ALLOCATION OF TIME EXPRESSIONS FOR TEXTS IN NATURAL LANGUAGE Russian patent published in 2016 - IPC G06F17/00 

FIELD: information technology.

SUBSTANCE: invention relates to method of extracting time expressions in natural language texts. Method involves dividing a text into two non-overlapping subsets: non-marked text data for testing and non-marked text data for training; marking untagged text data for testing to obtain “golden” plurality; creating a list of regular expressions and a mechanism for marking text data by means of a list of regular expressions; and marking non-marked text data for training to produce a grammatically marked text and partially marked time expressions; providing training of machine learning algorithm using marked text data, marking non-marked text data for testing, by means of the machine learning algorithm.

EFFECT: technical result consists in enabling and using marking of non-marked text data in machine learning algorithm for marking time expressions in natural language text.

7 cl, 2 dwg