Abstract

As generative AI has popularized natural language–based conversational interfaces, chatbot services have gained attention as an intelligent alternative to single-turn, keyword-based searches in archival information services. This study implements an intelligent archival information service chatbot Proof of Concept (PoC) based on a previously established question-type classification framework and examines its applicability in archival institutions. The chatbot was developed with five response stabilization rules. Performance was evaluated through quantitative assessment using 50 test queries and qualitative focus group interviews (FGIs) with 3 records management professionals. Results showed 96% accuracy in question-type classification and 84% in sentence structure appropriateness. Qualitative findings indicated suitability for handling repetitive inquiries in public institutions and potential use as a preliminary guidance tool prior to filing an information disclosure request. Based on dual evaluations, four improvement directions were identified: question-type refinement, linguistic rule enhancement, complex sentence handling advancement, and task-data integration expansion. This study presents an empirical model for archival chatbot development by implementing a question-type–based response structure in an actual system environment and validating its functionality and performance through dual quantitative and qualitative evaluations.