Paper2020
Dense Passage Retrieval for Open-Domain Question Answering
Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis et al. · EMNLP · 2020
Abstract
Replaces keyword search with learned dense embeddings for question answering: two encoders, one for questions and one for passages, trained so that a question lands near the passages that answer it. Trained on relatively few question-passage pairs, it beat the long-standing BM25 baseline by a wide margin on retrieval accuracy.
Why it matters
The retrieval half of RAG, and the reason vector search became the default. Read it with the RAG paper here — this is the component that made that architecture work.
ragretrievalgroundingevals
Read the source
https://arxiv.org/abs/2004.04906