words in their base form. After implementing the new algorithm, grammatical variations and conjunctions will also be taken into account, as they influence the query’s context.

However, a thorough analysis of its effectiveness will require some time. The new solution has so far been fully introduced only for the English language. From December 9, 2019, it has also been gradually implemented for the Polish language.