嵌入
表示一段文本含义的稠密数值向量。
概览
Embedding 将一段文本映射为一个固定长度的数值向量，其位置安排使得含义相近的文本彼此靠近。它解决了关键词匹配无法判断两个措辞不同的句子表达相同含义这一问题。它之所以存在，是因为基于向量的最近邻搜索速度快且可以建立索引，从而使语义检索在大规模场景下变得可行。Embedding 是检索得以运作的根本：查询被嵌入后，最接近的已存储文档会作为候选结果被返回。
技术图谱
嵌入 如何与技术栈的其他部分相连。
使用与依赖
- Recommendation application Applications Recommendation based on 嵌入 High ✓ source
被用于
- AI Search application Applications 嵌入 used by AI Search High ✓ source
- Agent Memory technique AI Agents 嵌入 used by Agent Memory High ✓ source
- Chroma database Vector Databases 嵌入 used by Chroma High ✓ source
- Elasticsearch database Search Engines 嵌入 used by Elasticsearch High ✓ source
- GraphRAG technique Techniques 嵌入 used by GraphRAG High ✓ source
- HyDE technique Techniques 嵌入 used by HyDE High ✓ source
- Hybrid Search technique Techniques 嵌入 used by Hybrid Search High ✓ source
- LanceDB database Vector Databases 嵌入 used by LanceDB High ✓ source
- Milvus database Vector Databases 嵌入 used by Milvus High ✓ source
- OpenSearch database Search Engines 嵌入 used by OpenSearch High ✓ source
- Pinecone database Vector Databases 嵌入 used by Pinecone High ✓ source
- Qdrant database Vector Databases 嵌入 used by Qdrant High ✓ source
- RAG technique Techniques 嵌入 used by RAG High ✓ source
- Redis database Databases 嵌入 used by Redis High ✓ source
- Typesense database Search Engines 嵌入 used by Typesense High ✓ source
- Vespa database Search Engines 嵌入 used by Vespa High ✓ source
- Weaviate database Vector Databases 嵌入 used by Weaviate High ✓ source
- pgvector library Vector Databases 嵌入 used by pgvector High ✓ source
应用于
- BEIR benchmark Benchmarks BEIR used for 嵌入 High ✓ source
- Bi-Encoder concept Techniques Bi-Encoder used for 嵌入 High ✓ source
- Chunking technique Techniques Chunking used for 嵌入 High ✓ source
- Contrastive Learning technique Algorithms Contrastive Learning used for 嵌入 High ✓ source
- HNSW algorithm Algorithms HNSW used for 嵌入 High ✓ source
- MS MARCO dataset Datasets MS MARCO used for 嵌入 High ✓ source
- MTEB benchmark Benchmarks MTEB used for 嵌入 High ✓ source
- SQLite database Databases SQLite used for 嵌入 High ✓ source
- Sentence Transformers framework Frameworks Sentence Transformers used for 嵌入 High ✓ source
- Tokenizer tool Infrastructure Tokenizer used for 嵌入 High ✓ source