국어교육과 디지털 리터러시 ; 문서 자동 요약의 현황과 과제

강인수 ( In Su Kang ) 국어교육학회 2010 國語敎育學硏究 Vol.39 No.-

인간이 다루어야 할 정보가 기하급수적으로 증가하는 문제를 다루기 위해 전산언어학 및 자연어처리 커뮤니티에서는 문서 요약의 자동화 기법이 연구되고 있다. 1950년대부터 시작된 자동 문서 요약 연구는 여러 유형의 문서를 다루면서 단일/다중 문서 요약, 질의 관련 다중 문서 요약 등의 다양한 태스크에 적용하기 위한 추출 및 추상 방식 요약 기법을 시도해 왔다. 이 논문은 추출 방식을 중심으로 텍스트 자동 요약 기술의 현황을 제시하고 요약 평가 방법과 대규모 자동 요약 대회에 대한 개괄 및 향후 과제에 대해 기술한다. Information that human should read grows exponentially. To deal with this problem, computational linguistics and natural language processing communities have attempted to automate summarizing text. Since its start in 1950`s, automated text summarization has handled single-/multi-document summarization using extracting and abstracting techniques, and nowadays specialized its tasks to query-focused multi-document summarization. This paper gives the current state of automatic text summarization techniques focusing on robust, practical extraction-based methods, and describes evaluation methodologies and large-scale summarization evaluation conferences. Finally, future issues are discussed.

Multi-layered attentional peephole convolutional LSTM for abstractive text summarization

Rahman, Md. Motiur,Siddiqui, Fazlul Hasan Electronics and Telecommunications Research Instit 2021 ETRI Journal Vol.43 No.2

Abstractive text summarization is a process of making a summary of a given text by paraphrasing the facts of the text while keeping the meaning intact. The manmade summary generation process is laborious and time-consuming. We present here a summary generation model that is based on multilayered attentional peephole convolutional long short-term memory (MAPCoL; LSTM) in order to extract abstractive summaries of large text in an automated manner. We added the concept of attention in a peephole convolutional LSTM to improve the overall quality of a summary by giving weights to important parts of the source text during training. We evaluated the performance with regard to semantic coherence of our MAPCoL model over a popular dataset named CNN/Daily Mail, and found that MAPCoL outperformed other traditional LSTM-based models. We found improvements in the performance of MAPCoL in different internal settings when compared to state-of-the-art models of abstractive text summarization.

심층학습을 이용한 문서요약의 연구방법 분석

지인영,김희동 국제언어인문학회 2018 인문언어 Vol.20 No.1

In this study, we discuss the basic technology of Text Summarization based on the deep neural network for natural language processing(NLP). The text summarization task is divided into an extractive summary and an abstractive summary. The extractive summary is a method of extracting a summary of the words used in the input document in the output text, and the abstractive summary is a problem of understanding the input statement and generating a sentence of the same content. The abstractive sentence generation system is based on the encoder-decoder model with attention mechanism, and a selector that can select input sentence is added. The Copy network and Pointer network are the special mechanisms for selector. Such selector systems can make text summarization to be the hybrid form of abstractive and extractive summary. In the future, we expect that accuracy of text summarization will be improved by adding reinforcement learning method.

한국어 국회 회의록 생성 요약 말뭉치 구축 및 모델 개발

함영균,강예지,박서윤,정용빈,서현빈,이이슬,서혜진,서샛별,김한샘 한국정보과학회 2024 정보과학회논문지 Vol.51 No.3

The mainstream of summary research has been targeting documents, but recently, interest in meeting summary research has significantly increased. As part of the National Institute of Korean Language’s big data construction project, a study on the summary of the National Assembly minutes, which have not yet been studied in Korea, was conducted and a summarization dataset for the National Assembly minutes was constructed. Qualitative intrinsic human evaluation was conducted to verify the quality of the constructed dataset. In addition, by conducting quantitative and qualitative evaluations of datasets built through the generative summarization model, the evaluation of the National Assembly Minutes Summarization dataset and the research direction of future generative and minutes summaries were sought.

한국어 텍스트 문서의 자동 요약에 관한 연구

이유빈(You-Bin Lee),온병원(Byung-Won On) 한국정보기술학회 2022 Proceedings of KIIT Conference Vol.2022 No.6

4차 산업혁명의 발전과 함께 자연어 처리 분야에서는 방대한 양의 데이터를 의미 있고 가치 있는 정보로 변화하기 위한 자동 문서 요약 연구가 활발히 진행되고 있다. 최근 문서 요약은 주어진 문서의 의미를 유지하면서 중요하고 핵심적인 내용을 포함한 요약을 생성함으로서 자연어 처리에서 주목받는 분야 중 하나이다. 과거에는 추출 요약 방법이 주로 연구되었지만, 최근에는 심층 신경망을 학습하기 위한 시퀀스 투 시퀀스 모델이 개발되면서 생성 요약에 대한 연구가 진행되고 있다. 본 논문에서는 한국어 텍스트 문서를 요약하는 문서요약 기법들을 살펴보고, 최신 방법론에 대해 자세히 논의한다. With the development of the 4th industrial revolution, automatic text summarization methods have been actively studied in the field of natural language processing, to transform a vast amount of data to meaningful and valuable information. These days, text summarization is one of the fields receiving attention in natural language processing as it generates a summary including important and core contents while maintaining the meaning of a given document. In the past, extractive summarization methods were mainly studied, but now, with the development of various sequence-to-sequence models for learning deep neural networks, abstractive summarization methods have been in progress. In this paper, major text summarization techniques that summarize Korean text documents are surveyed and the state-of-the-art methods are discussed in detail.

