CARA: Causal Retrieval Agent for Knowledge Graph QA
October 16, 2026
CARA is a training-free retrieval agent that improves question answering over knowledge graphs by evaluating evidence through consequentiality and essentiality. It uses causal relevance rather than simple textual similarity to guide multi-hop exploration and aggregate evidence.
HOW THIS AFFECTS YOU
●
builderYou can implement this causal logic to improve the accuracy of RAG systems using knowledge graphs.
●
researcherThis offers a new approach to moving beyond similarity-based retrieval in agentic workflows.