METHOD AND SERVER FOR TRAINING MACHINE LEARNING ALGORITHM IN OBJECT RANKING Russian patent published in 2022 - IPC G06F16/245 G06N3/08 

FIELD: computer technology.

SUBSTANCE: computer method for training a machine learning algorithm in object ranking in response to a request, performed by a server, to which objects and the request are available, includes: formation of relevance assessments for a set of training objects based on data related to the set of training objects and to a training request; formation of noisy relevance assessments for the set of training objects by combination of relevance assessments and noise values; formation of an efficiency assessment for MLA algorithm based on noisy relevance assessments; determination of a policy gradient value for correction of relevance assessments, which should be formed by MLA algorithm for objects of a usage stage in response to a request of the usage stage; and use of the policy gradient value in training MLA algorithm in object ranking of the usage stage in response to the request of the usage stage.

EFFECT: optimization of the quality of object ranking by a machine learning algorithm in response to a search request.

26 cl, 5 dwg