Donghyun Kim

Donghyun Kim

Auto-generated profile, based on publicly available information

Donghyun Kim

Naver Corporation, Seongnam, Republic of Korea

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 August 2024
Verified
Conference Paper
Open access
SKSein Kim
Sein Kim
HKHongseok Kang
Hongseok Kang
SCSeungyoon Choi
Seungyoon Choi
DKDonghyun Kim
Donghyun Kim
MYMinchul Yang
Minchul Yang
CPChanyoung Park
Chanyoung Park
Collaborative filtering recommender systems (CF-RecSys) have shown successive results in enhancing the user experience on social media and e-commerce platforms. However, as CF-RecSys struggles under cold scenarios with sparse user-item interactions, recent strategies have focused on leveraging modality information of user/items (e.g., text or images) based on pre-trained modality encoders and Large Language Models (LLMs). Despite their effectiveness under cold scenarios, we observe that they underperform simple traditional collaborative filtering models under warm scenarios due to the lack of collaborative knowledge. In this work, we propose an efficient All-round LLM-based Recommender system, called A-LLMRec, that excels not only in the cold scenario but also in the warm scenario. Our main idea is to enable an LLM to directly leverage the collaborative knowledge contained in a pre-trained state-of-the-art CF-RecSys so that the emergent ability of the LLM as well as the high-quality user/item embeddings that are already trained by the state-of-the-art CF-RecSys can be jointly exploited. This approach yields two advantages: (1) model-agnostic, allowing for integration with various existing CF-RecSys, and (2) efficiency, eliminating the extensive fine-tuning typically required for LLM-based recommenders. Our extensive experiments on various real-world datasets demonstrate the superiority of A-LLMRec in various scenarios, including cold/warm, few-shot, cold user, and cross-domain scenarios. Beyond the recommendation task, we also show the potential of A-LLMRec in generating natural language outputs based on the understanding of the collaborative knowledge by performing a favorite genre prediction task. Our code is available at https://github.com/ghdtjr/A-LLMRec.

3 August 2025
Verified
Conference Paper
Open access
SKSein Kim
Sein Kim
HKHongseok Kang
Hongseok Kang
KKKibum Kim
Kibum Kim
JKJiwan Kim
Jiwan Kim
DKDonghyun Kim
Donghyun Kim
CPChanyoung Park
Chanyoung Park

4 August 2023
Verified
Conference Paper
Open access
SKSein Kim
Sein Kim
NLNamkyeong Lee
Namkyeong Lee
DKDonghyun Kim
Donghyun Kim
MYMinchul Yang
Minchul Yang
CPChanyoung Park
Chanyoung Park
User modeling, which learns to represent users into a low-dimensional representation space based on their past behaviors, got a surge of interest from the industry for providing personalized services to users. Previous efforts in user modeling mainly focus on learning a task-specific user representation that is designed for a single task. However, since learning task-specific user representations for every task is infeasible, recent studies introduce the concept of universal user representation, which is a more generalized representation of a user that is relevant to a variety of tasks. Despite their effectiveness, existing approaches for learning universal user representations are impractical in real-world applications due to the data requirement, catastrophic forgetting and the limited learning capability for continually added tasks. In this paper, we propose a novel continual user representation learning method, called TERACON, whose learning capability is not limited as the number of learned tasks increases while capturing the relationship between the tasks. The main idea is to introduce an embedding for each task, i.e., task embedding, which is utilized to generate task-specific soft masks that not only allow the entire model parameters to be updated until the end of training sequence, but also facilitate the relationship between the tasks to be captured. Moreover, we introduce a novel knowledge retention module with pseudo-labeling strategy that successfully alleviates the long-standing problem of continual learning, i.e., catastrophic forgetting. Extensive experiments on public and proprietary real-world datasets demonstrate the superiority and practicality of TERACON. Our code is available at https://github.com/Sein-Kim/TERACON.

1 November 2019 in INFORMS Journal on Applied Analytics
Verified
Journal Article
Closed access
DKDonghyun Kim
Donghyun Kim
NKNamyong Kim
Namyong Kim
JCJunoh Cho
Junoh Cho
HSHayong Shin
Hayong Shin
The admission decision process is an important operational management problem for many universities. Admission control processes may, however, differ among universities. In this paper, we focus on the problem at Korea Advanced Institute of Science and Technology (KAIST). We assume that individual applications are evaluated and ranked based on paper evaluations and (optional) interview results. We use the term “university admission decision” to mean determining the number of admission offers that will meet the target number of enrollments. The major complexity of an admission decision comes from the enrollment uncertainty of admitted applicants. In the method we propose in this paper, we use logistic regression with past data to estimate the enrollment probability of each applicant. We then model the admission decision problem as a Markov decision process from which we formulate optimal decision making. The proposed method outperformed human expert results in meeting the enrollment target for the validation data in 2014 and 2015. KAIST successfully used our method for its admission decisions in academic year 2016.

31 January 2021 in Journal of Communication Design
Verified
Journal Article
Closed access
ZWZe-Jun Wang
Ze-Jun Wang
DKDong-Hyun Kim
Dong-Hyun Kim
영화와 건축은 각각 독자적인 예술 형태로 언어표시 시스템을 갖추고 있다. 영화에서 건축은 보통 필수적인 요소이며, 인물 활동의 지점으로서 이야기가 일어나는 지점을 알려준다. 그러나 가장 큰 시각적 요소인 건물이 기본적인 기능 외에 영화적 요소들과 깊이 결합해 더 중요한 역할을 할 수 있는가? 본 연구는 이를 출발점으로 건축요소 중 유리 문창으로 대상을 한정하였는데, 이는 투명, 반사, 통과 가능, 깨짐 가능, 스위치 가능 다양한 조형 등의 특징이 있어 가소성이 높기 때문이다. 본고의 연구 사례로서 프랑스 감독 자크 타티(Jacques Tati)의 작품 <나의 삼촌>(Mon Oncle)과 <플레이타임>(Play Time)에도 여러 차례 유리 요소가 활용된 사례가 있다. 분석을 통해 건축요소인 유리 문창이 카메라 앵글, 인물 행동과 동선 등과 결합할 때 시각적 재미 표현, 인물의 성격 표현, 스토리 구성, 영화적 주제 표현 등 다양한 역할을 한다는 사실을 발견했다.

12 January 2015 in IEEE Micro
Verified
Journal Article
Closed access
GKGyeonghoon Kim
Gyeonghoon Kim
DKDonghyun Kim
Donghyun Kim
SPSeongwook Park
Seongwook Park
YKYouchang Kim
Youchang Kim
KLKyuho Lee
Kyuho Lee
HYHoi-Jun Yoo
Hoi-Jun Yoo
For all day long operable markerless augmented reality system, low-power BONE-AR processor is implemented to execute object recognition, camera pose estimation, and 3D graphics rendering in real-time for a HD resolution video input. BONE-AR adopts 6 clusters of heterogeneous SIMD processors distributed on the mesh topology network-on-chip (NoC) to exploit data-level parallelism and task-level parallelism. Visual attention algorithm reduces overall workload by removing background clutters from the input video frames, but also incurs NoC congestion due to dynamically fluctuating workload. We propose a congestion-aware scheduler (CAS) that detects and resolves the NoC congestion to prevent throughput degradation of task-level pipeline.