본문 바로가기
  • Home

Advancing Digital Reference Services through a Hybrid RAG-Based Model: Integrating the “Ask a Librarian” Knowledge Base with Generative AI

  • Journal of the Korean Biblia Society for Library and Information Science
  • 2026, 37(1), pp.287~310
  • DOI : 10.14699//kbiblia.2026.37.1.287
  • Publisher : Journal Of The Korean Biblia Society For Library And Information Science
  • Research Area : Interdisciplinary Studies > Library and Information Science
  • Received : February 19, 2026
  • Accepted : March 18, 2026
  • Published : March 30, 2026

Jeonghoon Lim 1,  Kim Sun Tae 2

1계명대학교
2전북대학교

Accredited

ABSTRACT

This study aims to experimentally analyze the performance of a hybrid search system that combines keyword-based and vector-based retrieval, utilizing the Knowledge Information Database accumulated through the National Library of Korea’s “Ask a Librarian” collaborative digital reference service. The dataset consists of 5,898 question-answer records from the Knowledge Information Database, categorized according to the ten main classes of the Korean Decimal Classification (KDC). The search system was implemented in a Python-based experimental environment, employing an inverted-index-based full-text search engine (Whoosh) for keyword retrieval and a sentence-embedding-based vector database (ChromaDB) for vector retrieval. A total of 100 test queries were constructed, with 10 queries for each of the 10 main classes, and both the proposed system and the “Ask a Librarian” service were invoked under identical conditions to collect the top 10 results and response times. The results showed that the proposed system achieved a mean response time of 0.21 seconds and a 100% search success rate, demonstrating stable retrieval performance, whereas the “Ask a Librarian” service recorded a mean response time of 13.12 seconds and an 81% search success rate. This study experimentally confirmed that the hybrid search and Retrieval-Augmented Generation (RAG) approach is more effective than the existing method in terms of search success rate and retrieval stability, and suggests future research directions including the application of Korean-language-specific embedding models and the expansion of relevance-based evaluation frameworks.

Journal Copyright Policy

BY-NC-ND 아이콘

This license enables reusers to copy and distribute the material in any medium or format in unadapted form only, for noncommercial purposes only, and only if attribution is given to the creator.

Citation status

* References for papers published after 2025 are currently being built.