Comparative evaluation demonstrates superior query accuracy and transparency in Chinese traditional music retrieval, indicating the value of linking knowledge graphs with language models.
A persistent challenge in Music Information Retrieval (MIR) lies in the underdevelopment of the metadata tier, which remains overshadowed by the fields of audio, score, and image-based music analysis. This paper argues that music metadata, when formalized through linked data and ontology, can serve as a powerful bridge between musicological theory and computational AI. We introduce the Chinese Traditional Music (CTM) Linked Data Knowledge Graph platform, a public repository of controlled musicological vocabularies—including traditional music species, instruments, and ethnic classifications—organized through a custom OWL ontology. Building on this foundation, we present a semantic-metadata-based AI Question-Answering (QA) agent (Enhanced Version 2.0) that leverages large language models (LLMs) to convert natural language questions into SPARQL queries. Our approach addresses two core challenges: (1) the ambiguity of natural language, through multi-round dialogue and graph-structure-based reasoning; and (2) the scalability of ontology-driven retrieval, through an ontology-subgraph extraction method. We demonstrate the agent's efficacy through case studies that span T-box/A-box hybrid queries, ontological advanced semantics, and generation-augmented retrieval (GAR) for fine-grained musicological inquiry. Comparative tests between LLM-only and KG-augmented responses reveal the superior accuracy and transparency of the integrated approach, particularly in handling polysemic Chinese musicological terms and geographically distributed music cultures. We conclude by proposing a prospective music digital humanities sub-discipline cluster, with knowledge graphs as the core and metadata ontology and network analysis as the two supporting pillars—offering a vision for bridging musicological theory and AI through semantic formalization.
No takes yet. Share an insight, caveat, or question.
Shaojun et al. (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: