- Adoption rate approximately 20%: This achievement is recognized at one of the world's most prestigious conferences in computer science.

A paper by the research team led by Professor Jong-June Jeon of the Department of Statistics at the University of Seoul has been accepted in the information retrieval track at CIKM 2025 (Conference on Information and Knowledge Management), the world's premier international conference in computer science.

CIKM is organized by the Association for Computing Machinery (ACM) and is a world-renowned conference where the latest research achievements in the fields of artificial intelligence, databases, and information retrieval are presented. It is classified as a top-tier conference by both the National Research Foundation of Korea and the Korean Institute of Information Scientists and Engineer, and has an acceptance rate of approximately 20-30% among thousands of papers submitted by scholars worldwide every year.

The selected paper, "Generalizing Query Performance Prediction under Retriever and Concept Shifts via Data-driven Correction," is a study on an artificial intelligence model that proactively predicts whether a given query in an information retrieval system can find relevant documents, and proposes a new approach based on Multi-label Classification (MLC). Unlike regression-based models that directly score query performance, this method demonstrated that it can accurately predict various target evaluation metrics (RR@10, nDCG@10, etc.) by directly learning the relevance of query-document pairs. This study confirmed that the existing regression-based models have a systematic bias toward information retrieval systems, which reduces their generalizability, whereas the proposed method can adjust the bias.

▶ Professor Jong-June Jeon, Department of Statistics; Doctoral Candidate JaeHwan Jung

The study was supported by the Data Science Convergence Talent Development Project Doctoral candidate JaeHwan Jung of the Department of Statistics and Data Science at the University of Seoul, the first author, and Professor Jong-June Jeon of the Department of Statistics, the corresponding author. The research team stated, "This achievement is significant in that it enables stable, high-accurate query performance prediction even when the dataset or search system varies," and added, "It is expected to have practical implications for large-scale search engines as well as generative AI-based search systems that have recently gained traction."

Furthermore, the team is conducting follow-up research to extend the technology beyond text-based search to a multi-modal retrieval system. It aims to predict performance in real-world environments where diverse types of queries are used (e.g., Google Lens, voice-based search), by focusing on enhancing generalization capabilities. Simultaneously, the team is working on designing lightweight models that enable rapid inference at low computational cost.