딥러닝 기반 문장 중요도를 고려한 중심 문장 추출 방법

Abstract

As data increases exponentially in the era of the 4th industrial revolution, it is becoming more important to efficiently acquire important information from among a number of data. In particular, in the case of text data, since the length of the text data is long and important sentences and non-critical sentences are mixed in the document, it is difficult to grasp the topic. Therefore, a method of efficiently extracting only sentences of high importance in the document is required. Methods of extracting sentences with high importance include a method of extracting important words and important sentences by calculating the frequency of words in a sentence, and a method of calculating a sentence score by calculating the similarity between sentences based on a graph to determine a ranking. Recently, various studies are being conducted to summarize documents using deep learning models. In this paper, we propose a method of extracting the central sentence by defining the sentence with high importance as a topic sentence. The proposed method is to calculate the probability of the central sentence by using the pre-learned language models ELMo, BERT, and the LSTM model that can classify sentences by learning the flow of context. We propose a method of extracting the central sentence by combining the probability of the central sentence and the other features that affect the importance of the sentence.