• Annual Conference of KIPS
• /
• 2023.11a
• /
• pp.656-659
• /
• 2023

정서적 지원 대화를 위한 챗봇 개발 시, 사용자의 챗봇에 대한 사용성 및 대화 적절성을 높이기 위해서는 사용자 감정에 적합한 지원 콘텐츠를 제공하는 것이 중요하다. 이를 위해, 본 논문은 사용자 입력 텍스트의 감정 강도 예측 모델을 제안하고, 사용자 발화 맞춤형 정서적 지원 대화에 적용하고자 한다. 먼저 입력된 한국어 문장에서 키워드를 추출한 뒤, 이를 각성도 (arousal)과 긍정부 정도(valence) 공간에 투영하여 키워드가 내포하는 각성도-긍정부정도에 가장 근접한 감정을 예측하였다. 뿐만 아니라, 입력된 전체 문장에 대한 감정 강도를 추가로 예측하여, 핵심 감정 강도 - 문맥상 감정강도를 모두 추출하였다. 이러한 통섭적 감정 강도 지수들은 사용자 감정에 따른 최적 지원 전략 선택 및 최적 대화 콘텐츠 생성에 공헌할 것으로 기대한다.

• Journal of Korea Society of Industrial Information Systems
• /
• v.29 no.3
• /
• pp.1-12
• /
• 2024

In this paper, we fine-tuned KoAlpaca, a large-scale Korean language model, and implemented a blog text generation system utilizing it. Blogs on social media platforms are widely used as a marketing tool for businesses. We constructed training data of positive reviews through emotion analysis and refinement of collected sponsored blog texts and applied QLoRA for the lightweight training of KoAlpaca. QLoRA is a fine-tuning approach that significantly reduces the memory usage required for training, with experiments in an environment with a parameter size of 12.8B showing up to a 58.8% decrease in memory usage compared to LoRA. To evaluate the generative performance of the fine-tuned model, texts generated from 100 inputs not included in the training data produced on average more than twice the number of words compared to the pre-trained model, with texts of positive sentiment also appearing more than twice as often. In a survey conducted for qualitative evaluation of generative performance, responses indicated that the fine-tuned model's generated outputs were more relevant to the given topics on average 77.5% of the time. This demonstrates that the positive review generation language model for sponsored content in this paper can enhance the efficiency of time management for content creation and ensure consistent marketing effects. However, to reduce the generation of content that deviates from the category of positive reviews due to elements of the pre-trained model, we plan to proceed with fine-tuning using the augmentation of training data.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.01a
• /
• pp.85-88
• /
• 2023

미국 콜로라도주 박람회 미술전에서 신인 디지털 아티스트 부문에서 1위를 차지한 게임 디자이너인 제이슨 앨런의 작품 스페이스오페라 극장'이 AI Image generator Midjourney를 활용해서 완성된 작품이라는 것이 알려지면서 창작과 예술 분야에 AI 활용이라는 논쟁이 가속화되고 있다. 창작과 예술을 돕는 탁월한 기능을 가진 툴로 바라보거나 창작과 예술 활동에 아이디어를 제공하고 작품을 구체화하는 과정의 조력자로 환영하는 입장과 예술가의 작품을 허가 없이 훔쳐서 만들어 낸 이미지일 뿐이라는 이상도 이하도 아니며 도덕적으로 허락되어서는 안되다는 입장이 크게 충돌하고 있다. 하루가 다르게 빠르게 발전하고 있는 주요 AI Image generator를 살펴보고 창작과 예술 분야에 AI 활용은 어떤 변화를 가져올지, AI 활용의 긍정적인 측면을 예측하고 연구해 보고자 한다.

• The Journal of the Korea Contents Association
• /
• v.20 no.7
• /
• pp.97-106
• /
• 2020

Due to the rapid development and dissemination of 5G communication and IoT technologies, there are increasing demands for big data analysis techniques and service systems. In particular, explosively growing demands on AI technology adoption are also causing high competitions to take advantages of machine/deep-learning models to extract novel values from enormously collected data. In order to adopt AI technology to various research and application domains, it is necessary to prepare high-performance GPU-equipped systems and perform complicated settings to utilze deep learning models. To relieve the efforts and lower the barrier to utilize AI techniques, AIaaS(AI as a service) platform is attracting a great deal of attention as a promising on-line service, where the complexity of preparation and operation can be hidden behind the cloud side and service developers only need to utilize the high-level AI services easily. In this paper, we propose an AIaaS system which can support the creation of AI services based on Docker and OpenFaaS from the registration of models to the on-line operation. We also describe a case study to show how AI services can be easily generated by the proposed system.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.722-725
• /
• 2023

대규모 언어 모델(LLM)은 대규모의 데이터를 학습하여 얻은 지식을 기반으로 텍스트와 다양한 콘텐츠를 인식하고 요약, 번역, 예측, 생성할 수 있는 딥러닝 알고리즘이다. 초기 공개된 LLM은 영어 기반 모델로 비영어권에서는 높은 성능을 기대할 수 없었으며, 이에 한국, 중국 등 자체적 LLM 연구개발이 활성화되고 있다. 본 논문에서는 언어가 LLM의 성능에 영향을 미치는가에 대하여 한국어 기반 LLM과 영어 기반 LLM으로 KoBEST의 4가지 Task에 대하여 성능비교를 하였다. 그 결과 한국어에 대한 사전 지식을 추가하는 것이 LLM의 성능에 영향을 미치는 것을 확인할 수 있었다.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.431-436
• /
• 2022

