IFCMemoryBench gives LLM agents 4,016 prior chat sessions plus live IFC model queries; the best vector-, graph-, or file-based memory system reaches only 32.4% answer accuracy, versus 83.2% when all relevant user messages are shown directly.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
IFCMemoryBench: Evaluating Long-Term Memory of LLM-Based Agents in BIM Information Retrieval
IFCMemoryBench gives LLM agents 4,016 prior chat sessions plus live IFC model queries; the best vector-, graph-, or file-based memory system reaches only 32.4% answer accuracy, versus 83.2% when all relevant user messages are shown directly.