METHOD, APPARATUS AND DEVICE FOR NATURAL LANGUAGE PROCESSING Russian patent published in 2020 - IPC G06F16/33 

FIELD: data processing.

SUBSTANCE: invention relates to the processing of natural language. Method of natural language processing, applicable to a chat-bot in a human-machine communication system and including stages, on which determining the result of slots marking, issued by the used model of the bidirectional recurrent neural network with long short-term memory based on the conditional random fields BiLSTM-CRF, after marking slots in speech data input by user; determining reinforcement information based on the slots marking result and user reinforcement response to the slots marking result; and training with reinforcement on BiLSTM-CRF model in accordance with reinforcement information.

EFFECT: technical result consists in dynamic self-learning model and higher accuracy of slot layout due to the fact that model BiLSTM-CRF provides for external use the result of marking slots, chat-bot receives corresponding reinforcement information corresponding to the result of slots marking, and performs training with reinforcement on BiLSTM-CRF model in accordance with reinforcement information.

12 cl, 5 dwg