CARA: Causally Guided Retrieval Agent for Question Answering
What it is
CARA is a training-free retrieval agent designed for text-rich knowledge graphs that uses causal relevance metrics to guide search. It evaluates evidence based on consequentiality and essentiality to determine if adding or removing information affects answerability, rather than relying solely on textual similarity.
The problem it solves
Existing agentic retrievers rely primarily on textual similarity, which fails to identify evidence that is causally relevant to answering a question rather than just textually matching it.
What you could use it for
- Question answering over text-rich knowledge graphs
- Multi-hop reasoning in retrieval-augmented generation
- Improving faithfulness in LLM-based QA systems
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.