이 연구는 MBTI의 심리 기능지표 조합인 ST, SF, NT, NF의 유형별 특징을 반영한 마케팅 문체 프레임워크를 정의하고 모델 학습을 통해 성격유형별 맞춤화 된 광고 메시지로 생성하는 것을 목적으로 한다. 활용되는 광고 메시지 자동 생성 기술은 BART 모델에 성격유형을 Prefix로 포함한 광고문을 학습시켜 성격유형에 따라 맞춤형 광고 메시지를 생성하는 방식이다. 학습된 모델은 Prefix 조작만으로 MBTI 성격유형별 문체 특징을 갖춘 광고 메시지로 변환되는 것을 실험을 통해 확인할 수 있었다. 본 연구는 성격유형의 특징을 문체 프레임워크로써 정의하고 이에 기반한 모델 학습을 통해 성격유형별 특징을 반영한 광고 메시지를 재현해 낼 수 있다는 점에서 의의가 있다. 또한 성격유형과 연관 feature를 함께 학습하여 유형별 문체 특징과 소구점을 포함한 광고 메시지를 생성했다는 기술적 가치가 있다. 이 연구 결과를 기반으로 차후 타겟 고객층의 성격유형과 광고 도메인을 고려한 효과적인 광고 콘텐츠를 생성해 내는 모델을 개발하여 타겟 마케팅 분야는 물론이고 지역별 또는 언어별 문체 간 차이를 구조화하거나 재현해야 하는 문제에서 기반이 되는 연구로 활용될 수 있을 것으로 기대된다.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.10
• /
• pp.77-87
• /
• 2024

With the recent advancements in artificial intelligence, various chatbots have emerged, efficiently performing everyday tasks such as hotel bookings, news updates, and legal consultations. Particularly, generative chatbots like ChatGPT are expanding their applicability by generating original content in fields such as education, research, and the arts. However, the training of these AI chatbots requires large volumes of conversational text data, such as customer service records, which has led to privacy infringement cases domestically and internationally due to the use of unrefined data. This study proposes a methodology to quantitatively assess the re-identification risk of personal information contained in conversational text data used for training AI chatbots. To validate the proposed methodology, we conducted a case study using synthetic conversational data and carried out a survey with 220 external experts, confirming the significance of the proposed approach.

• KIPS Transactions on Computer and Communication Systems
• /
• v.12 no.1
• /
• pp.25-40
• /
• 2023

Recommendation systems are frequently used by users to generate intelligent automatic decisions. In the study of movie recommendation system, the existing approach uses largely collaboration and content-based filtering techniques. Collaborative filtering considers user similarity, while content-based filtering focuses on the activity of a single user. Also, mixed filtering approaches that combine collaborative filtering and content-based filtering are being used to compensate for each other's limitations. Recently, several AI-based similarity techniques have been used to find similarities between users to provide better recommendation services. This paper aims to provide the prospective expansion by deriving possible solutions through the analysis of various existing movie recommendation systems and their challenges.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.9
• /
• pp.453-459
• /
• 2024

Recent advances in generative AI technologies using natural language processing have critically impacted text, image, and video production. Despite these innovations, we still need to improve the consistency and reusability of AI-generated outputs. These issues are critical in cartoon creation, where the inability to consistently replicate characters and specific objects can degrade the work's quality. We propose an integrated adaption of language analysis-based requirement engineering and cartoon engineering to solve this. The proposed method applies the linguistic frameworks of Chomsky and Fillmore to analyze natural language and utilizes UML sequence models for generating consistent 3D representations of object interactions. It systematically interprets the creator's intentions from textual inputs, ensuring that each character or object, once conceptualized, is accurately replicated across various panels and episodes to preserve visual and contextual integrity. This technique enhances the accuracy and consistency of character portrayals in animated contexts, aligning closely with the initial specifications. Consequently, this method holds potential applicability in other domains requiring the translation of complex textual descriptions into visual representations.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.3
• /
• pp.99-105
• /
• 2021

In addition to the OTT video production service represented by Nexflix and YouTube, a personalized recommendation system for content with artificial intelligence has become common. YouTube's personalized recommendation service system consists of two neural networks, one neural network consisting of a recommendation candidate generation model and the other consisting of a ranking network. Netflix's video recommendation system consists of two data classification systems, divided into content-based filtering and collaborative filtering. As the online platform-led content production is activated by the Corona Pandemic, the field of virtual influencers using artificial intelligence is emerging. Virtual influencers are produced with GAN (Generative Adversarial Networks) artificial intelligence, and are unsupervised learning algorithms in which two opposing systems compete with each other. This study also researched the possibility of developing AI platform based on individual recommendation and virtual influencer (metabus) as a core content of OTT in the future.