嵌入
定义
嵌入是文本（或其他模态）的密集向量表示。 Similar content maps to nearby vectors, enabling semantic search and 检索 in RAG.
它们是 the bridge between raw text and vector databases: both documents and queries are embedded so similarity can be computed as vector distance (例如 cosine). Choice of embedding model affects RAG 检索 quality and cost; see semantic search for the broader use of embeddings in search.
工作原理
Text (a sentence, paragraph, or chunk) 被输入到一个 encoder (例如 OpenAI embeddings, Cohere, or open-source sentence-transformers). The encoder outputs a fixed-size vector (例如 768 or 1536 dimensions). Training uses contrastive or similar objectives so that semantically related texts get nearby vectors; similarity at query time is usually cosine or dot product. Models can be multilingual or domain-specific. For RAG, use the same encoder for documents and queries so distances are meaningful.
应用场景
Embeddings 是原始文本和相似性搜索之间的桥梁; 这里的质量决定了检索召回率和 RAG 准确性.
- Converting documents and queries to vectors for RAG 检索
- Semantic similarity and clustering (例如 duplicate detection)
- Cross-lingual and multimodal 检索 when using matching encoders