• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.167-171
• /
• 2020

문서 요약은 길이가 긴 원본 문서에서 의미를 유지한 채 짧은 문서나 문장을 얻어내는 작업을 의미한다. 딥러닝을 이용한 자연어처리 기술들이 연구됨에 따라 end-to-end 방식의 자연어 생성 모델인 sequence-to-sequence 모델을 문서 요약 생성에 적용하는 방법들이 연구되었다. 본 논문에서는 여러 자연어처리 분야에서 높은 성능을 보이고 있는 BERT 모델을 이용한 자연어 생성 모델에 복사 메커니즘과 강화 학습을 추가한 문서 요약 모델을 제안한다. 복사 메커니즘은 입력 문장의 단어들을 출력 문장에 복사하는 기술로 학습데이터에서 학습되기 힘든 고유 명사 등의 단어들에 대한 성능을 높이는 방법이다. 강화 학습은 정답 단어의 확률을 높이기 위해 학습하는 지도 학습 방법과는 달리 연속적인 단어 생성으로 얻어진 전체 문장의 보상 점수를 높이는 방향으로 학습하여 생성되는 단어 자체보다는 최종 생성된 문장이 더 중요한 자연어 생성 문제에 효과적일 수 있다. 실험결과 기존의 BERT 생성 모델 보다 복사 메커니즘과 강화 학습을 적용한 모델의 Rouge score가 더 높음을 확인 하였다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.228-233
• /
• 2023

인간의 수동 평가 시 시간과 비용의 소모, 주석자 간의 의견 불일치, 평가 결과의 품질 등 불가피한 한계가 발생한다. 본 논문에서는 맥락을 고려하고 긴 문장 입출력이 가능한 ChatGPT를 활용한 한국어 요약문 평가가 인간 평가를 대체하거나 보조하는 것이 가능한가에 대해 살펴보았다. 이를 위해 ChatGPT가 생성한 요약문에 정량적 평가와 정성적 평가를 진행하였으며 정량적 지표로 BERTScore, 정성적 지표로는 일관성, 관련성, 문법성, 유창성을 사용하였다. 평가 결과 ChatGPT4의 경우 인간 수동 평가를 보조할 수 있는 가능성이 있음을 확인하였다. ChatGPT가 영어 기반으로 학습된 모델임을 고려하여 오류 발견 성능을 검증하고자 한국어 오류 요약문으로 추가 평가를 진행하였다. 그 결과 ChatGPT3.5와 ChatGPT4의 오류 요약 평가 성능은 불안정하여 인간을 보조하기에는 아직 어려움이 있음을 확인하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.514-517
• /
• 2023

본 연구는 인 컨택스트 러닝 (In-Context Learning)을 오디오-언어 작업에 적용하기 위한 멀티모달 (Multi-Modal) 딥러닝 모델을 다룬다. 해당 모델을 통해 학습 단계에서 오디오와 텍스트의 소통 가능한 형태의 표현 (Representation)을 학습하고 여러가지 오디오-텍스트 작업을 수행할 수 있는 멀티모달 딥러닝 모델을 개발하는 것이 본 연구의 목적이다. 모델은 오디오 인코더와 언어 인코더가 연결된 구조를 가지고 있으며, 언어 모델은 6.7B, 30B 의 파라미터 수를 가진 자동회귀 (Autoregressive) 대형 언어 모델 (Large Language Model)을 사용한다 오디오 인코더는 자기지도학습 (Self-Supervised Learning)을 기반으로 사전학습 된 오디오 특징 추출 모델이다. 언어모델이 상대적으로 대용량이기 언어모델의 파라미터를 고정하고 오디오 인코더의 파라미터만 업데이트하는 프로즌 (Frozen) 방법으로 학습한다. 학습을 위한 과제는 음성인식 (Automatic Speech Recognition)과 요약 (Abstractive Summarization) 이다. 학습을 마친 후 질의응답 (Question Answering) 작업으로 테스트를 진행했다. 그 결과, 정답 문장을 생성하기 위해서는 추가적인 학습이 필요한 것으로 보였으나, 음성인식으로 사전학습 한 모델의 경우 정답과 유사한 키워드를 사용하는 문법적으로 올바른 문장을 생성함을 확인했다.

• Journal of information and communication convergence engineering
• /
• v.21 no.4
• /
• pp.287-293
• /
• 2023

With the rapid development of domestic and international over-the-top markets, a large amount of video content is being created. As the volume of video content increases, consumers tend to increasingly check data concerning the videos before watching them. To address this demand, video summaries in the form of plot descriptions, thumbnails, posters, and other formats are provided to consumers. This study proposes an approach that automatically generates posters to effectively convey video content while reducing the cost of video summarization. In the automatic generation of posters, face recognition and clustering are used to gather and classify character data, and keyframes from the video are extracted to learn the overall atmosphere of the video. This study used the facial data of the characters and keyframes as training data and employed technologies such as DreamBooth, a text-to-image generation model, to automatically generate video posters. This process significantly reduces the time and cost of video-poster production.

