DEEP-LEARNING BASED MORPHEME ANALYSIS DEVICE AND METHOD OF OPERATING MORPHEME ANALYSIS APPLICATION

According to an embodiment of the present invention, a deep-learning based morpheme analysis device includes: a morpheme candidate column generation module for generating a plurality of morpheme candidate columns from an input word phrase; a morpheme analysis module for recognizing each of a plurali...

Full description

Saved in:
Bibliographic Details
Main Authors KO, EUI YUL, PARK, WEI JIN, OH, SUNG SIK, LEE, SEUL KI
Format Patent
LanguageEnglish
Korean
Published 02.01.2017
Subjects
Online AccessGet full text

Cover

Loading…
More Information
Summary:According to an embodiment of the present invention, a deep-learning based morpheme analysis device includes: a morpheme candidate column generation module for generating a plurality of morpheme candidate columns from an input word phrase; a morpheme analysis module for recognizing each of a plurality of sequential morpheme candidates included in the morpheme candidate columns by using a morpheme dictionary; a junction probability calculation unit for calculating junction probability of the recognized sequential morpheme candidates by using a deep neural network (DNN); and a morpheme column determination module for calculating each generation probability of the morpheme candidate columns on the basis of the calculated junction probability to determine a morpheme candidate column having the highest generation probability after the calculation as a morpheme column with respect to the input word phrase. 본 발명의 실시 예에 따른 딥-러닝(deep-learning) 기반 형태소 분석 장치는, 입력 어절로부터 복수의 형태소 후보열들을 생성하는 형태소 후보열 생성 모듈과, 형태소 사전을 이용하여 상기 복수의 형태소 후보열들 각각에 포함된 복수의 순차적인 형태소 후보들 각각을 인식하는 형태소 분석 모듈과, 인식된 상기 복수의 순차적인 형태소 후보들의 연결 확률을 DNN(deep neural network)을 이용하여 계산하는 연결 확률 계산 모듈과, 계산된 연결 확률에 기초하여, 상기 복수의 형태소 후보열들 각각의 생성 확률을 계산하고, 계산된 생성 확률이 가장 높은 형태소 후보열을 상기 입력 어절에 대한 형태소 열로 결정하는 형태소 열 결정 모듈을 포함한다.
Bibliography:Application Number: KR20150089121