• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.118-123
• /
• 2021

현대 사회에서 소셜 네트워킹 서비스의 증가와 확산은 많은 정보를 쉽고 빠르게 얻을 수 있도록 하였지만 허위·과장 정보의 확산이 큰 문제로 자리잡고 있다. 최근 해외에서는 이들을 자동으로 분류 및 판별하고자하는 Fact 검증 모델에 관한 연구 및 모델 학습을 위한 데이터의 제작 및 배포가 활발히 이루어지고 있다. 그러나 아직 국내에서는 한국어 Fact 검증을 위한 데이터가 많이 부족한 상황이기 때문에 본 논문에서는 최근 좋은 성능을 보이는 openai 의 GPT-3를 한국어 태스크에 적용시킨 HyperCLOVA 를 이용하여 한국어 Fact 검증 데이터 셋을 자동으로 구축하고 이를 최신 Fact 검증 모델들에 적용하였을 때의 성능을 측정 및 분석 하고자 하였다.

• Journal of Information Science Theory and Practice
• /
• v.11 no.1
• /
• pp.1-13
• /
• 2023

This study determines how social media users (SMUs) verify the information they come across on the Internet. It determines SMUs' perception of online fact-checking services in terms of their ease of use, usefulness, and trust. By conducting a focus group discussion and key informant interviews, themes were derived in determining fact-checking practices while a survey was further conducted to determine such perceived ease of use, usefulness, and trust in fact-checking services. The thematic analysis revealed major information verification practices, such as cross-checking and verifying with other sources, inspecting comments and reactions, and confirming from personal and social networks. The results showed that SMUs considered fact-checking services easy to use. However, a concern was raised about their usefulness stemming from the delayed action in addressing the information issues that need to be verified. As to perceived trust, it was found that SMUs have reservations about fact-checking services. Finally, it is believed that fact-checking services are expected to be credible and need to be promoted to mitigate any form of fake news, particularly on social media platforms.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.227-232
• /
• 2021

사실 검증(Fact verification) 문제는 문서 검색(Document retrieval), 증거 선택(Evidence selection), 증거 검증(Claim verification) 3가지 단계로 구성되어있다. 사실 검증 모델들의 주요 관심사인 증거 검증 단계에서 많은 모델이 제안되는 가운데 증거 선택 단계에 집중하여 강화 학습을 통해 해결한 모델이 제안되었다. 그래프 기반의 모델과 강화 학습 기반의 사실 검증 모델을 소개하고 각 모델을 한국어 사실 검증에 적용해본다. 또한, 두 모델을 같이 사용하여 각 모델의 장점을 가지는 부분을 병렬적으로 결합한 모델의 성능과 증거의 구성 단위에 따른 성능도 비교한다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.524-527
• /
• 2021

최근 국외에서 사실 검증 연구가 활발하게 이루어지고 있지만 한국어의 경우 데이터 집합의 부재로 인하여 사실 검증 연구가 이루어지는데 큰 어려움을 겪고 있다. 이러한 어려움을 해소하고자 자동 생성 모델을 통하여 데이터 집합을 생성하는 시도도 있으나 생성 모델의 특성 상 부정확한 데이터가 생성되어 사실 검증 연구의 퀄리티를 떨어뜨린다는 문제점이 있다. 이러한 문제점을 해소하기 위해 수동으로 구축한 100건의 데이터 집합으로 최근에 이루어진 퓨-샷(Few-Shot) 사실 검증을 확장한 학습이 필요없는 제로-샷(Zero-Shot) 질의 응답에 대한 사실 검증 연구를 제안한다.

• 시스템엔지니어링워크숍
• /
• s.1
• /
• pp.9-14
• /
• 2003

The paper aims at presenting a systematic approach process and a method of requirement validation and system verification. Validation is applied during concept development to ensure conceptual validity, requirements validity, and design validity. Verification work is applied subsequent to the design work on test articles and early production items to produce evidence that the design solutions do, in fact, satisfy th requirements. In this paper, we present a requirements validation model and a system verification model. This models are applied to the development of TTX(Tilting Train Express)system with systems engineering tool, CORE.