• Journal of the military operations research society of Korea
• /
• v.3 no.2
• /
• pp.49-77
• /
• 1977

Weapon systems are curcial instruments for the security of a nation and critical elements for the victory in a war. Since modern weapon systems tend to be capital-intensive with high precision and quality, they become more and more complex and diversified; their acquisition costs become huge; and their technological obsolescence becomes accelerated. Therefore, the systematic management of weapon acquisition process would be one of the most important defense tasks at the national level. To analyze such problems and find solutions, this paper has studied various aspects related to the efficient management of weapon system acquisition. After brief summarization of the general characteristics of weapon systems, their effectiveness, and developmental trend, the paper discusses the defense management policies and techniques for the weapon systems. Specifically, four alternative acquisition methods such as indigenous R & D, foreign purchase, co-production and joint-production are discussed and analyzed by systems approach. The systems analysis procedure to evaluate and select weapon acquisition method is as follows; 1) to analyze the merits and demerits of the alternative methods, 2) to screen unrealistic alternatives through the consideration of significant factors such as political, economic, military, technological, and social constraints, 3) to evaluate and select an optimal one among the remaining acquisition methods after the cost-effectivenss analysis. For the base of cost-effectivess analysis, cost analysis model as well as effectiveness analysis model of each acquisition method are developed.

• Journal of the Korea Convergence Society
• /
• v.9 no.11
• /
• pp.23-28
• /
• 2018

Recent approaches to Relation Extraction methods mostly tend to be limited to mention level relation extractions. These types of methods, while featuring high performances, can only extract relations limited to a single sentence or so. The inability to extract these kinds of data is a terrible amount of information loss. To tackle this problem this paper presents an Augmented External Memory Neural Network model to enable Global Relation Extraction. the proposed model's Global relation extraction is done by first gathering and analyzing the mention level relation extraction by the Augmented External Memory. Additionally the proposed model shows high level of performances in korean due to the fact it can take the often omitted subjects and objectives into consideration.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.8
• /
• pp.31-39
• /
• 2022

Stock prices are influenced by a number of external factors, such as laws and trends, as well as number-based internal factors such as trading volume and closing prices. Since many factors affect stock prices, it is very difficult to accurately predict stock prices using only fragmentary stock data. In particular, since the value of a company is greatly affected by the perception of people who actually trade stocks, emotional information about a specific company is considered an important factor. In this paper, we propose a deep learning-based stock price prediction model using sentiment analysis with news data considering temporal characteristics. Stock and news data, two heterogeneous data with different characteristics, are integrated according to time scale and used as input to the model, and the effect of time scale and sentiment index on stock price prediction is finally compared and analyzed. Also, we verify that the accuracy of the proposed model is improved through comparative experiments with existing models.

• Journal of the Korean Society for Library and Information Science
• /
• v.58 no.1
• /
• pp.225-254
• /
• 2024

The study aims to understand undergraduates' and graduate students' perceptions and satisfaction with online learning platforms and to verify the relationship between usability factors and satisfaction with online courses. The literature review facilitated the summarization of major factors to be considered in the online learning platform development process and established the research model. The follow-up survey verified the perceptions of university constituents regarding the fulfillment of the university online learning platforms' user interface principles, platforms' usability, satisfaction with platforms, and satisfaction with online courses. Causal relationships between variables were tested and modeled by analyzing survey results. We also confirmed that the same model can be applied to different types of learners and various types of online learning methods. This study is significant in verifying that the fulfillment of the platforms' user interface design principles can affect satisfaction with online courses using the platforms based on learners' evaluation results. We expect that the research model proposed in this study can contribute to the improvement and development of online learning environments in the future.

• Journal of Intelligence and Information Systems
• /
• v.28 no.4
• /
• pp.309-328
• /
• 2022

This study empirically analyzed a Korean pre-trained language models (PLMs) designed for natural language generation. The performance of two PLMs - BART and GPT - at the task of abstractive text summarization was compared. To investigate how performance depends on the characteristics of the inference data, ten different document types, containing six types of informational content and creation content, were considered. It was found that BART (which can both generate and understand natural language) performed better than GPT (which can only generate). Upon more detailed examination of the effect of inference data characteristics, the performance of GPT was found to be proportional to the length of the input text. However, even for the longest documents (with optimal GPT performance), BART still out-performed GPT, suggesting that the greatest influence on downstream performance is not the size of the training data or PLMs parameters but the structural suitability of the PLMs for the applied downstream task. The performance of different PLMs was also compared through analyzing parts of speech (POS) shares. BART's performance was inversely related to the proportion of prefixes, adjectives, adverbs and verbs but positively related to that of nouns. This result emphasizes the importance of taking the inference data's characteristics into account when fine-tuning a PLMs for its intended downstream task.

• The Journal of Society for e-Business Studies
• /
• v.14 no.4
• /
• pp.59-73
• /
• 2009

Keyword extraction is an important and essential technique for text mining applications such as information retrieval, text categorization, summarization and topic detection. A set of keywords extracted from a large-scale electronic document data are used for significant features for text mining algorithms and they contribute to improve the performance of document browsing, topic detection, and automated text classification. This paper presents a keyword extraction technique that can be used to detect topics for each news domain from a large document collection of internet news portal sites. Basically, we have used six variants of traditional TF-IDF weighting model. On top of the TF-IDF model, we propose a word filtering technique called 'cross-domain comparison filtering'. To prove effectiveness of our method, we have analyzed usefulness of keywords extracted from Korean news articles and have presented changes of the keywords over time of each news domain.