YATI Machine Learning: What Is It?

Yandex Webmaster introduced a new tool to check brocken links

The goal of YATI is to improve the ability of robots to find a relevant answer to a user’s question.

The algorithm is designed to better assess the semantic proximity of the request and the document (web page), almost like a person.

And what happened before that? Until 2016, the Yandex search engine took into account only 5-10% of all text, the rest of the material was simply ignored. Of course, there were certain attempts: an advanced language model, latent semantic indexing (LSA), etc.

But in fact, the SE read superficially, understood only the title, the entry of keywords, some synonyms of the main query and common terms.

This situation was the impetus for the introduction of a smarter robot. According to the assurances of Yandex specialists, YATI has completed its task by 96% – the ranking has improved significantly compared to what it was over the past 10 years.

Yandex did not immediately evolve. Before that, there were two attempts to enter neural network search:

  • Palekh (2016) – he knew how to understand the request and the title, using alphabetic trigrams and bigrams of words for this, but all this is only for 150 top pages (in other words, he worked only with the top ten results plus 5 sites);
  • Korolev (2017) – used streams and was able to take into account not only the title and key phrase, but also the content itself (albeit only some important excerpts) for 200 thousand documents.

And finally, a new architecture or new text analysis technology that has modernized search better than Palekh and Korolev combined.

With YATI, it became possible to use even more streams – anchor list, query index for URLs by clicks – and understand articles up to 10 sentences in full, without breaking into separate fragments.

Optimizing For YATI

Do not forget that YATI only sorts TOP results by 50%. These factors continue to play a major role in ranking:

  • the total number of pages of the resource (website)
  • structure
  • quality and uniqueness of content
  • link profile
  • number of visitors (traffic)
  • PF

Many experts just recommend working in 2021 to improve behavioral factors. Of course, without frank cheating, since last year, before the launch of YATI, the Russian search engine mowed down a large number of sites for the unnaturalness of PF.