BROWSE
Detail
Conference Paper
ArtMISE: Multi-Intent Search and Exploration for Art Collections
Cited - time in 
Share 
Share
- Authors
- Jehak Ha, Yewon Min, Jae Woo Kim, Jin-Seo Kim, Seung-Hyun Moon, Soonchul Jung
- Issue Date
- 2026-10
- Citation
- International Conference on Information and Communication Technology Convergence (ICTC) 2026, pp.1-4
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- Art collection exploration involves heterogeneous query intents, including visual similarity search, contextual interpretation, record lookup, filtering, comparison, and aggregation. Existing keyword-based collection interfaces and text-oriented conversational retrieval systems typically address only subsets of these requirements. We present ArtMISE (Multi-Intent Search and Exploration), a collection-grounded system that integrates image similarity search, document-based retrieval-augmented generation (RAG), and structured metadata querying within a unified conversational interface. Institution-assigned artwork identifiers connect results across these mechanisms to corresponding records, enabling artwork-level traceability across searches and follow-up questions. We quantitatively evaluate the document RAG and structured metadata pathways using 500 natural-language questions over 11,760 artworks from the National Museum of Modern and Contemporary Art, Korea. Under automated LLM-based judging, ArtMISE achieved 92.40% question-level correctness, compared with 39.20% for the Document RAG baseline. The largest differences were observed in collection-level tasks involving aggregation, ranking, and comparison, whereas the two approaches showed similar performance on exact single-record lookup. Within the scope of this automated benchmark, the results indicate that structured metadata querying can complement document RAG for collection-level queries requiring exact or exhaustive computation.
- Keyword
- Artwork Retrieval, Digital Art Collections, Image Similarity Search, Structured Metadata Querying, Retrieval-Augmented Generation, Conversational Search
- KSP Keywords
- Contemporary art, Conversational Search, Digital art, Follow-up, National museum, conversational interface, image similarity search, retrieval system, visual similarity