At the end of last year 2020, Yandex introduced a new ranking algorithm, the effect of which is based on the neural transformation of queries. The abbreviation YATI is translated from English as “another transformer with improvements.” The new technology, based on the semantic component, evaluates the proximity of the query and the searched page.
This algorithm undergoes self-training according to the transfer principle, that is, first of all, one problem is solved, for which data is collected that allows the transformer to solve it. After that, the same information is used a second time, but for solving other problems.
Initially”Yati” is pre-trained What is the
The new algorithm analyzes text queries and pages displayed on them. According to publishers, it does this more effectively than the Palekh and Korolev algorithms, which, when working together, have a smaller impact on search asia mobile number list results. It is worth understanding that the use of neural networks does not cancel the general search rules by which ranking occurs. However, if search results are displayed only by the YATI algorithm, the quality of selection (by offline metrics) will decrease by a maximum of 5%.
Continuous improvement of search engines and Core Updates
Each search engine notifies its CEO about major updates in different ways. For example, Google , which had three major algorithm updates last year, gave brief warnings before they were introduced and on social media. Yandex does not announce releases itself, but provides more detailed information in articles published about them.
“Yati” was announced in November last year. However, according to “Pihel Tools”, there were no errors in the growth of the average parameters of the search engine emissions that month. The results were observed at the end of September, which could indicate the introduction of this algorithm.
Principles of ranking and neural networks
Neural networks are a type of machine learning that has been used in search engines since the 2000s.
In 2009, Yandex introduced “What is the Snezhinsk” – an algorithm whose operation is based on MatrixNet (the same principle of machine learning).
In 2016, when introducing the creative comparative infographic example Palekh algorithm, Yandex first announced the use of neural networks. Frankly speaking, the search engine began using neural networks earlier – for example, in the Yandex.Translate service.
The Palekh algorithm was a significant . Step in the development of the search engine towards the Yati technology, just like Korolev, introduced in 2017. It was created to compare text queries entered into the line and page titles. Training was carried out on several examples (both positive and ineffective) from previously collected statistical data. Since search engine algorithms cannot read texts, the search for a match was carried out by numerical comparison.
- the autonomous algorithm performs calculations of the page vector, storing it in the index database,
- the user enters a query into the line,
- The query is formatted into a vector, multiplied by the prepared side vectors, and its relevance is calculated.
If the vectors were not calculated in malaysia data advance, it .
In addition to vector comparison actions . Korolev began comparing new query vectors with other query vectors for which the best answer was clear. If they were close to each other, the result would be the same.