Earticle

This study analysed the characteristics of perceptual similarity judgements for colour-category boundary pairs in a digital display environment. Ten adjacent colour pairs on the Munsell hue circle were selected and a repeated-measures experiment was conducted in which participants rated, on a 7-point scale, the similarity of the right-hand stimulus relative to the left-hand stimulus. The Pearson correlation coefficient between Session 1 and Session 2 (r = .796) and the ICC (.80) indicated high repeat reliability and the paired-samples t-test also showed no significant difference. ANOVAs revealed statistically significant differences in similarity judgements across colour pairs for Session 1, Session 2, and their combined mean, supporting systematic effects of category boundaries. The regression analysis on lightness difference found no significant effect, as all colour pairs were controlled for lightness. Pairs such as Y–GY, B–PB, and GY–G consistently exhibited lower similarity scores, while P–RP and G–BG were perceived as more similar. This study provides evidence that perceptual sensitivity at colour-category boundaries exerts a systematic influence on colour perception. These results imply that boundary-based colour choices can impact user experience by unintentionally increasing or decreasing perceived similarity in digital design contexts.

This study aimed to provide foundational data for developing exercise strategies to mitigate postprandial blood glucose spikes by comparing the effects of different intensities of circuit weight training on heart rate, exercise volume, and blood glucose responses in healthy men in their 20s. Ten participants completed three conditions in a randomized crossover design: non-exercise, hypertrophy-oriented (70–80% 1RM), and muscular endurance-oriented (50–60% 1RM) circuit training. Each session included eight resistance exercises over two circuit cycles with differing rest intervals. Repeated-measures two-way ANOVA and paired t-tests were used for analysis. No significant heart rate difference was found between exercise intensities, though heart rate increased in the second cycle across both. Total exercise volume was significantly higher in the muscular endurance condition. Blood glucose peaked 20 minutes post-meal and declined, followed by a rebound postexercise. At 40 and 60 minutes postprandially, both exercise groups showed significantly lower glucose levels than the non-exercise group, with no difference between the two training intensities. These results suggest that exercise volume and rest intervals may impact postprandial glucose more than intensity alone.

This study aims to analyze the key trends in the global restaurant industry and the changes in consumer dining behavior, providing implications for effective global marketing strategies. In recent years, the foodservice industry has undergone structural transformations driven by factors such as health consciousness, sustainability, and digital transition, while the COVID-19 pandemic has accelerated the adoption of contactless services and digital platforms. Using big data sources including online reviews, news articles, and social media posts, this study applies text mining techniques to identify major themes and patterns. The results indicate that consumer values have shifted from price and quality to a focus on health, convenience, experience, and sustainability. Furthermore, differences in preferences and value orientations across countries and brands highlight the importance of balancing localization and standardization in global brand strategies. This research complements the limitations of traditional survey-based studies by providing empirical evidence of changes in industry trends and consumer perceptions through big data analysis. We expect to impact the field by offering strategic insights into how global restaurant businesses can achieve sustainable competitiveness in a rapidly evolving market.

We investigated the effects of a single-session health tourism program, combining forest walking and hot spring bathing, on heart rate variability (HRV) and electroencephalographic (EEG) activity. Eight healthy male athletes with no history of cardiovascular or respiratory disease participated in this study. The intervention consisted of one hour of forest walking followed by one hour of aquatic exercise and hot spring bathing. HRV indices—including mean heart rate, the root mean square of successive differences (RMSSD), low-frequency (LF) and high-frequency (HF) power, and the LF/HF ratio—were assessed using an SA-3000P analyzer, and EEG activity (theta, alpha, and beta waves) was recorded before and after the intervention with a ProComp Infiniti system. Pre- and post-intervention values were compared using paired-samples t-tests (α = 0.05). The results indicated a significant decrease in mean heart rate and a significant increase in RMSSD (p < 0.05). However, no significant changes were observed in the other HRV parameters (LF, HF, LF/HF) or EEG components (theta, alpha, beta). In conclusion, the integrated forest walking and hot spring program elicited beneficial autonomic nervous system responses, as evidenced by improvements in HRV, although it did not significantly affect EEG indicators.

