5 Takes on Yandex Neural Network-Based Palekh Algorithm

Being a global agency iProspect UK is proud to service not just UK-based clients, but also brands that operate across Europe. For the majority of the EU markets Google is the main search engine, but not in Russia where Yandex dominates the organic search market. Russian language alphabet and structure differs from English and most other European languages which makes it challenging for Google to process Russian queries and content. Hence from the UX perspective, Yandex provides better local results in Russia compared Google.

As Russia is one of the biggest EU markets and it has a lot of potential for e-commerce brands, the awareness and timely reaction to Yandex updates can be a game changer in the local market.

Yandex, the largest search engine in Russia and one of the key Google’s competitors, introduced a new “Palekh” search algorithm with neural networks at core. With Palekh, Yandex aims at a better long-tail queries processing and finding a right match not just by keywords, but also by a meaning.

Here are the main 5 things you need to know about Yandex Palekh:

1 How many queries are affected?

The new algorithm will affect over 100M search queries per day or approximately 40% of its overall 280M daily search flow. Those are mostly long-tail queries, including the unique ones and the ones asked in a natural language (e.g. “dear Yandex tell me about the movie where the man was growing potatoes on another planet”). 

2How does the new algorithm work?

Yandex taught its search engine to hash the billions of crawled page titles into numbers – or the groups of 300 numbers each. As a result, all the pages from Yandex database get their coordinates in a 300-dimensional space. The search queries are hashed in the same manner.

Then it’s only a question of finding a page with a closest hash number to a query. Named a “semantic vector”, this technology supports the evolution of Yandex towards the semantic search.

3How are neural networks utilised in the algorithm?

Neural networks are essentially a machine learning method. While “learning”, a network processes a huge amount of content examples and develops an ability to recognise an object behind search query in images, sounds and other entities.

Currently Yandex Palekh can apply neural networks only to analyse text in search queries and page titles. Using Yandex’ search and user behaviour stats neural networks learn to “understand” a meaning connection between a page and a query.

4. What does it mean for SEO?

At the moment, Yandex Palekh algorithm looks only at page titles – thus, it might be worth checking them for being content relevant and user-friendly. Yandex is working to expand the technology at the body content in the same way, giving all webmasters a heads-up to prioritise content development. If this update will go forward - it might negatively affect e-commerce sites that do not contain much content apart from product listings.

As the algorithm mostly covers long-tail queries optimisation for natural language queries and voice search might be beneficial. In addition, Yandex uses neural networks to process images, which emphasises the importance of rich content and image optimisation.

On the side note, Yandex is not planning to get rid of 1,5K other ranking factors. Hence, the majority of best practices such as quality content, avoiding black-hat techniques, local relevancy etc. should be still taken into consideration.

5 And finally - what’s the story behind that Palekh name?

Since 2008 Yandex names its algorithms after Russian cities. Palekh is a small town in Russia famous for the folk handicraft of miniature painting inspired by fairy tales and folk songs. The most distinctive image of Palekh art is a firebird with a long beautiful tail. Yandex used it as a metaphor to showcase a variety and uniqueness of the search queries.

comments powered by Disqus