• MALSORI
• /
• no.47
• /
• pp.131-140
• /
• 2003

We propose a new method of generating client models for HMM based text-dependent speaker verification system with only a small amount of training data. To make a client model, statistical methods such as segmental K-means algorithm are widely used, but they do not guarantee the quality or reliability of a model when only limited data are avaliable. In this paper, we propose a blind speech segmentation based on level building DTW algorithm as an alternative method to make a client model with limited data. In addition, considering the fact that voiced sounds have much more speaker-specific information than unvoiced sounds and energy of the former is higher than that of the latter, we also propose a new score evaluation method using the observation probability raised to the power of weighting factor estimated from the normalized log energy. Our experiment shows that the proposed methods are superior to conventional HMM based speaker verification system.

• The KIPS Transactions:PartD
• /
• v.16D no.2
• /
• pp.265-272
• /
• 2009

Although the various crime involved numerous digital evidence, the digital evidence is hard to be acknowledged as a evidence to proof the crime fact in court. We propose the method of verification for the legal admissibility of digital evidence using digital forensics ontology. In order to verify the legal admissibility of digital evidence, we will extend the digital ontology by standard digital forensics process from Digital Forensics Technical Manual defined by KNPA and set up the relation properties and the rule of property constraint to process class in the digital forensics ontology. It is possible for proposed ontology to utilize to plan the criminal investigation and to educate the digital forensics.

• KIISE Transactions on Computing Practices
• /
• v.23 no.4
• /
• pp.256-261
• /
• 2017

The notion of System of Systems (SoS), which is composed by many independent systems (i.e., Constituent Systems, CS), has emerged in various domains including social infrastructure. It is widely expected that complex requirements, which cannot be achieved in each CS-level, will be achieved in an SoS-level. While verification of SoS-level goal achievement is one of the most important problems, concrete case studies on SoS modeling and verification are still rare. In this paper, we focus on the fact that each CS performs an action for its own purpose by its own decision-making mechanism. We propose a novel Action-Benefit-Cost (ABC) SoS model which caters to the independent decision-making mechanisms of CSs. Using an abstract SoS example, this proposal provides a case study for the modeling and quantitative verification of the ABC SoS model.

• The Journal of Korean Medicine
• /
• v.19 no.2
• /
• pp.411-419
• /
• 1998

For the purpose of external and internal morphological study on the origin of Folium, writing on herbolagy have been thoroughly reviewed. The object herb medicine(9 kinds) is Lophatheri Folium(Herba), Isatidis Folium, Eoriobotryae Folium, Mori Folium, Perillae Folium, Artemisiae argi Folium, Phyllostachys Folium, Clerodendri Folium, Biotae Folium(Cacumen). After this study, additional experimental verification is required as a matter of fact. As comparison materials, this thesis will be utilized.

• Korean Journal of Logic
• /
• v.17 no.1
• /
• pp.109-136
• /
• 2014

Every (semantic) antirealist accepts one or another form of verification principle. The principle has strong and weak forms, the strong form being highly counterintuitive but the weak one being more plausible. Understandably, antirealists have preferred the weak form of verification principle. Unfortunately, the socalled knowability paradox shows that those two forms are indeed equivalent. To solve this problem, Edgington suggests a yet new form of verification principle. Unfortunately, her new principle has its own difficulty. To overcome this difficulty, Edgington provides a new model of knowledge, according to which every true proposition is somehow associated with a known counterfactual conditional. In this paper, I shall argue that even this new model of knowledge confronts with an insurmountable problem. It is a well-known fact that, in the microscopic levels, some facts manage to occur despite very low physical chances. I will argue that the counterfactuals linked with those facts cannot be known due to the existence of epistemic defeaters. Hence, Edgington's knowledge model does not work in all cases.