We focused on evaluating and comparing two NPC escape strategies—NavMesh, a traditional pathfinding algorithm, and ML-Agents, a reinforcement learning-based system—within a Unity-based simulation environment. We conducted 100 controlled simulations under identical environmental conditions, setting average survival time, distance to the player, and number of obstacle collisions as quantitative performance indicators. The experimental results demonstrated that the ML-Agents model achieved superior adaptability and spatial utilization, maintaining longer escape durations and broader movement paths. In contrast, the NavMesh method produced more stable performance in precise control, particularly in obstacle avoidance. Through this study, we highlight the flexibility and learning capability of reinforcement learning-based AI in dynamic game scenarios and propose the importance of compensation design and environmental diversity for enhancing its behavioral realism. This research provides practical insights for developing adaptive NPCs that balance precision and adaptability in game AI design.

This study presents a reinforcement learning-based simulation of agent behaviors within a synthetic ecosystem environment. Using Unity ML-Agents and the Proximal Policy Optimization (PPO) reinforcement learning algorithm, we designed three distinct agents—a chicken, a dog, and a tiger—each with unique survival objectives such as foraging or hunting. These agents autonomously learned their behaviors through interaction and reward feedback, without predefined rules. We investigated how variations in reward structure and environmental complexity influenced policy convergence and strategic behavior formation. Experimental results showed that each agent developed distinct behavioral strategies aligned with its reward design. Furthermore, despite the high complexity of the simulated environment—including uneven terrain and multiagent interactions—all agents achieved stable learning convergence when reward signals were properly calibrated. This work contributes to the modeling of reward-driven adaptive strategies in multi-agent ecological simulations and offers a foundation for future studies involving cooperation, emergent behavior, or survival-based competition.

Most current video summarization techniques focus on converting long videos into short summary clips by detecting semantically important parts of the original video. However, many summarization applications, such as personal summaries of weddings, travel videos, movie trailers, or short summaries for search, require summaries that include the entire content of the original video, not just the important parts. This paper proposes a technique for generating summary videos by combining highlight scenes representing important parts of the video with story-conveying scenes capturing the overall flow of the video. To achieve this objective, we introduce the concept of 'diversity contribution,' which numerically quantifies how much each individual scene contributes to creating a diverse scene composition in the summary video. The higher the proportion of scenes with high diversity contribution values, the more comprehensively the summary video incorporates the entire content of the original video. In this study, we developed an algorithm to create a summary video by adjusting the diversity contribution and importance of scenes in proportion, and verified its operation by implementing an actual summary application system. Furthermore, we confirmed through experiments that the accuracy of the summary technique presented in this paper was over 90%.

With growing risks to student data, there is a greater need for digital wallets that respect privacy in online education. Our goal is to keep information private, hard to link, and usable, without using blockchain. We are ensuring that our solution is easy to use and meets the standards for both schools and students. The wallet utilizes basic public and private keys to safeguard information, and a standard login system to simplify the login process. We use a method that only displays the necessary information, helping to keep other details private. A special signature method stops different services from linking your information together. We track credential status using a small list of credentials. The system has a short memory to keep things fast and upto- date. We ensure that the phone or computer can be used even when offline, for rentals and payments. Checks still work even if the internet is unreliable. We control spending using clear rules and limit digital payments by item, amount, region, and time. We start with basic logins, then add new ways to share credentials, and finally add advanced signing tools. We also explore methods to verify information online without relying on complex technology, such as using trusted computers or hidden signatures, to meet specific needs. More complicated methods are saved for special cases. This keeps our design simple but flexible. As a result, the system limits the amount of data that is shared or connected, while still allowing for checks and growth. This leads to easier verification and strong privacy in different situations.

This study aims to address the need for more research on the interplay between two types of social norms, namely descriptive and injunctive norms, and two components of perceived behavioral control, namely capacity and controllability. Specifically, we examined their direct influence on the intention to reduce disposable product use. Furthermore, the moderating roles of capacity and controllability were explored. Analyzing the survey responses of 160 college students revealed that capacity exerted direct influence on the intention to reduce disposable product use and acted as a moderator in the attitude–intention and injunctive norm–intention relations. Controllability did not directly influence the intention; rather, it served as a moderator in the injunctive norm–intention relation. The results of this study indicate that sustainability campaign messages should be structured differently depending on the target group’s level of capacity and controllability regarding the behavior.

