Reranking
A second-stage retrieval step that re-scores initially retrieved documents using a more powerful model to improve the relevance of the final results.
Initial retrieval (vector search + BM25) is optimized for speed and recall — it retrieves a broad set of potentially relevant passages. Reranking applies a cross-encoder model that jointly processes the query and each candidate passage to produce a relevance score that captures more nuanced semantic relationships than the bi-encoder approach used for initial retrieval. The top-k passages after reranking are then passed to the LLM for answer generation.
Reranking improves answer quality significantly for complex, multi-part questions where the initial retrieval may surface passages that are topically related but not specifically responsive to the query. Cross-encoder models like Cohere Rerank, BGE Reranker, and ColBERT achieve substantially higher precision than bi-encoder retrieval alone, at the cost of higher latency and API fees. For professional document intelligence applications in legal and compliance contexts — where missing a critical passage can have real consequences — the quality improvement typically justifies the additional cost.
More ai/ml Terms
Retrieval-Augmented Generation (RAG)
An AI architecture that combines information retrieval with text generation to produce answers grounded in source documents.
Vector Embedding
A numerical representation of text as a high-dimensional vector, enabling semantic similarity comparisons between passages.
BM25
A probabilistic keyword-ranking algorithm that scores documents by term frequency and inverse document frequency.
Chunking
The process of splitting large documents into smaller, overlapping segments optimized for retrieval and embedding.
Hallucination
When an AI model generates plausible-sounding but factually incorrect or fabricated information.
Large Language Model (LLM)
A neural network trained on massive text corpora that can understand and generate human language.
Analyze Documents Related to Reranking
Upload any document and get AI-powered analysis with verifiable citations.
Start Free