DEVICE AND METHOD TO GENERATE ABSTRACTIVE SUMMARIES FROM LARGE MULTI-PARAGRAPH TEXTS RECORDING MEDIUM FOR PERFORMING THE METHOD

• Main Authors: MOIRANGTHEM DENNIS SINGH, MINHO LEE
• Format: Patent
• Language: English; Korean
• Published: 25.07.2018
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

A device for creating an abstract of multi-paragraph texts includes: an input unit for automatically dividing a document into paragraphs and transmitting the same; an encoding unit for transforming the paragraphs received from the input unit into an internal representation vector, generating, as an internal representation, a vector transformed through a recurrent neural network (RNN) including a multiple timescales grated recurrent unit (MTGRU) having multiple constants, and transmitting the internal representation; a decoding unit for decoding the internal representation received from the encoding unit through the RNN including the MTGRU and generating sentences by using linguistic modeling; and an output unit for collecting an abstract output of each paragraph to output the final abstract. Accordingly, an abstract expression is generated so that an abstract more similar to a man-made abstract can be created. 복수 문단 텍스트의 추상적 요약문 생성 장치는, 문서를 문단들로 자동 구분하여 전달하는 입력부; 상기 입력부로부터 전달된 문단을 내부표현 벡터로 변환하고, 다중 시상수를 갖는 GRU(Multiple Timescales Gated Recurrent Unit, 이하 MTGRU)를 포함하는 회귀신경망을 통해 변환된 벡터를 내부표현(representation)으로 생성하여 전달하는 부호화 처리부; MTGRU를 포함하는 회귀신경망을 통해 상기 부호화 처리부로부터 전달받은 내부표현을 복호화하고, 언어 모델링을 이용하여 문장들을 생성하는 복호화 처리부; 및 각 문단의 요약 출력을 수집하여 최종 추상적 요약을 출력하는 출력부를 포함한다. 이에 따라, 추상적 표현을 생성함으로써 보다 사람이 작성한 요약에 가까운 요약문을 생성할 수 있다.