This paper investigates the impact of virtual reality technology on surrealist art and explores its potential for convergence. Surrealism endeavors to visualize the unconscious, dreams, and irrational emotions in order to express a world that transcends reality; virtual reality, as an emerging medium, has attracted significant attention for its capacity to actualize such artistic imagination. Anchored in Freud’s concept of dreams, this paper explores the philosophical foundations of Surrealism and provides a theoretical account of the characteristics and aesthetic functions of virtual reality as a medium. It then draws on case studies such as Timo Helgert’s AR art projects and the VR immersive experiences at the Dalí Museum to analyze how Surrealist aesthetics are concretely materialized in digital environments. Building on these discussions, the paper argues that virtual reality constitutes a pivotal medium for reinterpreting and expanding Surrealism, while also reflecting on future trajectories of artistic creation and aesthetic experience.

This paper presents a comprehensive comparative analysis of two optimization algorithms: the widely-adopted Adam optimizer and the recently proposed MuonClip algorithm. Optimization algorithms play a crucial role in deep learning, directly influencing both convergence speed and final model performance. Through systematic comparison of mathematical structures, computational complexity, convergence characteristics, and empirical performance, we analyze the strengths and limitations of each algorithm. Our findings suggest that MuonClip offers superior memory efficiency and stability, while Adam provides finer adaptivity and broader applicability. These insights provide practical guidelines for algorithm selection in different scenarios.

This paper measures the end-to-end impact of inserting a Jejueo→Standard-Korean normalization layer before a retrieval-augmented generation (RAG) QA system and of merging results from two retrieval paths. Under an identical stack, we compare: (i) the original-Jejueo retrieval path, retrieve with the unnormalized query (RAW), (ii) the normalized-to-Standard-Korean retrieval path, retrieve with the query rewritten by our Jejueo→Standard-Korean normalizer (NORM), and (iii) the merged path, deduplicate and re-rank the union of RAW and NORM candidates (DUAL). Using the Jejueo Interview Transcripts, lightweight preprocessing, and a compact LoRA-based normalizer, we evaluate retrieval/ranking (Hit@5, MRR@5), evidence attribution (Attrib-F1), answer quality (Token-F1/ROUGE-L), diversity (ILD@5), and hallucination. Normalization alone improves top-rank concentration and citation precision over RAW (Token-F1 +6–7; Attrib-F1 +0.09; Hit@5 ≈ +0.11), and DUAL adds further gains (≈ +4 Token-F1; +0.06 Attrib-F1), increases diversity, and lowers hallucination (≈ −1.7%p). Error analyses trace benefits to aligning Jejueo sentence-final endings, discourse markers, and idioms that otherwise destabilize retrieval; DUAL also hedges over/under-substitution from imperfect normalization. Our intent is practical: use DUAL by default for robustness, and prefer NORM alone when normalization confidence is high and latency/compute budgets are tight.

In the digital preservation and exhibition of cultural heritage, existing photogrammetry techniques face limitations in real-time applications due to long processing times and large file sizes. This study proposes a methodology for producing high-quality holographic content for Gyeongju Cheomseongdae by integrating drone-based data collection with Gaussian Splatting technology. Experimental results show that Gaussian Splatting reduces processing time by approximately 60% compared to photogrammetry, increases file size by 416%, and improves performance by 11.4% and 9.0% in PSNR and SSIM metrics, respectively. The holographic output achieved a resolution of 51 DPI, color accuracy of ΔE76=2.3, a 120° field of view, and a 12 ms parallax response time, meeting the practical requirements for museum exhibition environments. This study presents an empirical solution for the efficient digital preservation and immersive exhibition of cultural heritage.

We explored the cultural and narrative “scripts” of sexual intercourse conveyed in user-produced Korean Internet pornographic stories from the early digital era. A total of one hundred stories were collected from Sora.net and analyzed using the framework of fantasy theme analysis. The analysis revealed six dominant themes: sexual drive as an irresistible force, male gaze, men as leaders of sexual interaction, gendered communication, female vulnerability, and goals of intercourse. Together, these themes construct a rhetorical vision of sexual conduct centered on male dominance and female passivity, occasionally disrupted by alternative accounts. We interpret these findings as reflecting the broader gender ideologies and cultural discourses of early online sexual storytelling in Korea. By examining how Internet users collectively imagined and narrated sexual relations, we highlight the historical, cultural, and interpretive significance of these early digital narratives as cultural precursors to later controversies surrounding the representation of sexuality, gender, and ideologies in digital communication.

