Yandex has announced the Vega update, a significant update containing more than 1,500 improvements.
One of the more prominent changes is the introduction of human expert feedback into the algorithm, to improve its learning. The update has also allowed Yandex to almost double the size of its search index without hampering the user experience and speed in which results are returned to users.
Vega Update Breakdown
Quality Raters/Assessors
Much like Google, Yandex users quality raters, known as assessors, to test and provide feedback on new algorithm updates.
Whilst Google has said the raters don’t directly impact results and more provide feedback on algorithm changes, Yandex employs their assessors and subsequent feedback to improve accuracy. The assessors (experts) verify Yandex training data, so they can directly impact and influence results.
By training our machine learning algorithms with expert assessments, our search engine learns to rank relevant information higher in results thanks to the work of a highly qualified group of individuals.
Crowd Sourcing Raters
Google’s quality raters are hired, contractors, who are familiar with the quality rater guidelines (and a general sense of how Google should work for the user) to judge and provide feedback on search results.
Yandex, however, relies on Yandex.Toloka, a crowdsourcing platform. Whilst there is less control, Yandex does provide raters guidelines.
Expanding The Search Index
Previously, Yandex would search through its full index in order to answer a query, however, now Yandex’s index utilizes topical clustering.
Instead of searching the full index, Yandex now pulls data from the most topically relevant clusters. Thanks to this clustering technology, Yandex has been able to almost double its capabilities.
In the initial press release, it was thought that the index now contained 200-billion data points, but this number is false, as confirmed by Yandex Head of Communications, Melissa McDonald
