Explaining MatrixNet: Machine Learning Behind Yandex Search

Yandex’s goal is to help users and businesses better navigate the online and offline world. Yandex is most well-known for the way it’s search service helps users find all sorts of answers to their questions.  As one of three companies worldwide with a global web-index, Yandex maintains the ability to crawl the entire web for answers to users’ search queries.  Today Russian Search Marketing looks at the technology behind Yandex search that helps provide the most relevant search results to Yandex users.

Yandex Search for Beginners

The Yandex database consists of known web pages from around the global web index. When a user enters a search query – a phrase into the search box- results are pulled from the Yandex database and shown in the form of links related to the content of the query. The number of search results that answer the query have grown quickly, following the rapid growth of the Internet. Most Yandex searches will retrieve millions of links from its database of known web pages. From there, search results are ranked in order that best answer a given query and to ensure user satisfaction.

yandex search image

Yandex Search Technology

Yandex search uses machine learning to give users the results that match their queries. Machine learning or ML means that the machine can make its own decisions based on input algorithms, empirical data, and experience. Through machine learning, Yandex search is able to create and apply a rule that helps the engine decide what web pages in its databases are good answers to a query.

The technology behind Yandex search includes the help of human assessors who can teach the engine about relevant search results to queries.   The Yandex search team creates learning samples for the algorithm to help the search engine make the best decisions when picking from its database. Learning samples are multiple corresponding queries with search responses that essentially teach the search engine what to look for when finding web pages for queries related to that specific topic.  The more queries the engine gets, the better it becomes in using the learning samples and its experiences to provide strong search results to users.

Yandex search organizes webpages by their ranking factors, which ensure that they are the most relevant results to searches. The search engine also considers static factors, like the number of links leading to that web page. As well as dynamic factors, like if the web page has keywords that match the search query.

Yandex MatrixNet for Search

In 2009, Yandex created its own proprietary machine learning method MatrixNet.  One benefit of MatrixNet for search is preventing the engine from overfitting. Overfitting is when the search engine is looking at too many unrelated factors when deciding the search results, and therefore starts to look for dependencies in webpages that are unrelated to the actual query. MatrixNet helps Yandex search to look at many factors of a webpage to find the most relevant search results and stops it from creating dependencies that do not exist among the results.

Another key feature of MatrixNet besides prevention of overfitting is its ability to create a custom ranking formula for a specific class of search queries. For example, the search engine is able to have a specific ranking algorithm just for music searches. Unlike other machine learning tools, MatrixNet adjustments to specific search query classes without creating changes to the overall search engine.

Read on to learn more about the recent Yandex SEO update, Palekh, which is based on neural networks, a machine learning method, to help Yandex search improve its results across the board but especially for long-tail queries!