FIELD: data processing.
SUBSTANCE: invention relates to means for detecting text fields in documents using neural networks. Multiple sequences of symbols of a document containing a plurality of text fields are obtained. Determining plurality of vectors belonging to one of multiple sequences of symbols. Multiple vectors are processed using a first neural network to derive, based on values of a plurality of vectors, a plurality of recalculated vectors. Links between the first counted vector of the plurality of recalculated vectors and the first text field from the plurality of text fields are determined, wherein the first recalculated vector belongs to the first sequence of symbols from the plurality of sequences of symbols. Also, links between the first sequence of symbols and the first text field are determined based on the connection between the first recalculated vector and the first text field.
EFFECT: technical result consists in improvement of efficiency of data processing.
20 cl, 9 dwg
