Spectrum allowed for inference of implicit queries and returning matching search results.
The system automatically analyzed users’ searches to identify objects such as personal names, films, or cars.
Sections of the search results pages responded and augmented to different user intents are based on the user demand for these results.
Ilya Segalovich took to Quora in December 2010 to provide this explanation as to how the technology worked:
Yandex analyzes a very big search log (5 bln queries) to find ‘objects’ in queries, categorizes them (60 categories) and maps each query into one of possible ‘user intents’ according to a category of the object. Then it counts the percentage of users looking for this object with each of the potential intents. Then it solves an NP-hard optimization problem when it maximizes the probability of user satisfaction for all relatively frequent ambiguous queries.
E.g. for [jaguar] and [beethoven], it will show results about car/animal/drink and movie/composer categories, while for product searches it will typically bring results for ‘buy’, ‘reviews’ and ‘feedback’ intents.