The modern retail environment is no longer limited to transactional exchanges; it has evolved into a space for experiences and engagement. This study examines the impact of visual merchandising (theme-centered vs. product-centered) and shoppertainment (related vs. unrelated) environments on two key consumer behaviors: impulse purchase and social sharing. This study employs a 2 × 2 experimental design in the context of skin care products, examining how these strategies independently and jointly influence consumer responses. The findings highlight the importance of aligning visual presentation with relevant entertainment to maximize both unplanned buying and shareable experiences. This study contributes to the retail and consumer behavior literature, providing practical guidance for retailers seeking to design environments that promote both immediate sales and long-term brand engagement.

This paper argues the emergence of artificial intelligence (AI)-based acting performance opens new artistic horizons that existing concepts of representation and expression cannot fully capture. While digital humans mimicking human appearance and emotional expression spread across the content industry, this phenomenon raises fundamental questions about acting as a human-centric art form. Traditional representation presupposes an original and expression is based on a human subject's inner experience, whereas AI acting is characterized by imitation without an original and emotion simulation without interiority. This study employs conceptual analysis and literature review as primary methodologies, focusing on media theory, performance theory, and posthumanism to build a new analytical framework. This paper proposes "Generative Simulation" as a comprehensive superordinate concept, which encompasses three key sub-components: "Algorithmic Corporeality" defining the AI's data-driven body, "Simulated Affect" describing its imitation of emotion devoid of inner experience, and "Generative Agency" explaining its role as a non-conscious participant in the creative process. This framework provides a theoretical foundation for re-evaluating human actors' unique artistic value and redefining the art-technology relationship.

Recently, research has been actively conducted to determine which drugs are the most effective treatments without side effects for patients with Parkinson's disease, heart disease, and chronic diseases, using artificial intelligence generation technology and virtual model-based technology. In this paper, we performed a computer simulation to self-diagnose patients at risk of heart disease and predict risk levels on a web-based platform. Furthermore, this paper established a hypothesis theory on the probability of heart disease risk occurrence, and performed factor analysis, correlation analysis, and fuzzy inference computer simulations to predict heart disease risk probability and improve performance. The results of the computer simulations confirmed that the fuzzy inference method improved heart disease risk prediction by over 10% compared to existing methods.

Filmmaking is often demanding in both financial and human resources, posing challenges for smallscale creators—students or independent filmmakers. However, artificial intelligence (AI) now helps streamline procedures and improves efficiency. This study aims to assess the effectiveness of AI assistance in tackling the hurdle of casting multiple actors, which is often financially demanding. In doing so, this study focuses on evaluating AI-assisted face-swapping and its as a cost-effective alternative, by comparing two AI platforms, ComfyUI and Akool. Combined with AI’s ability to generate videos and digital humans, this study evaluates each platform’s accessibility, workflow, performance, and output quality. The findings reveal that Akool provides higher visual consistency and user-friendliness, while ComfyUI offers extensive customization at no cost, but requires more technical skill and results in lower clarity. Both struggle in replicating complex expressions and facial obscurity, though they may be mitigated through narrative design. By highlighting the strengths, weaknesses, and ethical implications of each platform, this study demonstrates how AI-assisted face-swapping can enhance narrative possibilities in low-budget filmmaking.

This paper explores how slow cinema reconstructs romantic storytelling, focusing on temporal duration, emotional realism, and embodied perception. Findings suggest that aesthetic slowness fosters emotional depth and reflective engagement by slowing narrative time and encouraging immersive spectatorship. While many viewers reported heightened empathy and affective resonance, others expressed frustration with the film’s pacing, revealing how audience reception is shaped by cultural context and media habits. The study argues that slow cinema challenges dominant romantic storytelling by privileging process over resolution, presence over spectacle. In doing so, it opens new possibilities for presenting intimacy in film, not as a climactic narrative arc, but as an evolving, embodied condition shaped through time.

