Your digital business partner
shikdar is a global it service provider, providing all types of online-based services at a low cost. We are ready to transform your imagination into reality.

Yandex launched the Korolev algorithm. What is it and what to do now


On August 22, an impressive presentation of Yandex took place at the Moscow Planetarium. They spoke a lot of beautiful words, philosophized about the Internet, compared search with space exploration, talked about ranking algorithms and artificial intelligence. And they also pressed the big red button together to launch a new search algorithm – “Korolev”.

What is Korolev and how does it work?

You will not notice any external changes. Search is still an empty line in a yellow frame, and around it is news, exchange rates, weather, program, poster, traffic jams, maps, advertising … The approach to query processing and the relevance of search results is changing. We are sure that having heard the latter, the majority of SEO specialists tensed, waiting for the next Minusinsk, but, apparently, in vain.

Now the search will not search by words, it will compare meanings. The task of the “Korolev” algorithm, which is based on a neural network, is to understand the user’s request, compare it with the information on the pages, and generate relevant results.

Although the neural network was already involved in the previous algorithm – Palekh – the difference is that Palekh used AI to process the titles of 150 pages at the last stage of page ranking, and Korolev will use neural networks at earlier stages, covering the entire text of 200,000 pages. But SEO experts were very skeptical about this high-profile figure, because “95% of clicks are on the first page only.”

For example, by requesting “What to wear”, a user might be interested in fashion trends, style tips, wardrobe apps, or … the weather forecast.

The Korolev converts the semantics of each text into an array of 300 numbers and then compares the resulting arrays. Thus, the neural network assesses the semantic proximity of the texts, and also takes into account all the requests for which users went to each of them.

That is, we ourselves must teach search to search?
Yes, each Yandex user will learn how to search correctly (but, in fact, we have been doing this for a long time), and Korolev will analyze the big data collected for each user in order to further tune in not just to semantic, but to personalized results …

You don’t have to do anything special to start teaching search, just keep (start?) Using it. As 500,000 assessors and tolokers did earlier, helping to develop Korolev (usually in this place they ask a question about the quality of the work of these people, but no one can answer it).

At the presentation, the phrase was heard: “Now it has become much more convenient to search, for example, films whose names you have forgotten, but remember the key points”, which immediately caused a surge of various requests for cinema in Yandex. We have also joined:

And what about “Korolev” they say on the Internet?

Have you noticed that Yandex has become somewhat similar to Sberbank or Russian Post, only on the Internet? No, it does not collect queues, it attracts comments from disgruntled users in huge quantities.

For example, not only testers appeared on the Yandex blog and on habrahabr immediately after the release of Korolev, who began to check the work of the search, coming up with queries “pozakovyrestee”, but also a crowd of offended people whose texts are not searched, sites are ranked poorly, and in technical support, they do not answer for years.

We asked our head of the Internet marketing department, Alexei Basov, to comment on the most frequent complaints about Yandex’s work, which we found under articles about Korolev.

Yandex does not look for the original source or gives it out far from the first page. How is it treated?

I am afraid that the problem with the definition of the primary source is now not cured in any way. The service “Original Texts” in Yandex. The webmaster did not work and does not work correctly. The primary source can be either a more “trustworthy” site (the level of trust in which Yandex has higher than that of your site), and vice versa, a less trustworthy, even spammy site, but which, according to the search engine, is more suitable for this search query. When the content is the same, ranking is based on other factors. But, in any case, we try to add our lyrics to the “Original Texts” and hope that someday it will work.

Problem # 01:

Yandex does not struggle with cheating, which leads to irrelevant results.

Nowadays, there are fewer and fewer cases of cheating on commercial requests. They are “cleaned up” after complaints about resources and inspections by assessors. But, I think that with the introduction of the Korolev algorithm, the results of markups will deteriorate in a natural way – after an increase in the number of bounces and a decrease in CTR.

Problem # 02:
Yandex has problems with geography.

Yes, we have encountered such a problem at work. This is due to a large number of information resources in the search results for commercial queries. A large number of specific companies appear in the search results only after the request has been clarified. But, if you refine your queries with the phrases “at home”, “folk remedies”, etc., the user himself will “retrain” the search to rank information sites higher for queries without these qualifications. If we consider the example from the commentary, even in the search results in Moscow, only half of the sites from the TOP10 for such a request are clinics. The rest is the same informational articles.

On the other hand, this gives rise to even more comments about Yandex’s “conspiracy theory”, so that the owners of commercial companies, not finding themselves in the TOP10 for such queries, launch ads in Yandex. Direct, hoping to take the first advertising lines before the organic search starts. But this moment, of course, cannot be verified in any way.

Problem # 03:

How is it right to search here?

Yes, in this example, the user simply does not know how to cook. The language of queries in Yandex and Google is significantly different. Even in the simplest situation from the comment: in order for the user to see only the results in which he needs a Russian name, it was necessary to put an exclamation mark in front of the desired word in addition to quotes. That is, “! Jeriko” – and everything works correctly. And for those who want to use search wisely, there are official instructions from Yandex.

Problem # 04:

They didn’t say anything about Direct. How will Korolev affect advertising?

The organic search results that Korolev should display and the work of Direct are completely different algorithms. Perhaps, there will be an increase in the number of informative sites in organic search results and it will become more difficult for commercial sites to “breakthrough” to their TOP. Then it is logical that more commercial companies will use Yandex. Direct to compete with information resources on the first search page. But a similar trend has already been observed for a long time and without Korolev.

Problem # 05:

And what should I do now? Strongly Yandex your site?

Do you think no one has done this before you? For quite some time now, there has been a topic with cheating behavioral factors. Those who wanted to “Yandex” their business for many years, those who overdid it – fell under the filter. Yandex has not presented anything fundamentally new at the moment. So you just need to keep working on improving the technical condition of sites, creating quality content, working on usability, and improving conversions. Concurrently conduct contextual advertising and, if necessary, use other channels to attract targeted visitors.

And where to look here?

Korolev, first of all, is aimed at improving the search results for non-commercial queries, therefore, improvements must occur in the TOP for informational low-frequency queries. And, as we remember, the movements here began with the Palekh algorithm at the end of last year.

Taking into account the peculiarities of the gradually learning algorithm, we will be able to notice some changes not earlier than in a month. And they will be the same gradual, so it’s better to stock up on screenshots of the issue today so that you have something to compare within six months or a year.


Leave A Reply

Your email address will not be published.