METHOD AND SYSTEM OF SELECTING POTENTIALLY ERRONEOUSLY RANKED DOCUMENTS WITH USE OF MACHINE TRAINING ALGORITHM Russian patent published in 2018 - IPC G06F17/30 G06N5/00 

FIELD: data processing.

SUBSTANCE: invention relates to means for selecting a potentially erroneously ranked document in a set of search results in response to a query. Get a set of search results from the search engine server, and each document from the set of search results has a relevance score and a property vector created by the machine learning algorithm. Calculate for each possible pair of documents the first parameter indicating the level of difference in the relevance assessment of documents from a pair of documents, and a second parameter indicating the level of difference in the property vector of documents from a pair of documents. Result of the verification is calculated based on the first parameter and the second parameter, the result of the check indicating the level of inconsistency between the relevance scores and the property vectors. Select and mark out a couple of documents related to the extreme test result for verification.

EFFECT: technical result consists in increased accuracy of machine learning.

30 cl, 5 dwg