The development of highly sensitive and selective biosensors is essential for applications ranging from medical diagnostics and environmental monitoring to food safety and biosecurity. In this paper, a silicon-based dielectric diffraction grating of molar shape composed of symmetrical and asymmetrical gratings is proposed for designing a high-efficiency optical biosensor. The reflection characteristics are analyzed using a rigorous Eigenvalue Problem of Modal Transmission-Line Theory (EP-MTLT) in the visible wavelength region. The effects of grating period, thickness of the homogeneous reflecting layer, and grating thickness on the reflectance spectra are investigated to optimize the structure. Simulation results show strong guided-mode resonance (GMR) characteristics with narrow linewidths and high sensitivity to the surrounding refractive index. The optimized design demonstrates a linear wavelength shift in response to refractive index variation, achieving sensitivities on the order of tens of nm/RIU, which is suitable for label-free biosensing applications. This study provides theoretical guidance for the design of compact, highly sensitive dielectric biosensors based on double-layered molar-shaped gratings.

In this paper, we propose a novel, fast model for detecting fire flame and smoke using object detection and 3D classification, referred to as FastFireDet3D. This model uses NanoDet to quickly identify potential areas representing fire and smoke, followed by a novel 3D classification model based on a spatio-temporal convolutional neural network (STCNN). This two-step process allows for efficient and accurate detection. The average processing time for FastFireDet3D is approximately 40-90ms when run on a CPU, and it achieves an accuracy improvement of 3.45% over traditional Convolutional 3D (C3D) models.

This study presents the design and implementation of an educational simulation game that reproduces fire situations in a virtual environment to improve users’ safety response capabilities. The proposed system employs a probability-based cellular partitioning algorithm, which divides indoor spaces into grid-shaped cells and probabilistically determines fire spread based on each cell’s material, temperature, and location. This approach enables the efficient representation of realistic fire spread patterns while maintaining performance suitable for real-time gameplay. The game is set in a 3D graphic environment and played from a first-person perspective, allowing players to experience the entire process from fire outbreak to evacuation using various suppression and evacuation tools such as fire extinguishers and escape ladders. This paper analyzes the structure and learning effects of the proposed system, and discusses both the educational applicability and technical limitations of the probability-based simulation model. Future work will explore enhancing educational effects through the inclusion of more complex fire scenarios, user behavior AI, and VR interface integration.

The use of camera images for estimating object locations and distances has increased substantially, resulting in greater sensitivity to image distortion caused by camera lenses. Existing approaches to estimating and correcting lens distortion coefficients are limited because they apply uniform corrections to the entire image. This study introduces a method that enhances correction performance by approximately 30 percent compared to conventional techniques by estimating lens distortion coefficients for individual regions within the image and correcting objects within those regions. The proposed approach enables the accurate estimation of two-dimensional coordinates for target objects in specific image regions, yielding significantly improved image quality when the entire image is divided into regions and corrected sequentially.

MIMO (multi-input multi-output)–OFDMA (orthogonal frequency division multiple access) systems have been recognized as a key technology for next-generation wireless communications, offering high spectral efficiency and multiuser support. However, their performance is constrained when the number of users exceeds the allocated bandwidth and further deteriorates under adverse wireless environments. In this paper, we propose a novel scheme that integrates beamforming and distributed subcarrier allocation (DSA) in MIMO–OFDMA systems to address performance degradation caused by interference and adverse channel conditions. We mitigate interference through beamforming and enhance robustness against channel impairments via distributed allocation, thereby improving BER performance and increasing user capacity. We further conduct computer simulations to validate the effectiveness of the proposed scheme in terms of both user capacity and BER performance, demonstrating its potential as a valuable contribution to next-generation wireless communication research.

This study proposes a distributed reinforcement learning system that incorporates the Safe Proper Time (SPT) protocol to address latency issues in cloud-based environments. The system is architected to operate efficiently under limited computational resources, making it suitable for small-scale enterprises. By combining physicallevel technologies such as InfiniBand and TOE with software-level optimization, the SPT protocol enables low-latency, high-throughput data transmission across distributed nodes. Experimental results show that the proposed system reduces response failure rates and achieves faster processing times compared to centralized models. Furthermore, a comparative analysis demonstrates that the system offers competitive advantages over existing machine learning platforms in terms of deployment flexibility and initial cost efficiency. This research contributes to the field by presenting a scalable and resource-efficient approach to distributed reinforcement learning. Future work will focus on enhancing the security and stability of data transmission in SPT-based systems.

