Today, Yandex released a new version of its search platform, incorporating two important elements:
· An upgraded version of a deep neural network based search algorithm called “Korolyov,” named after a town northeast of Moscow that has long served as the center of Russia’s space exploration program.
· Incorporating Yandex.Toloka, a mass-scale crowd-sourced platform for search assessors into Yandex MatrixNet.
“Korolyov” builds on “Palekh,” Yandex’s first neural network based search algorithm released in late 2016. The update improves how Yandex handles infrequent and complex queries, known as long-tail queries, in two distinct ways.
First, “Korolyov” is better at understanding user intent than its predecessor because it examines the entirety of web pages rather than just their headlines. Second, “Korolyov” can scale to analyze a thousand times more documents in real time than “Palekh.”
Like all modern AI-based systems, “Korolyov” improves itself with each incremental data point. Yandex’s position as the largest search engine in Russia creates a positive feedback loop for our deep neural network algorithm, which leads to superior search results for our users.
"Korolyov" results feed into MatrixNet, Yandex’s proprietary machine learning ranking algorithm, where a number of other ranking factors are considered before results are returned to a user.
Recently, MatrixNet started incorporating data from Yandex.Toloka, Yandex's crowdsourcing platform, in addition to anonymized user data to train the machine learning algorithms.
These updates help us to enhance the quality of our search services for our users. To learn more about “Korolyov,” you can watch the event broadcast in Russian only.