ReCite: Agentic Reasoning for Faithful Citation
ReCite tries to fix real-paper, wrong-support citations by checking whether evidence actually backs the claim.
The paper argues that citation recommendation still misattributes sources when systems lean on semantic similarity alone. ReCite separates the task into location perception, intent-aware query planning, and reflective verification. Its agent can reject retrieved candidates that do not logically support the claim and loop back for correction. The authors report that the lightweight framework beats larger generative models on strict citation accuracy. ArXiv · AI/CL/LG's note
The paper argues that citation recommendation still misattributes sources when systems lean on semantic similarity alone. ReCite separates the task into location perception, intent-aware query planning, and reflective verification. Its agent can reject retrieved candidates that do not logically support the claim and loop back for correction. The authors report that the lightweight framework beats larger generative models on strict citation accuracy. ArXiv · AI/CL/LG's note
score 4