Cloud computing has become a critical infrastructure for large-scale data processing and machine learning applications. However, task failures frequently occur due to distributed resources and dynamic operating environments, leading to service delays and resource waste. To address this issue, this paper proposes a task failure prediction system based on stacking ensemble learning. The proposed system is structured in three layers: log collection, prediction, and result application. Base model predictions are integrated by a metamodel to produce the final outcome. Experiments conducted with the Google Cluster Trace 2019 dataset demonstrate that the proposed system outperforms single models in terms of prediction accuracy and stability, providing a robust and scalable framework suitable for real-world cloud environments.

In Business Intelligence (BI) environments, organizations increasingly face challenges in integrating data distributed across multiple heterogeneous domains. Traditional systems mainly address structural or schemalevel differences. However, they often overlook inconsistencies in the meanings of domain-specific terms. These limitations can lead to misinterpretation and reduce the effectiveness of data analysis. To address these issues, we designed and propose a meaning-based inter-domain mapping system. This system enables semantic integration across multiple domains. First, it collects distributed domain information from various sources. Then, it measures semantic similarity between fields using text embeddings and cosine similarity, and clusters related fields into groups. Representative meaning units are selected for each cluster and stored in a central domain repository, which supports consistent interpretation and semantic queries. By focusing on meaning rather than solely on structure, the system ensures more accurate and efficient data utilization. It also reduces redundancy and semantic ambiguity, supporting better decision-making in BI analyses. Experimental evaluation shows that the system effectively consolidates semantically similar domains and enhances analytical reliability.

This study presents a real-time interactive VR logistics training system by integrating YOLOv8-based object detection and tracking into a Unity + Sentis platform. To support realistic logistic scenarios, we constructed a custom VR logistics dataset consisting of approximately 15,000 images across six key object categories: boxes, forklifts, conveyors, people, AGVs, and robotic arms. Leveraging a YOLOv8 model trained on our domain-specific logistics dataset, we integrated real-time object detection with the Unity 3D coordinate system to enable synchronized interactions such as AGV obstacle avoidance and robotic arm operations. Experimental results demonstrate that the proposed system achieves high detection accuracy and real-time inference speeds exceeding 30 FPS (achieving 50-60 FPS on Apple M3 GPU with MPS acceleration), ensuring smooth performance in immersive VR environments. This approach addresses the limitations of real-time object recognition and responsiveness in conventional VR logistics training systems, expanding the potential for an intelligent and immersive training platform that dynamically adapts to learner behavior and training scenarios. Our research provides the foundation for establishing a digital twin/metaverse-based smart logistics education ecosystem and contributes to cultivating next-generation logistics professionals with enhanced safety awareness and operational efficiency.

As modern society faces global problems such as climate change, resource depletion, and environmental pollution, interest in sustainability and eco-friendly consumption is expanding across all industrial groups. In particular, the fashion industry has such a big impact on the environment that it is pointed out as the world's second-largest environmental pollution industry, that sustainable fashion and the use of eco-friendly materials are emerging as important alternatives. This study intends to analyze how consumers perceive the use of ecomaterials by Korean fashion companies. Using Textom, "Eco Fashion" data is collected through SNS, blogs, and news comments to derive keyword frequency and related words. After that, UCINET is used to conduct visualization through network analysis between keywords and CONCOR analysis. This study intends to provide basic data for the spread of eco-friendly fashion by understanding how actual consumers perceive this and how it is affected by purchasing decisions. Through this, the fashion industry will be able to derive more effective communication strategies and product planning directions, and it will also be able to present practical implications for consumer education and policy development.

This study analyzed the factors influencing satisfaction with Korean dining experiences using online review data written by Chinese tourists. Reviews collected using Textom were categorized into atmosphere, hygiene, and service. Sentiment analysis, TF-IDF keyword analysis, and CONCOR semantic network analysis were then employed to understand the characteristics and structure of each factor. The analysis revealed differing proportions of positive and negative sentiment, distinct core keyword structures, and varying levels of perceived connectivity between words for each factor. This research provides practical data for the Korean food service industry to design marketing strategies and service improvement directions that align with tourists' expectations.