Google has developed a new SMITH algorithm - more efficient than BERT


Google recently published a research paper on a new algorithm called SMITH that outperforms BERT in understanding long queries and documents.

In particular, the new model is able to understand fragments in documents in the same way that BERT understands words and sentences, which makes it possible to better understand longer documents.

What is SMITH

SMITH is a new model that tries to understand whole documents. BERT is trained to understand words in the context of sentences.

According to a very simplified description, the SMITH model is focused on understanding fragments in the context of the entire document.

While algorithms like BERT are trained on datasets to predict randomly hidden words from context in sentences, SMITH is trained to predict the next block of sentences.

According to the researchers, such training helps the new algorithm better understand long documents than the BERT algorithm.

Is Google using a new algorithm

Google does not usually disclose what specific algorithms it uses. While the researchers claim that SMITH is superior to BERT, until Google officially announces the use of this algorithm, one can only speculate whether it is being used in search engine performance or not.

SMITH test results

"The SMITH model, which has a longer input text length than other standard models, is the best choice for examining and comparing long document representations," the article says.

As a result, the researchers concluded that the SMITH algorithm is better suited for analyzing long documents.

You can read the article here.