Retrieval-Based Methods

Architecture pattern combining information retrieval components with neural NLP models. Instead of relying solely on parametric knowledge in model weights, these methods explicitly retrieve relevant documents or evidence at inference time. Common approaches include retrieve-and-rank (retrieving candidates and ranking them) and retrieval-augmented generation (retrieving passages to condition generation).

Key papers

• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks — foundational work on augmenting generation with retrieval

Related topics

• Information Retrieval (technical foundation)
• Knowledge-Intensive NLP (motivation)
• Question Answering Systems (primary application domain)
• Neural language models (generation component)