• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.41-46
• /
• 2022

본 논문에서는 중요도를 반영한 긴 회의록 요약 모델을 제안한다. 제안한 모델은 먼저 회의록을 일정 크기로 구분한 후 구분된 텍스트에 대해 중간 요약문을 생성하고 각 요약문의 중요도를 계산한다. 다음으로 생성된 중간 요약문과 중요도를 함께 사용하여 최종 요약문을 생성한다. 제안 방법은 최종 요약문을 생성할 때 중간 요약문을 다르게 반영하므로 중요한 중간 요약문에서는 핵심 내용을 중점적으로 생성하도록 한다. 실험에서 제안한 요약 모델은 BART기반 요약 모델과, 중요도를 고려하지 않는 요약 모델(SUMM^N)보다 핵심 내용을 포함한 요약문을 생성하였고, 평가 데이터에 대해 ROUGE-1 기준 1.37, 0.29 향상된 성능을 보였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2014.11a
• /
• pp.841-844
• /
• 2014

텍스트 자동 요약은 수많은 텍스트 데이터를 처리함에 있어 중요한 연구 분야이다. 이중 추출요약은 현재 가장 많이 연구가 되고 있는 자동 요약 분야이다. 본 논문은 추출 요약의 선두 연구인 TextRank는 문장 간 유사도를 계산할 때 문장 내 단어 간의 의미적 유사성을 충분히 고려하지 못하였다. 본 연구에서는 의미적 유사성을 고려한 새로운 단어 간 유사도 측정 방법을 제안한다. 추출된 문장 간 유사도는 그래프로 표현되며, TextRank의 랭킹 알고리즘과 동일한 랭킹 알고리즘을 사용하여 실험적으로 평가하였다. 그 결과 문장 간 유사성을 고려할 때 단어의 의미적 요소를 충분히 고려하여 정보의 유실을 최소화하여야 한다는 것을 실험 결과로써 확인할 수 있었다.

• Korean Journal of Cognitive Science
• /
• v.34 no.3
• /
• pp.197-226
• /
• 2023

Generative Text Summarization is one of the Natural Language Processing tasks. It generates a short abbreviated summary while preserving the content of the long text. ROUGE is a widely used lexical-overlap based metric for text summarization models in generative summarization benchmarks. Although it shows very high performance, the studies report that 30% of the generated summary and the text are still inconsistent. This paper proposes a methodology for evaluating the performance of the summary model without using the correct summary. AggreFACT is a human-annotated dataset that classifies the types of errors in neural text summarization models. Among all the test candidates, the two cases, generation summary, and when errors occurred throughout the summary showed the highest correlation results. We observed that the proposed evaluation score showed a high correlation with models finetuned with BART and PEGASUS, which is pretrained with a large-scale Transformer structure.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.694-695
• /
• 2023

최근 ChatGPT 를 각 분야에 활용하는 연구가 활발하게 이루어지고 있다. ChatGPT 는 최신 자연어 처리 모델로, 텍스트를 통해 입출력을 진행한다. 본 논문에서는 이러한 ChatGPT 를 활용하여 영상을 효과적으로 요약할 수 있는 새로운 접근 방식을 제시한다. STT 기술을 사용하여 영상의 자막에 대한 텍스트 파일을 추출하고 이를 ChatGPT 로 요약한다. 최종적으로 기존 텍스트와의 유사도 분석을 통해 유사도가 높은 부분을 선택하여 영상을 편집하고 요약한다.

• Journal of the Korean Society for information Management
• /
• v.26 no.3
• /
• pp.169-188
• /
• 2009

The study proposed the algorithm for automatically summarizing the audio information from a video and then conducted an experiment for the evaluation of the audio extraction that was constructed based on the proposed algorithm. The research results showed that first, the recall and precision rates of the proposed method for audio summarization were higher than those of the mechanical method by which audio extraction was constructed based on the sentence location. Second, the proposed method outperformed the mechanical method in summary making tasks, although in the gist recognition task(multiple choice), there is no statistically difference between the proposed and mechanical methods. In addition, the study conducted the participants' satisfaction survey regarding the use of audio extraction for video browsing and also discussed the practical implications of the proposed method in Internet and digital library environments.

