New Google algorithm "SMITH" - How it's better than BERT

Google has been updating the way it fetches results for you every now and then. Google keeps on updating its algorithms to give a better and optimized search result. Lately, Google has been testing out its new algorithm “SMITH”. This is an improvised version of the existing version of this algorithm BERT. Let’s see how SMITH is better than BERT and demystify its working.

To understand this completely, you need to know how google searches work. You basically search in Google’s index of the web. This is done using a software called “spiders”. Initially, spiders start by fetching a few web pages, and then they follow links on those pages and then fetch the pages mentioned in that link. They repeat this process of searching the pages that the links in searched pages point to. This happens till Google has searched quite a big chunk of pages to present you as the search result.

These spiders look for pages that include the keywords you entered in the search. It also looks for the number of times your keywords are contained in the page. Spiders check if the words appear in the title, the URL, directly adjacent. It ranks the pages higher if it includes synonyms of the keywords you used and if it is from a quality website. It usually discourages spammy websites. All this is handled by the SEO agency you are working with.

Page rank is a very important aspect of SEO indexing. It is a formula invented by Larry Page and Sergey Brin. This formula rates a web page’s importance by looking at how many outside links point to it and how important those links are. Digital marketing companies take care of all these factors while publishing a page. Hmm... So now that you know all the basics, let’s dive right into what SMITH and BERT actually are.

Before we tear into it, if you are searching for a digital marketing company or an SEO agency, don’t forget to check out the services Fuerte Developers offer. You can find professional web developers and SEO managers here.

So, the most basic thing which SEOs need is good content. But it’s not easy for SEOs or Google to crack this content easily. SMITH helps google to understand this “content”. Language plays a very important role in helping SMITH to understand the content. It is a powerful tool that helps us to convey the right thing in the right way. But for search engines, it is a rather tedious task. It is an even greater problem for the SEOs because their users speak numerous languages. Hence, to understand all this and in the end answer all the questions that satisfy your query, that too under a matter of millisecond, seems to be a problematic task for Google.

SEO agencies and digital marketing companies make sure that the content is not too vague and to the point. This also helps to index the page on top.

This is the reason why big tech companies like Google and Microsoft research quite intensively on this natural language processing (NLP) and language understanding. This explains the fact that computers’ capacity to understand language is like that of a child. It understands small sentences better. Google understands that this is an issue. Hence, Google tries to solve this problem using Artificial Intelligence (AI).

BERT stands for Bidirectional Encoder Representations from Transformers. Intimidating right? Anyway, it was launched by Google on 25 October 2019. This basically helps Google to understand the natural meaning of sentences. In normal computer systems, a sentence’s meaning is usually understood word by word. This could be way different from the natural language we speak. But BERT, on the other hand, considers the words written before and after the keyword to understand its complete meaning. All thanks to AI.

Even though BERT is powerful in terms of understanding our language, it is not as effective. Sure, it does help Google to understand what a sentence means, but it fails to explain google the meaning of the content in its entirety. Google’s new algorithm SMITH is working this problem out. SMITH means Siamese Multi-Depth Transformer-Based Hierarchical Encoder. Phew!

SMITH helps to understand the relations among sentences in big paragraphs and passages. It then extracts the underlying meaning of it. In light of this, Google came up with passage indexing and ranking. Officially, Google has yet not announced if SMITH is used in the search ranking or not. But as per Google’s data, SMITH can produce better data than BERT.

As per the research papers, SMITH processes 400% more text than BERT. Eventually, it comes with its underlying benefits of doing so. Being able to process 4 times more data gives SMITH the ability to understand long texts properly. It can also recommend search users with similar articles and news content. There would be no wonder if SEO agencies and digital marketing companies start upskilling themselves because SMITH is the future.

SMITH works on a two-tower structure. Herein, one of the two towers divides a big paragraph into many sentences and blocks and the second tower understands the meaning lies in between the sentences. This tells us that one tower works on the sentence level while the other on the document level.

This division is important for certain reasons. One of them being that language processing could be a very intensive task for CPU and GPU.

BERT and SMITH both are pre-trained models and can train themselves without any supervision and can fine-tune themselves. Conclusively, SMITH is not really a replacement for BERT. MatchBERT was a model which was made based on BERT and SMITH is simply an extension to it. Therefore, SMITH kicking into the search engines would bring the two most important advancements to this technology. One, it will be able to understand the content of pages easily and clearly and secondly, it will help Google to check the quality of a page’s content.