DETECTING SECTIONS OF TABLES IN DOCUMENTS BY NEURAL NETWORKS USING GLOBAL DOCUMENT CONTEXT

摘要

FIELD: data processing.;SUBSTANCE: invention relates to detection of text fields in documents. Data processing method includes obtaining a plurality of document symbol sequences comprising at least one table; determining a plurality of vectors representing sequences of symbols comprising at least one alphanumeric character or graphic element of a table; processing a plurality of vectors using a first neural network to obtain a plurality of recalculated vectors; determining a link between a first scaled vector and a second recalculated vector, where the first scaled vector belongs to the alphanumeric sequence, and the second scaled vector is associated with a table section, as well as determining communication between alphanumeric sequence and table section based on communication between first recalculated vector and second recalculated vector.;EFFECT: technical result is wider range of means.;20 cl, 9 dwg