• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.07a
• /
• pp.503-504
• /
• 2022

최근 자연어 처리 분야에서 언어 모델을 활용하여 문장 생성에 관한 연구가 이루어지고 있다. 기존 언어 모델을 활용하여 생성된 시나리오는 텍스트를 학습하여 활용하는 것 외에는 작가의 의도를 반영하는 것에 한계가 존재했고 문맥에 일관성 없는 모습을 보여주었다. 시나리오를 작성하는 것은 작가가 흐름을 주도하며 작업해야 하는 내용이다. 본 논문에서는 GPT-3 기반 언어 모델을 기반으로 다양한 시나리오 문장을 생성하여 작가가 선택하거나 원하는 문장을 직접 입력하는 등 작가의 의도에 부합하는 시나리오를 생성하는 보조 시스템을 제안한다. 본 연구를 통해 시나리오 생성을 포함한 문장 생성 분야의 보조 도구로 활용하여 작가의 의도를 반영하는 결과물을 생성하는 것을 목표로 한다.

• Annual Conference of KIPS
• /
• 2021.11a
• /
• pp.764-767
• /
• 2021

While Generative Adversarial Networks (GANs) have seen huge success in image synthesis tasks, synthesizing high-quality images from text descriptions is a challenging problem in computer vision. This paper proposes a method named Text-to-Face Generation Using Multi-Scale Gradients for Conditional Generative Adversarial Networks (T2F-MSGGANs) that combines GANs and a natural language processing model to create human faces has features found in the input text. The proposed method addresses two problems of GANs: model collapse and training instability by investigating how gradients at multiple scales can be used to generate high-resolution images. We show that T2F-MSGGANs converge stably and generate good-quality images.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.12
• /
• pp.3052-3062
• /
• 1997

An Information Retrieval (IR) is to retrieve relevant information that satisfies user's information needs. However a major role of IR systems is not just the generation of sets of relevant documents, but to help determine which documents are most likely to be relevant to the given requirements. Various attempts have been made in the recent past to use syntactic analysis methods for the generation of complex construction that are essential for content identification in various automatic text analysis systems. Unfortunately, it is known that methods based on syntactic understanding alone are not sufficiently powerful to Produce complete analyses of arbitrary text samples. In this paper, we present a document ranking method based on two-level ranking. The first level is used to retrieve the documents, and the second level to reorder the retrieved documents. The main keywords used in the first level can be defined as nouns and/or compound nouns that possess good document discrimination powers. The sub-keywords used in the second level can be also defined as adjectives, adverbs, and/or verbs that are not main keywords, and function words. An empirical study was conducted from a Korean encyclopedia with 23,113 entries and 161 Korean natural language queries collected by end users. 850% of the natural language queries contained sub-keywords. The two-level document ranking methods provides significant improvement in retrieval effectiveness over traditional ranking methods.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.425-429
• /
• 2018

지식 그래프는 많은 수의 개채와 이들 사이의 관계를 저장하고 있기 때문에 많은 연구에서 중요한 자원으로 활용된다. 최근에는 챗봇과 질의응답과 같은 연구에서 자연어 생성을 위한 연구에 활용되고 있다. 특히 자연어 생성에서 최근 발전 된 심층 신경망이 사용되고 있는데, 이러한 방식은 모델 학습을 위한 많은 양의 데이터가 필요하다. 즉, 심층신경망을 기반으로 지식 그래프에서 문장을 생성하기 위해서는 많은 트리플과 문장 쌍 데이터가 필요하지만 학습을 위해 사용하기엔 데이터가 부족하다는 문제가 있다. 따라서 본 논문에서는 데이터 부족 문제를 해결하기 위해 핵심어 시퀀스를 추출하여 학습하는 방법을 제안하고, 학습된 모델을 통해 트리플을 입력으로 하여 자연어 문장을 생성한다. 부족한 트리플과 문장 쌍 데이터를 대체하기 위해 핵심어 시퀀스를 추출하는 모듈을 사용해 핵심어 시퀀스와 문장 쌍 데이터를 생성하였고, 순환 신경망 기반의 인코더 - 디코더 모델을 사용해 자연어 문장을 생성하였다. 실험 결과, 핵심어 시퀀스와 문장 쌍 데이터를 이용해 학습된 모델을 이용해 트리플에서 자연어 문장 생성이 원활히 가능하며, 부족한 트리플과 문장 쌍 데이터를 대체하는데 효과적임을 밝혔다.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.71-76
• /
• 2022