• Journal of KIISE:Software and Applications
• /
• v.27 no.6
• /
• pp.678-689
• /
• 2000

This paper describes an automatic summarization approach that constructs a summary by extracting sentences that are likely to represent the main theme of a document. As a way of selecting summary sentences, the system uses a model that takes into account lexical and statistical information obtained from a document corpus. As such, the system consists of two parts: the training part and the summarization part. The former processes sentences that have been manually tagged for summary sentences and extracts necessary statistical information of various kinds, and the latter uses the information to calculate the likelihood that a given sentence is to be included in the summary. There are at least three unique aspects of this research. First of all, the system uses a text component identification model to categorize sentences into one of the text components. This allows us to eliminate parts of text that are not likely to contain summary sentences. Second, although our statistically-based model stems from an existing one developed for English texts, it applies the framework to individual features separately and computes the final score for each sentence by combining the pieces of evidence using the Dempster-Shafer combination rule. Third, not only were new features introduced but also all the features were tested for their effectiveness in the summarization framework.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.581-585
• /
• 2023

Question Answering(QA)은 질문과 문맥에 대한 정보를 토대로 적절한 답변을 도출하는 작업이다. 이때 입력으로 주어지는 문맥 텍스트는 대부분 길기 때문에 QA 모델은 이 정보를 처리하기 위해 상당한 컴퓨팅 자원이 필요하다. 이 문제를 해결하기 위해 본 논문에서는 요약 모델을 활용한 요약 기반 QA 모델 프레임워크를 제안한다. 이를 통해 문맥 정보를 효과적으로 요약하면서도 QA 모델의 컴퓨팅 비용을 줄이고 성능을 유지하는 것을 목표로 한다.

• Smart Media Journal
• /
• v.12 no.5
• /
• pp.17-27
• /
• 2023

Recently, automatic text summarization, which automatically summarizes only meaningful information for users, is being studied steadily. Especially, research on text summarization using Transformer, an artificial neural network model, has been mainly conducted. Among various studies, the GSG method, which trains a model through sentence-by-sentence masking, has received the most attention. However, the traditional GSG has limitations in selecting a sentence to be masked based on the degree of overlap of tokens, not the meaning of a sentence. Therefore, in this study, in order to improve the quality of text summarization, we propose SbGSG (Semantic-based GSG) methodology that selects sentences to be masked by GSG considering the meaning of sentences. As a result of conducting an experiment using 370,000 news articles and 21,600 summaries and reports, it was confirmed that the proposed methodology, SbGSG, showed superior performance compared to the traditional GSG in terms of ROUGE and BERT Score.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.22-27
• /
• 2021

언어 모델은 많은 데이터와 많은 파라미터로 오래 사전학습을 수행할수록 그 성능이 높아지지만, 그 크기가 큰 만큼 거대 언어 모델은 너무 큰 크기로 인해서 실사용에 많은 하드웨어 리소스를 필요로 한다. 본 논문에서는 거대 언어 모델 중 하나인 T5의 인코더-디코더 구조 대비 절반의 크기를 가지는 PrefixLM 구조에 기반한 한국어 모델을 학습하여 자연어 처리에서 중요한 태스크 중 하나인 텍스트 생성 요약 태스크에서의 성능평가를 하여 BART, T5와 비교하여 각각 0.02, 0.0859의 성능 향상을 보였다.

• Journal of the Korean Society for information Management
• /
• v.18 no.4
• /
• pp.37-55
• /
• 2001

The purpose of this study is to present a text summarization system using a knowledge base containing information about verbs and their arguments that are statistically obtained from a subject domain. The system consists of two modules: the training module and the summarization module. The training module is to extract cue verbs and their basic sentence patterns by counting the frequency of verbs and case markers respectively, and the summarization module is substantiate basic sentence patterns and to generate summaries. Basic sentence patterns are substantiated by applying substantiation rules to the syntactics structure of sentences. A summary is then produced by connecting simple sentences that the are generated through the substantiation module of basic sentence patterns. ‘robbery’in the daily newspapers are selected for a test collection. The system generates natural summaries without losing any essential information by combining both cue verbs and essential arguments. In addition, the use of statistical techniques makes it possible to apply this system to other subject domains through its learning capability.