RARe: Retrieval Augmented Retrieval with In-Context Examples
Topics: AI (Deep Learning), LLMO, Retrieval Augmented Generation (RAG)
This paper introduces RARe, a novel approach that enhances retrieval models by incorporating in-context examples during both training and inference. The method modifies query formats by adding semantically similar query-document pairs as examples, leading to performance improvements across various retrieval tasks. Testing showed gains of up to +2.72% nDCG on reasoning-oriented retrieval benchmarks.