https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/268 2026-07-24T08:54:21Z 2026-07-24T08:54:21Z Yeo, Haein Noh, Taehyung Han, Kyungsik https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/213162 2026-06-09T02:00:24Z 2026-12-01T00:00:00Z

Title: Llm-generated content-based explanations for user experience in fashion recommender systems Authors: Yeo, Haein; Noh, Taehyung; Han, Kyungsik Abstract: Recommendation explanations are crucial in helping users make informed and confident decisions, especially in domains such as fashion, where personal style and preferences play an important role. While previous studies have predominantly used review data for explanations, the review-based method requires the availability and quality of a good number of reviews. To address this issue, we investigate the effectiveness of content-based recommendation explanations in fashion recommender systems. Using a Large Language Model (LLM) and deep learning techniques trained on fashion attribute data, we developed a framework that extracts essential visual information from product images and generates user-tailored explanations. This approach allows us to generate customized explanations at various levels—basic, simple, and detailed—for each recommendation. We developed a My Own Style (MOS) interface that displays fashion products, recommendations, and explanations. Our user study with 211 participants showed that detailed explanations, especially when combined with diversity-based algorithms, significantly improved user satisfaction and trust in fashion recommendations. This study contributes to clothing and textile research by providing guidelines for fashion-specific LLM prompts and demonstrating the effectiveness of LLM-generated explanations in fashion e-commerce. Our findings point the way to more personalized and transparent AI-driven fashion recommender systems that improve user experience and style exploration in fashion e-commerce.

2026-12-01T00:00:00Z Li, Tao Lee, Dong Yeop Gwac, Hocheol Hyeon, Jae Sang Choi, Jung Gi Lee, Nilüfer Çakmakçı Jeong, Youngjin Choi, Changsoon Kim, Seon Jeong https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/212936 2026-06-02T02:00:22Z 2026-10-01T00:00:00Z

Title: Coil order-transformational actuation for high tensile stroke up to 65% with coiled CNT/PDMS fiber Authors: Li, Tao; Lee, Dong Yeop; Gwac, Hocheol; Hyeon, Jae Sang; Choi, Jung Gi; Lee, Nilüfer Çakmakçı; Jeong, Youngjin; Choi, Changsoon; Kim, Seon Jeong Abstract: Coiled fiber actuators have emerged as promising functional materials for soft robotics and intelligent systems. However, their inability to dynamically switch between primary and secondary coiled structures during operation severely limits their actuation performance. In this work, we propose a coil order transformational actuation mechanism based on twisted PDMS/CNT composite fibers that enables reversible switching between primary and secondary coils under dynamic loading and thermal stimulation. By exploiting temperature-induced variations in fiber diameter and mechanical properties, we modulate the critical twist density for secondary coiling and thereby trigger the spontaneous formation of higher-order coils. Under a load of 55 kPa at 200 °C, the actuator achieves a tensile stroke of 65% of its maximum value and can be fully reset under a load of 80 kPa. We further demonstrate a gas overheat protection valve based on this actuator, in which the structure autonomously regulates flow and shuts off at a critical temperature. This study introduces a new actuation paradigm based on reversible structural reconfiguration, thereby significantly enhancing the performance of coiled fiber actuators.

2026-10-01T00:00:00Z Lim, Jaemin Kim, Sohui Son, Seungyeon Lee, Jong-Min https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/218425 2026-07-08T11:00:27Z 2026-08-01T00:00:00Z

Title: A fully inductive inference protocol for population GNNs in single-subject brain disorder diagnosis Authors: Lim, Jaemin; Kim, Sohui; Son, Seungyeon; Lee, Jong-Min Abstract: Population graph-based Graph Neural Networks (GNNs) have demonstrated superior performance in brain disease diagnosis by modeling inter-subject relationships. However, most existing approaches rely on a transductive setting that achieves high performance on known subjects but suffers from significant performance degradation when applied to unseen subjects. While a few inductive population graph models have been proposed, they struggle with single-subject inference, either due to a reliance on test batches for graph construction or limited generalization capabilities for individual unseen nodes. In this paper, we propose a fully inductive inference protocol in population graphs designed for single-subject diagnosis. Our approach constructs a population graph exclusively with training nodes and dynamically establishes connections between a single unseen test subject and the training graph during the inference phase based on imaging and phenotypic similarities. We conducted extensive experiments on multiple neuroimaging datasets (ABIDE I, ABIDE II, and ADHD-200) to evaluate the proposed pipeline. The results demonstrate that our method outperforms both state-of-the-art transductive models and existing inductive baselines under a fully inductive evaluation protocol. Furthermore, our analysis reveals that single-subject inference tends to maximize diagnostic performance within our experimental settings by reducing potential interference between test subjects. Importantly, our approach obviates the prohibitive retraining bottleneck typically required by transductive models, thereby providing an operational advantage for deployment and facilitating efficient, real-time single-subject inference workflows. The source code is available at https://github.com/98jaemin/single_subject_popgnn .

2026-08-01T00:00:00Z Hyeon, Jae Sang Song, Gyu Hyeon Sim, Jieun Choi, Jinyeong Choi, Ji In Jeong, Youngjin Kim, Seon Jeong https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/219370 2026-07-21T02:30:14Z 2026-08-01T00:00:00Z

Title: Asymmetric Carbon Nanotube Yarns for Electrochemical and Mechanical Balance in Artificial Muscle Fascicle Authors: Hyeon, Jae Sang; Song, Gyu Hyeon; Sim, Jieun; Choi, Jinyeong; Choi, Ji In; Jeong, Youngjin; Kim, Seon Jeong Abstract: Artificial muscle fascicles that mimic the hierarchical structure of biological muscles are essential for translating the high performance of individual artificial muscles into scalable soft robotics applications. However, in electrochemical artificial muscles, the muscle fascicles that consist of anodic and cathodic muscles have asymmetric actuation due to electrochemical imbalance between the anodic and cathodic sides, including voltage, capacitance, and ion volume. This imbalance reduces the overall actuation of muscle fascicles and poses a challenge for soft robot design. We here demonstrate an asymmetric configuration for carbon nanotube (CNT) artificial muscles to resolve both the electrochemical imbalances and the subsequent mechanical imbalance. The ratio of cathodic-to-anodic muscles in the fascicles was tuned to achieve electrochemical balance, and the spring index of the coiled structure was adjusted to match the mechanical modulus between the muscles. This asymmetric strategy was further extended to multiplied structures, forming the basis of artificial muscle fascicles with improved performance. These results provide a scalable strategy for translating high-performance individual CNT artificial muscles into efficient and powerful artificial muscle fascicles for future soft robotic systems.

2026-08-01T00:00:00Z