Design and Implementation of an Emotional AI Game Character System
Abstract : This study designed and implemented an emotion-driven AI game character system that operates on-device, built upon a small language model integrating persona grounding and quantitative emotion scoring. To address the limitations of large language models, such as latency and inconsistency, a lightweight transformer-based encoder-decoder architecture was combined with a vector database to embed dialogue context, world-building information, and emotional states, thereby maintaining consistency in emotion inference and character identity. The emotion scoring mechanism was grounded in Plutchik's theory of emotion to quantify 10 emotion vectors to interpret user responses numerically. This emotion data was then used to dynamically adjust the dialogue tone and relational changes between the AI character and the user in real time. Furthermore, by separating system and user prompts within the emotional dialogue design, the system continuously accumulated and analyzed dialogue history and emotional flow, enabling a nonlinear narrative structure that triggers affinity-threshold events. By expanding AI game characterssimple ‘conversational agents’ to ‘storytelling agents that build and grow emotional relationships’, this study proposes a new interaction model that enhance immersion and emotional bonding within gameplay.
Keyword : AI, Game Character, SLM, Persona Grounding, Emotion Scoring
http://dx.doi.org/10.29056/jncist.2025.11.01
Skin Type and Aging Classification Model Using YOLO11: Potential and Limitations
Abstract : When artificial intelligence is used to assess skin type and aging, which are important for improving overall quality of life, it is expected that rapid and accurate results will be obtained. In particular, YOLO, which is simple and easy to use as an recognition model, is expected to be useful in clinical and industrial applications. In this research, we classified skin image data using facial image data of Korean women, labeled data of skin types, expert diagnoses, and quantitative measurement results, and evaluated the accuracy of the classification data using YOLO11. The accuracy of skin type classification was 57%, as as an indicator of aging stage, 56% for pigmentation and 66% for periocular wrinkles. In the case of aging stage indicators, the loss rate converged to the level around 2. Through this study, we confirmed the possibility that the YOLO model can be used not only for abnormality detection such as skin lesions but also for ive evaluation of aging stage. In addition, we were able to confirm the possibility of using it in industries such as suggesting customized cosmetics with rapid and convenient skin evaluation during extensive data collection and quantitative index development.
Keyword : YOLO11, Classification, C3k2, Pigmentation, Periocular Wrinkle
http://dx.doi.org/10.29056/jncist.2025.11.02
A Comparative Study on the Performance of Pan-Tompkins and SVM·RF Methods for ECG R-Peak Detection
Abstract : This study compares and analyzes the R-peak detection performance of the Pan–Tompkins algorithm, a Support Vector Machine (SVM; linear kernel, C=1), and a Random Forest (n=100) using 48 records (360 Hz)the MIT-BIH Arrhythmia Database under identical preprocessing and feature extraction conditions. A total of eight statistical features (mean, standard deviation, maximum, minimum, median, Q1, Q3, variance) were extracteda ±50-sample (100-point) window centered on each R-peak to construct feature vectors. To address the class imbalance, a 1:1 undersampling method was applied, and the dataset was divided into training and test sets at a ratio of 80:20. The same feature set and data-splitting strategy were applied across all three models to ensure fair comparison. Experimental results showed that the Random Forest achieved the highest performance in accuracy (0.9214) and precision (0.8961), while the SVM demonstrated relatively superior performance in sensitivity (0.9707). The Pan–Tompkins algorithm exhibited lower precision compared to the machine-learning models but maintained stable sensitivity (0.9384). These results indicate that traditional and machine-learning-based algorithms have different strengths, and the optimal choice may vary depending on the intended application. In particular, algorithms with higher sensitivity may be more suitable for real-time processing environments, while algorithms with higher precision are appropriate in scenarios where suppression of false positives is critical. This study provides quantitative evidence for selecting appropriate algorithms for real-time ECG-based diagnostic systems and clinical applications under uniform experimental conditions.
Keyword : electrocardiogram (ECG), R-peak detection, Pan-Tompkins algorithm, Support Vector Machine (SVM), Random Forest (RF)
http://dx.doi.org/10.29056/jncist.2025.11.03
Potential and Technical Limitations of ECG Zero-shot Learning for Rare Cardiac Disease Diagnosis
Abstract : Automatic diagnosis of rare cardiac diseases faces challenges due to insufficient training data. Zero-shot learning (ZSL) has been proposed as a potential solution by enabling prediction of classes absenttraining data. This study proposes a framework that aligns ECG signals with disease descriptions in a common space using a transformer-based ECG encoder and PubMedBERT. We conducted 5-fold cross-validation using the PTB-XL dataset (21,837 ECGs, 71 diagnoses), systematically divided into 45 seen and 26 unseen classes. The proposed method achieved 18.3±2.1% accuracy for unseen classes and 73.5±1.8% for seen classes, demonstrating a 37.6% relative improvement in harmonic mean (H-score) compared to baseline methods (p<0.01). However, the observed sensitivity of 12.5% for life-threatening conditions including ventricular tachycardia and ventricular fibrillation identifies sensitivity improvement for critical diseases as a priority research ive. As the first systematic study exploring zero-shot learning in ECG analysis, this research establishes an initial baseline (18.3%), with few-shot learning and multimodal learning emerging as promising future directions.
