Cross-Encoder Reranking
A reranking technique in two-stage retrieval that uses a cross-encoder to jointly encode the query and a passage for fine-grained relevance scoring. It captures finer-grained semantics but carries high computational cost, making it suitable for reranking a small number of candidate passages.
Timeline (last 90 days)
The study's retrieval pipeline fused dense vectors, BM25, and cross-encoder reranking, and systematically compared chunk sizes from 128 to 1024 tokens, overlap rates from 0 to 20%, and four retriever configurations.
Cross-encoder reranking was confirmed in the study to yield the highest retrieval scores, consistent with conclusions previously observed in the general domain.
All Facts (2)
The study's retrieval pipeline fused dense vectors, BM25, and cross-encoder reranking, and systematically compared chunk sizes from 128 to 1024 tokens, overlap rates from 0 to 20%, and four retriever configurations.
50%UnverifiedCross-encoder reranking was confirmed in the study to yield the highest retrieval scores, consistent with conclusions previously observed in the general domain.
50%