목적 지향 대화 시스템은 사용자가 원하는 목적을 달성하기 위해 사용하는 시스템으로 일상 대화와 다르게 시스템이 정보를 명확히 전달하는 것이 중요하다. 따라서 최근 연구에서 목적 지향 대화 시스템을 위한 자연어 생성 모델은 정해진 대화 정책에 따라 알맞은 응답을 생성할 수 있도록 의도와 슬롯 정보를 담은 대화 행위(Dialog Act)를 활용한다. 하지만 대화 행위는 생성하는 문장을 탁월하게 제어하는 반면에 대화의 흐름과 상황에 맞게 다양한 문장을 생성하기 어렵다는 문제점을 가지고 있다. 이러한 문제점을 해소하고자 본 논문에서는 목적에 부합하는 내용을 명확하게 자연어로 생성하기 위해 대화 행위를 사용하면서 동시에 일상 대화 생성 모델과 같이 문맥을 고려하여 대화 흐름에 어울리는 자연스러운 문장을 생성할 수 있는 문맥 기반의 제어 가능한 자연어 생성 모델을 제안한다. 실험에서는 KoGPT2 사전 학습 모델과 한국어 대화 데이터셋을 사용하였으며 실험을 통해 대화 행위 기반의 자연어 생성 모델과 본 연구에서 제안한 문맥 기반의 제어 가능한 자연어 생성 모델을 비교하였다. 결과적으로 대화 행위를 단독으로 학습한 모델보다 일정 문맥을 함께 학습한 모델이 유의미한 BLEU 점수 향상을 보인다는 점을 확인하였다.

• Korean Journal of Cognitive Science
• /
• v.1 no.2
• /
• pp.279-291
• /
• 1989

Major difficulties in the style selection,which guarantees the synthesis of good-styled natural expressions under the PIOVT MT system, are an absence of surface-level extra-information in the languageinde-pendent intermediate representation and the language-specific style of expressions due to cultural differences.This paper describes an attempt on the style selection with capabilities of guaranteeing more natural Korean expressions,which includes pragmatic and stylistic decision on target voice genertaion under heavy passive constraints,stylistic changes of sentence-structures,and meaning-supplementation of function words with content words.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.32 no.2
• /
• pp.417-437
• /
• 2022

The quality of the fuzzing seed file is one of the important factors to discover vulnerabilities faster. Although the prior seed generation paradigm, using dynamic taint analysis and symbolic execution techniques, enhanced fuzzing efficiency, the yare not extensively applied owing to their high complexity and need for expertise. This study proposed the DDRFuzz system, which creates seed files based on sequence-to-sequence models. We evaluated DDRFuzz on five open-source applications that used multimedia input files. Following experimental results, DDRFuzz showed the best performance compared with the state-of-the-art studies in terms of fuzzing efficiency.

• Journal of Korea Multimedia Society
• /
• v.11 no.1
• /
• pp.61-75
• /
• 2008

A script language, which represents and controls avatar behaviors in a natural language style, is especially remarkable, because it can provide a fast and easy way to develop an animation scenario script. However, the studies that consider avatar behavior interactions with various virtual objects and intuitive interface techniques to design scenario script have been lack. Therefore, we proposed a context-based avatar-object behavior model and layered script language. The model defines context-based elements to solve ambiguity problems that occur in abstract behavior interface and it provides user interface to control avatar in the object-based approach. Also, the proposed avatar behavior script language consisted of a layered structure that represents domain user interface, motion sequence, and implement environment information at each level. Using the proposed methods, the user can conveniently and quickly design an avatar-object behavior scenario script.

• Journal of Korean Society for Quality Management
• /
• v.52 no.3
• /
• pp.479-493
• /
• 2024

Purpose: The deterioration in the quality of failure history data due to differences in interpretation of failures among workers at power plants and the lack of consistency in the way failures are recorded negatively impacts the efficient operation of power plants. The purpose of this study is to propose a system that classifies power generation facilities failures consistently based on the failure history text data created by the workers. Methods: This study utilizes data collected from three coal unloaders operated by Korea Midland Power Co., LTD, from 2012 to 2023. It classifies failures based on the results of Soft Voting, which incorporates the prediction probabilities derived from applying the predict_proba technique to four machine learning models: Random Forest, Logistic Regression, XGBoost, and SVM, along with scores obtained by constructing word dictionaries for each type of failure using LIME, one of the XAI (Explainable Artificial Intelligence) methods. Through this, failure classification system is proposed to improve the quality of power generation facilities failure history data. Results: The results of this study are as follows. When the power generation facilities failure classification system was applied to the failure history data of Continuous Ship Unloader, XGBoost showed the best performance with a Macro_F1 Score of 93%. When the system proposed in this study was applied, there was an increase of up to 0.17 in the Macro_F1 Score for Logistic Regression compared to when the model was applied alone. All four models used in this study, when the system was applied, showed equal or higher values in Accuracy and Macro_F1 Score than the single model alone. Conclusion: This study propose a failure classification system for power generation facilities to improve the quality of failure history data. This will contribute to cost reduction and stability of power generation facilities, as well as further improvement of power plant operation efficiency and stability.

• Journal of KIISE:Software and Applications
• /
• v.37 no.10
• /
• pp.779-787
• /
• 2010

Since robots have been used widely recently, research about human-robot communication is in process actively. Typically, natural language processing or gesture generation have been applied to human-robot interaction. However, existing methods for communication among robot and human have their limits in performing only static communication, thus the method for more natural and realistic interaction is required. In this paper, an emotional gesture based dialogue management system is proposed for sophisticated human-robot communication. The proposed system performs communication by using the Bayesian networks and pattern matching, and generates emotional gestures of robots in real-time while the user communicates with the robot. Through emotional gestures robot can communicate the user more efficiently also realistically. We used behavior networks as the gesture generation method to deal with dialogue situations which change dynamically. Finally, we designed a usability test to confirm the usefulness of the proposed system by comparing with the existing dialogue system.