Keyword : Electrocardiogram, Zero-shot Learning, Transformer, Medical AI Baseline, Initial Research Achievement
http://dx.doi.org/10.29056/jncist.2025.11.04
Abstract : This study applied the Extended Unified Theory of Acceptance and Use of Technology (UTAUT2) to analyze key factors influencing user experience enhancement and continuous use intention for text-based AI platforms such as ChatGPT and Gemini. The research focused on three key questions: (1) the relative importance of UTAUT2's core factors on user experience, (2) the mediating effect of user experience enhancement on continuous use intention, and (3) the impact of price value and facilitating conditions on continuous use intention. Data were collected through a survey of 190 participants and in-depth interviews and usability tests with 7 participants. The analysis revealed that performance expectancy (β=0.259) and hedonic motivation (β=0.283) had the most significant impact on user experience enhancement, while performance expectancy (β=0.247), social influence (β=0.192), and habit (β=0.259) significantly influenced continuous use intention. However, the mediating effect of user experience enhancement on continuous use intention was not observed, and price value and facilitating conditions were not found to have significant effects on continuous use intention. This study highlights performance expectancy and habit as key factors driving continuous use intention and confirms the critical role of hedonic motivation in enhancing user experience. Future research should include diverse age groups and longitudinal data to expand these findings.
Keyword : Generative AI, User Experience Enhancement, Cotinuous Use Intention, UTAUT2 Model
http://dx.doi.org/10.29056/jncist.2025.11.05
Analysis of the Characteristics of Media Art: A Case Study Based on Materiality of the Medium
Abstract : Media art emphasizes the materiality of the medium itself, asserting that creative materiality can more profoundly transform society and human perception than content alone. However, its technological and algorithmic foundations often make it less accessible and marginal within contemporary art discourse. This study examines recent tendencies in media art amid the rapid integration of artificial intelligence (AI). The analysis reveals three major directions: first, the material properties at the intersection of art and engineering are being redefined through AI; second, immersive technologies enhance sensory engagement and audience experience regardless of conceptual intent; third, AI’s emergence foregrounds issues of creativity and autonomy, revitalizing debates on authorship and artistic agency. Consequently, media art-once on the periphery is gaining recognition within institutional art contexts and fostering new theoretical discourses. The study aims to deepen understanding of media art’s evolving nature and to support practitioners in anticipating technological shifts for more innovative creation.
Keyword : MediaArt, Media, Materiality, immersion, AI
http://dx.doi.org/10.29056/jncist.2025.11.06
A Study on the Character-Based User Experience Design of Pikmin Bloom
Abstract : This study examines the impact of character-based exercise game user experience design on users' exercise habits through the popular mobile game ‘Pikmin Bloom’. In particular, it identifies key elements contributing to the formation of bonds between users and Pikmin characters and verifies their influence on actual exercise habits by conducting in-depth, one-on-one face-to-face interviews with 10 subjects. The results show that users' affinity for Pikmin characters significantly increased due to game play experiences, and direct interactions with characters and realistic settings within the game played a crucial role in this change. Additionally, it was found that the bond users feel towards characters extends beyond immersion in the game to positive changes in actual exercise habits. However, users' immersion in the Pikmin Bloom game and Pikmin characters did not reach the level where characters felt like they exist in reality, and the study concludes by suggesting design development directions to overcome this limitation.
Keyword : Exergame, Character, Augmented Reality, User Experience, Pikmin Bloom
http://dx.doi.org/10.29056/jncist.2025.11.07
Abstract : This study thoroughly analyzed the impact of dark patterns commonly seen on accommodation booking platforms on user experience. It focused on two platforms, Agoda and Yeogieottae, and identified four types of dark patterns: information hiding and distortion, urgency notifications and psychological pressure, wasting time and effort, and action distortion and choice restriction. Participants of various age groups experienced the actual booking process on both platforms and were then interviewed in-depth. The results indicated that messages emphasizing urgency and hidden costs added at the payment stage caused discomfort and were major factors that weakened users' trust in the platforms. Repeated experiences with dark patterns made users suspicious of the platforms' commercial intentions, leading to psychological stress and distrust despite recognizing these tactics as marketing strategies. However, due to a lack of alternative platforms, users continued to use the platforms despite their discomfort. This study highlights the unethical design elements found in accommodation booking platforms, emphasizing the need to improve user experience and propose ethical design directions. These findings are expected to serve as foundational data for future related research and the development of design guidelines.
