Reducing Semantic Noise in Historical RAG through Knowledge-Graph Constraints

Key Points

• This paper aims to compare semantic vector-based retrieval with knowledge-graph constrained retrieval for historical knowledge.
• Conducted a comparative study of SVR and KGC retrieval methods.
• Integrated relational constraints from Wikidata and applied spatiotemporal filtering.
• Analyzed evidence retrieval precision in historical queries.
• Knowledge-graph constrained retrieval demonstrated a significant reduction in semantic noise.
• Improved precision for historical evidence retrieval was observed with ontology-guided methods.
• The study confirms the relevance of structural constraints in retrieval systems for Digital Humanities.

Abstract