Bi-Encoder
A Bi-Encoder is a retrieval architecture that encodes queries and documents independently into separate vectors and computes similarity via dot product. It supports offline pre-computation of document vectors and offers very high retrieval speed, making it suitable for coarse-recall from large-scale corpora.
Core Facts
A dense dual-encoder consists of two independent neural networks—a query encoder and a document encoder—that map queries and documents into the same vector space.
Cross-encoders are more accurate than bi-encoders because they concatenate [query, doc] and feed it into a Transformer, letting the attention mechanism compare directly at the word level, but they cannot pre-build an index and can only be used for fine ranking.
Rerankers are typically implemented using cross-encoders, which concatenate the query and document and feed them jointly into a Transformer model, achieving significantly higher ranking precision than the initial retrieval stage
Timeline (last 90 days)
Rerank models are usually cross-encoders, with higher accuracy than bi-encoders but high computational cost, suitable for finely ranking a small number of candidates such as the top-50.
The hybrid RAG pipeline of modern generative search engines typically first recalls 20 candidates with a bi-encoder, then re-ranks them precisely with a cross-encoder, while graph matching provides deterministic triples.
A cross-encoder concatenates the query and document and feeds them together into a Transformer, which can capture deeper relevance signals, but document representations cannot be precomputed, and each query-document pair requires one forward pass, so it is only suitable for reranking a small number of candidates.
Cross-encoders are more accurate than bi-encoders because they concatenate [query, doc] and feed it into a Transformer, letting the attention mechanism compare directly at the word level, but they cannot pre-build an index and can only be used for fine ranking.
RATED employs a frozen dual-encoder at the semantic matching layer, fine-tuned via contrastive learning.
A cross-encoder concatenates the query and document into a single sequence for joint encoding via full attention mechanisms, offering high accuracy but with O(n) computational complexity, making it suitable only for re-ranking the top-K results (typically 20–100 items) from initial retrieval.
Rerankers are typically implemented using cross-encoders, which concatenate the query and document and feed them jointly into a Transformer model, achieving significantly higher ranking precision than the initial retrieval stage
All Facts (9)
A dense dual-encoder consists of two independent neural networks—a query encoder and a document encoder—that map queries and documents into the same vector space.
75%VerifiedCross-encoders are more accurate than bi-encoders because they concatenate [query, doc] and feed it into a Transformer, letting the attention mechanism compare directly at the word level, but they cannot pre-build an index and can only be used for fine ranking.
70%VerifiedRerankers are typically implemented using cross-encoders, which concatenate the query and document and feed them jointly into a Transformer model, achieving significantly higher ranking precision than the initial retrieval stage
65%VerifiedA Cross-Encoder concatenates a query with each candidate document and feeds them jointly into the same model, capturing word-level interactions through the attention mechanism; its ranking precision is superior to vector similarity, but latency grows linearly with the size of the candidate set.
65%UnverifiedRerank models are usually cross-encoders, with higher accuracy than bi-encoders but high computational cost, suitable for finely ranking a small number of candidates such as the top-50.
50%UnverifiedThe hybrid RAG pipeline of modern generative search engines typically first recalls 20 candidates with a bi-encoder, then re-ranks them precisely with a cross-encoder, while graph matching provides deterministic triples.
50%UnverifiedA cross-encoder concatenates the query and document and feeds them together into a Transformer, which can capture deeper relevance signals, but document representations cannot be precomputed, and each query-document pair requires one forward pass, so it is only suitable for reranking a small number of candidates.
50%UnverifiedRATED employs a frozen dual-encoder at the semantic matching layer, fine-tuned via contrastive learning.
50%UnverifiedA cross-encoder concatenates the query and document into a single sequence for joint encoding via full attention mechanisms, offering high accuracy but with O(n) computational complexity, making it suitable only for re-ranking the top-K results (typically 20–100 items) from initial retrieval.
50%