Unverified95% confidenceFactTime unknown
Bi-Encoder将问题和文档分别独立编码成向量再计算相似度，Cross-Encoder将问题和候选文档拼接后一起输入模型通过注意力机制充分交互
1
Sources
95%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedA Bi-Encoder encodes queries and documents independently into vectors and computes relevance via cosine similarity or dot product; document vectors can be pre-computed offline and stored in a vector database.82% similarUnverifiedA 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.81% similarUnverified现代Sentence Embedding模型通常采用双塔（Bi-Encoder）结构，查询和文档分别独立编码为向量79% similarUnverifiedCross-Encoders concatenate the query and document into a single input sequence and perform deep interaction modeling through Transformer's full attention mechanism; representative models include Cohere Rerank, BGE-Reranker, and ms-marco-MiniLM.77% similarUnverifiedRetrieval BERT uses a Bi-Encoder (dual-tower model), where two sentences each generate a vector and similarity is then computed; vectors can be precomputed offline, enabling fast online matching.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/11825API
curl https://kongchang.com/api/v1/knowledge/claims/11825MCP
get_claim(id=11825)