- Trained academic reranker achieving state-of-the-art performance in the scholar domain
- Designed and conducted search system evaluation
- Built and maintained data ingestion pipeline
- Managed search infrastructure (Elasticsearch/Vespa)
Eunseong Choi (최은성)
I am a Research Scientist at Liner, working on helping people get smarter faster. I received my Ph.D. from DIAL Lab at Sungkyunkwan University, advised by Prof. Jongwuk Lee. My research lies in Information Retrieval (IR) and Natural Language Processing (NLP), with a focus on building robust and efficient Retrieval-Augmented Generation (RAG) frameworks. In particular, I study context–memory conflicts, prompt compression, and reasoning-intensive retrieval to enable scalable and reliable real-world applications.
- Retrieval-Augmented Generation: Mitigating context–memory conflicts, incorporating evidentiality, and guiding robust LLM reasoning
- LLM Efficiency & Compression: Reducing computational cost through hard and soft prompt compression while preserving semantic fidelity
- Document Retrieval: From efficiency-focused sparse retrieval to reasoning-intensive retrieval for complex information needs
Keywords: Retrieval-Augmented Generation, Efficient LLMs, Information Retrieval, NLP
Work Experience
Focus: Continual fine-tuning for recency-aware dense retrieval
Focus: Learned sparse retrieval with uni-encoder for efficient first-stage retrieval
Education
Advisor: Prof. Jongwuk Lee
Thesis: Improving Evidentiality and Compression for Retrieval-Augmented Generation
B.S., Samsung Convergence Software Course
Publications
Under Review
- Generative Log Anomaly Detection with Large Language Models
Youngbin Kim, Hyunsoo Kim, Jubong Park, Eunseong Choi, Jongwuk Lee
Under Review
International Conferences
- Multi-view-guided Passage Reranking with Large Language Models
[paper] [code]
Jeongwoo Na*, Jun Kwon*, Eunseong Choi, Jongwuk Lee (* : equal contribution)
EMNLP 2025, Suzhou, China - Conflict-Aware Soft Prompting for Retrieval-Augmented Generation
[paper] [code]
Eunseong Choi, June Park, Hyeri Lee, Jongwuk Lee
EMNLP 2025, Suzhou, China - GRAM: Generative Recommendation via Semantic-aware Multi-granular Late Fusion
[paper] [code]
Sunkyung Lee, Minjin Choi, Eunseong Choi, Hye-young Kim, Jongwuk Lee
ACL 2025, Vienna, Austria - From Reading to Compressing: Exploring the Multi-document Reader for Prompt Compression
[paper] [code] [poster] [slide]
Eunseong Choi, Sunkyung Lee, Minjin Choi, June Park, Jongwuk Lee
EMNLP Findings 2024, Miami, Florida - Multi-Granularity Guided Fusion-in-Decoder
[paper] [code] [poster] [slide]
Eunseong Choi, Hyeri Lee, Jongwuk Lee
NAACL Findings 2024, Mexico City, Mexico - Forgetting-aware Linear Bias for Attentive Knowledge Tracing
[paper] [code]
Yoonjin Im*, Eunseong Choi*, Heejin Kook, Jongwuk Lee (* : equal contribution)
CIKM 2023, Birmingham, UK - ConQueR: Contextualized Query Reduction using Search Logs
[paper] [code]
Hye-young Kim*, Minjin Choi*, Sunkyung Lee, Eunseong Choi, Young-In Song, Jongwuk Lee (* : equal contribution)
SIGIR 2023, Taipei, Taiwan - SpaDE: Improving Sparse Representations using a Dual Document Encoder for First-stage Retrieval
[paper] [code]
Eunseong Choi*, Sunkyung Lee*, Minjin Choi, Hyeseon Ko, Young-In Song, Jongwuk Lee (* : equal contribution)
CIKM 2022, Atlanta, Georgia, USA - Long-tail Mixup for Extreme Multi-label Classification
[paper]
Sangwoo Han, Eunseong Choi, Chan Lim, Hyunjung Shim, Jongwuk Lee
CIKM 2022, Atlanta, Georgia, USA - MelBERT: Metaphor Detection via Contextualized Late Interaction using Metaphorical Identification Theories
[paper] [code] [slide] [video]
Minjin Choi, Sunkyung Lee, Eunseong Choi, Heesoo Park, Junhyuk Lee, Dongwon Lee, Jongwuk Lee
NAACL 2021, Mexico City, Mexico (Virtual Event)
Domestic Journals & Conferences
- 지식 추적 모델의 성능 개선을 위한 양자화된 정답률 임베딩 방법
[paper]
임윤진, 문재완, 최은성, 이종욱
정보과학회논문지 (Journal of KIISE) Vol.50 No.4 [2023]: 329-336, April 2023 - 적은 양의 병렬 말뭉치를 가진 한국어 방언 간 딥 러닝 기반 기계번역
[paper]
한상민, 최은성, 이종욱
한국정보과학회 학술발표논문집 Vol.2022 No.12 [2022]: 1619-1621, Dec 2022 - 의사 문장 표현을 활용한 수학 문장형 문제 풀이 모델 (우수발표논문상)
[paper]
김지우, 이선경, 최은성, 이종욱
한국정보과학회 학술발표논문집 Vol.2022 No.06 [2022]: 446-448, Jun 2022 - 지식 추적 모델의 정확도 개선을 위한 양자화된 정답률 임베딩 방법 (우수논문상)
[paper]
임윤진, 문재완, 최은성, 이종욱
한국정보과학회 학술발표논문집 Vol.2022 No.06 [2022]: 1172-1174, Jun 2022 - 기계 독해 성능 개선을 위한 데이터 증강 기법
[paper]
이선경, 최은성, 정선호, 이종욱
정보과학회논문지 (Journal of KIISE) Vol.48 No.12 [2021]: 1298-1304, Nov 2021
Honors & Awards
- 🏆 1st Prize, Best Graduate Research Paper Award – SKKU, 2024 (KRW 5,000,000)
- 2nd Place, AI Grand Challenge for Policy Support AI – IITP, 2023
- 🥇 1st Place, AI Grand Challenge for Math Word Problem Solving – IITP, 2022
- 2nd Prize, Best Graduate Research Paper Award – SKKU, 2021 (KRW 6,000,000)