가짜 뉴스 탐지를 위한 텍스트 요약 탐색

변지에,이승언,카란딥 싱,차미영 한국정보과학회 2022 정보과학회논문지 Vol.49 No.11

Fake news detection models need to gather and ingest massive information from heterogeneous sources rapidly for solid verification. This paper demonstrates the feasibility of applying text summarization, to uncover useful information or evidence for fake news detection. Two popular deep learning-based summarization techniques, extractive and abstractive, were used to generate condensed textual information from lengthy news content. Experiments on popular rumor debunking datasets show that two lines of summarized text can extract critical information, while improving the classification performance and substantially reducing inference time. Text summarization can also bring explainability by providing evidence from three levels: words, sentences, and documents. 가짜 뉴스 탐지는 방대한 양의 이종 데이터를 빠르게 수집하여 확실한 검증을 요한다. 이 논문은 텍스트 요약 기술이 뉴스의 중요한 정보나 단서를 찾아내어 가짜뉴스 탐지 문제에 기여할 수 있음을 제시한다. 잘 알려진 벤치마크 데이터셋에서 검증해본 결과 주요한 두 가지 텍스트 요약 기법이 - 추출요약 및 추상요약 - 모두 가짜뉴스 탐지 문제 해결에 각기 도움을 줌을 확인할 수 있었다. 더 나아가 방대한 뉴스 데이터로부터 텍스트 요약은 문장, 단어, 문서의 단계에서 주요 정보를 압축함으로써 근거 자료로 활용되는데 있어 모델의 설명 가능성도 보여준다.

KI-HABS: Key Information Guided Hierarchical Abstractive Summarization

( Mengli Zhang ),( Gang Zhou ),( Wanting Yu ),( Wenfen Liu ) 한국인터넷정보학회 2021 KSII Transactions on Internet and Information Syst Vol.15 No.12

With the unprecedented growth of textual information on the Internet, an efficient automatic summarization system has become an urgent need. Recently, the neural network models based on the encoder-decoder with an attention mechanism have demonstrated powerful capabilities in the sentence summarization task. However, for paragraphs or longer document summarization, these models fail to mine the core information in the input text, which leads to information loss and repetitions. In this paper, we propose an abstractive document summarization method by applying guidance signals of key sentences to the encoder based on the hierarchical encoder-decoder architecture, denoted as KI-HABS. Specifically, we first train an extractor to extract key sentences in the input document by the hierarchical bidirectional GRU. Then, we encode the key sentences to the key information representation in the sentence level. Finally, we adopt key information representation guided selective encoding strategies to filter source information, which establishes a connection between the key sentences and the document. We use the CNN/Daily Mail and Gigaword datasets to evaluate our model. The experimental results demonstrate that our method generates more informative and concise summaries, achieving better performance than the competitive models.

실시간 뇌파반응을 이용한 주제관련 영상물 쇼트 자동추출기법 개발연구

김용호(Yong Ho Kim),김현희(Hyun Hee Kim) 한국멀티미디어학회 2016 멀티미디어학회논문지 Vol.19 No.8

To obtain good summarization algorithms, we need first understand how people summarize videos. Semantic gap refers to the gap between semantics implied in video summarization algorithms and what people actually infer from watching videos. We hypothesized that ERP responses to real time videos will show either N400 effects to topic-irrelevant shots in the 300∼500ms time-range after stimulus on-set or P600 effects to topic-relevant shots in the 500∼700ms time range. We recruited 32 participants in the EEG experiment, asking them to focus on the topic of short videos and to memorize relevant shots to the topic of the video. After analysing real time videos based on the participants’ rating information, we obtained the following t-test result, showing N400 effects on PF1, F7, F3, C3, Cz, T7, and FT7 positions on the left and central hemisphere, and P600 effects on PF1, C3, Cz, and FCz on the left and central hemisphere and C4, FC4, P8, and TP8 on the right. A further 3-way MANOVA test with repeated measures of topic-relevance, hemisphere, and electrode positions showed significant interaction effects, implying that the left hemisphere at central, frontal, and pre-frontal positions were sensitive in detecting topic-relevant shots while watching real time videos.

Construction of Text Summarization Corpus in Economics Domain and Baseline Models

Sawittree Jumpathong,Akkharawoot Takhom,Prachya Boonkwan,Vipas Sutantayawalee,Peerachet Porkaew,Sitthaa Phaholphinyo,Charun Phrombut,Khemarath Choke-mangmi,Saran Yamasathien,Nattachai Tretasayuth,Kasi 한국정보통신학회JICCE 2024 Journal of information and communication convergen Vol.22 No.1

Automated text summarization (ATS) systems rely on language resources as datasets. However, creating these datasets is a complex and labor-intensive task requiring linguists to extensively annotate the data. Consequently, certain public datasets for ATS, particularly in languages such as Thai, are not as readily available as those for the more popular languages. The primary objective of the ATS approach is to condense large volumes of text into shorter summaries, thereby reducing the time required to extract information from extensive textual data. Owing to the challenges involved in preparing language resources, publicly accessible datasets for Thai ATS are relatively scarce compared to those for widely used languages. The goal is to produce concise summaries and accelerate the information extraction process using vast amounts of textual input. This study introduced ThEconSum, an ATS architecture specifically designed for Thai language, using economy-related data. An evaluation of this research revealed the significant remaining tasks and limitations of the Thai language.

Adversarially Regularized Autoencoder를 이용한 텍스트 요약 문장 생성

공혜수,김우주 한국지능정보시스템학회 2018 한국지능정보시스템학회 학술대회논문집 Vol.2018 No.6