Keyword : hotel booking platforms, dark patterns, user experience, in-depth interviews, ethical design
http://dx.doi.org/10.29056/jncist.2025.11.08
Abstract : As global warming intensifies and numerous animal species face extinction, the importance of protecting endangered animals through everyday environmental protection activities is being emphasized. This study aims to develop a separate collection record application that enables users to participate in environmental protection easily and enjoyably. To this end, seven endangered animal species, including polar bears, Adélie penguins, and sea turtles, were selected, and a system was established where users take photos of their separate collection after performing it, and an AI verifies it and accumulates points. The accumulated points are used to improve the habitatsprovide food for the selected animals. The application encourages user participation through gamification elements and visual feedback, and is designed to be easily accessible to a diverse user base by applying persona theory and inclusive design. In conclusion, this application provides both fun and meaning of environmental protection activities through virtual connections between users and endangered animals, and will establish itself as an innovative way to promote sustainable environmental behavior.
Keyword : Global Warming, Endangered Animals, Recycling Behavior, Gamification, Sustainable Design
http://dx.doi.org/10.29056/jncist.2025.11.09
A Survey on Generative AI-driven 2D Game Graphic Asset Production
Abstract : In game development, two-dimensional (2D) graphic assets are core elements that define player experience and game identity, yet the production of character sprites, tilesets, backgrounds, effects, and UI icons still depends heavily on manual work, so that it results in substantial production costs and repetitive labor. Traditional procedural content generation (PCG) has proven effective for automating structural content such as levels, rules, and stories, but it has shown clear limitations in domains like 2D graphic assets, where strict control over art style and direction is required. Recent studies in generative adversarial networks, image-to-image translation models, and diffusion-based generative AI have rapidly expanded the possibilities for automatic generation, enhancement, and transformation of 2D game graphic assets. Focusing on research into generative AI–based 2D game graphic asset production, this paper surveys prior work along five dimensions the theoretical background of PCG, generative models, and graphical asset frameworks technical trends centered on characters, animations, and effects mixed-initiative asset production pipelines that integrate generative AI evaluation metrics and benchmarks that reflect the characteristics of 2D games and game artists' acceptance of these technologies and emerging structures of human–AI collaboration. The analysis indicates that generative AI has the potential to simultaneously enhance efficiency and expressiveness in 2D game graphic asset production, while also revealing remaining challenges in domain generalization, frame consistency, tiling and palette constraints, copyright and dataset provenance, and the reconfiguration of artist roles.
Keyword : Generative artificial intelligence, Pocedural content generation, Game graphic assets, 2D sprites, Pixel art, GANs, Diffusion models
http://dx.doi.org/10.29056/jncist.2025.11.10
Abstract : This study analyzes the construction process and operational performance of the Daegu Export Industries Biz Platform to explore the potential and limitations of metaverse platforms in the digital transformation of regional export industries. As the transition to non-contact trade environments accelerated after the COVID-19 pandemic, the metaverse has emerged as a core element of a new trade paradigm. The platform created differentiated value compared to existing export platforms by integrating advanced technologies such as cloud SaaS-based web accessibility, AI buyer matching systems, and NFT contract systems. In particular, an industry-specific strategy was implemented through the production of customized immersive 3D content for Daegu's three major industries.the Technology Acceptance Model (TAM) perspective, analysis of the platform's ease of use and usefulness showed high levels of user satisfaction and education satisfaction, demonstrating excellent acceptance. Operational results exceeded targets in export contracts, new buyer discovery, and buyer matching, proving quantitative performance. SWOT analysis identified technological superiority and regional specialization strategy as strengths, while shortage of specialized personnel and global network limitations were identified as weaknesses. The study suggests the potential for expansion of regionally customized metaverse platform models, along with the need to create an integrated digital trade ecosystem and strengthen digital capabilities of small and medium-sized enterprises.
Keyword : Digital Transformation, Metaverse Platform, Export Industries, Regional Specialization, Technology Acceptance Model
http://dx.doi.org/10.29056/jncist.2025.11.11
Abstract : This study focuses on the need for new educational methodologies to foster creativity, which has emerged as a key competency in the Fourth Industrial Revolution, and the potential for integrating Design Thinking, an innovative problem-solving framework, with Generative AI, which has emerged as a powerful tool to support human creative work. Therefore, this study aims to investigate the feasibility of designing and applying a GenAI-DT (Generative AI Integrated Design Thinking) workshop model to a curriculum that systematically integrates generative AI into each step of the design thinking process. Through the redesign of the traditional design thinking steps (empathy, problem definition, ideation, prototyping, and testing) with generative AI-based tools, it was found that generative AI can function as a multidimensional creative collaboration partner, serving as an idea generator, rapid visualization tool, and feedback simulator, beyond a simple information retrieval tool. This study offers important educational insights, presenting a new methodology and possibilities for enhancing creativity via human-AI collaboration.
Keyword : Generative AI, GenAI-DT, Creative Problem Solving, AI Literacy, Human-AI Collaboration
http://dx.doi.org/10.29056/jncist.2025.11.12
Table of Contents