Embedding
Embedding is a representation learning technique that maps discrete objects—such as text, images, or user behaviors—to dense vectors in a low-dimensional continuous vector space. Its core characteristic is that semantically similar objects are positioned close together in the vector space, enabling computers to capture semantic relationships between objects. Embeddings are widely used in natural language processing, recommendation systems, and information retrieval, and serve as a foundational method for deep learning models to handle non-numerical data.
Core Facts
混合检索（Hybrid Search）将向量检索与BM25关键词检索的结果通过RRF（倒数排名融合）等算法合并，兼顾语义理解与精确匹配
RAG（检索增强生成）技术架构将文档切片、向量化、存入向量数据库，查询时先检索相关片段再交给模型生成回答
The working principle of RAG is divided into a retrieval phase (converting the question into a vector to recall relevant content from the knowledge base) and a generation phase (inputting the recalled context together with the question into the model to produce an answer).
Timeline (last 90 days)
The difficulty of RAG lies not in getting the demo running but in retrieval accuracy; chunking granularity, embedding model selection, and recall strategy all affect the final result.
Semantic Memory stores conversation content in a vector database via Embedding and recalls it by semantic similarity at retrieval time; its pipeline is the same as RAG, except that the knowledge source is historical conversation rather than external documents.
The standard practice is to generate one embedding vector per chunk; when a chunk mixes multiple topics, that vector is the average of the semantics of each topic, making it hard for a query targeting a single detail to match precisely.
The RAG retrieval process is: question → rewriting → embedding → top-k → reranking → context → LLM; the model does not perform the search—it is the application code that searches and puts the results into the context; embedding uses a specially trained model to convert text into fixed-length vectors, and cosine similarity measures semantic distance.
Embedding call prices are typically one to two orders of magnitude lower than LLM call prices, which is the premise for semantic caching being economically viable.
Embedding is a technique that maps text, URLs, or page content into high-dimensional vectors and can be used to measure semantic similarity; in the SEO internal linking scenario it can discover linking opportunities where the wording differs but the topic is consistent, but the output still requires human review.
In that to-do app, every task is converted into an embedding vector, so that when searching for "milk" the system can recall semantically similar tasks such as "buy dairy products" and "emotional support yogurt".
Embedding is a technique that maps unstructured data such as text and images into high-dimensional dense vectors.
The project loads the original model files via the Transformers framework to run Embedding on local compute, no longer relying on the platform's one-click conversion.
Embedding models are deep neural networks that map text into a fixed-dimensional dense vector space (such as 768 dimensions or 1536 dimensions); cosine similarity is commonly used to measure semantic closeness, with a value range from -1 to 1.
33 more timeline events
All Facts (20)
混合检索（Hybrid Search）将向量检索与BM25关键词检索的结果通过RRF（倒数排名融合）等算法合并，兼顾语义理解与精确匹配
90%VerifiedRAG（检索增强生成）技术架构将文档切片、向量化、存入向量数据库，查询时先检索相关片段再交给模型生成回答
90%VerifiedThe working principle of RAG is divided into a retrieval phase (converting the question into a vector to recall relevant content from the knowledge base) and a generation phase (inputting the recalled context together with the question into the model to produce an answer).
80%VerifiedSemantic search converts text into high-dimensional vectors and finds the most relevant conclusions by computing cosine similarity, allowing queries phrased in different ways to match semantically equivalent results.
80%VerifiedVector databases such as FAISS, Chroma, and Weaviate build indexes by encoding text as high-dimensional vectors, enabling semantic similarity retrieval in milliseconds.
80%VerifiedRAG（检索增强生成）的核心思想是在大模型生成回答之前，先从外部知识库中检索相关文档片段作为上下文注入Prompt
80%VerifiedVector databases store high-dimensional vectors converted from text by an Embedding model, typically 768 or 1536 dimensions.
80%VerifiedEmbedding models typically map text into high-dimensional vector spaces of 768 or 1536 dimensions.
80%VerifiedEmbedding是将文本转换为高维数值向量的过程，语义相近的文本在向量空间中距离更近
80%Verified向量数据库将文本内容通过嵌入模型（Embedding Model）转化为高维数值向量，检索时通过余弦相似度等算法找出语义最接近的内容
80%VerifiedVoyage AI is known for its high-quality text embedding models, which are widely used in RAG, semantic search, and vectorization scenarios
75%VerifiedEach Transformer layer contains two core sub-modules: multi-head self-attention and a feed-forward neural network.
75%VerifiedRAG（检索增强生成）是一种结合信息检索与生成式大模型的架构范式，模型生成答案前先根据用户查询从外部知识库检索最相关的文档片段，再将其作为上下文注入模型
70%VerifiedProduction-grade RAG systems typically build and fully fuse three retrieval paths: vector retrieval, knowledge graphs, and keyword inverted indexes.
70%VerifiedA complete RAG system typically consists of five stages: document chunking, vector embedding, vector storage, semantic retrieval, and augmented generation.
70%VerifiedLong-term memory retrieval relies on vector database technology at its core, converting text into high-dimensional vectors via embedding models.
70%VerifiedThe value of the private-document-based RAG model is highly dependent on the quality of the user's preliminary materials; if documents are poorly organized, outdated, or inconsistently formatted, retrieval quality will decrease accordingly.
65%VerifiedThe knowledge base relies on RAG (Retrieval-Augmented Generation) technology at its foundation, slicing private documents and converting them into high-dimensional vectors via an Embedding model for storage in a vector database.
65%Unverified索引阶段和查询阶段必须使用同一个Embedding模型，否则向量空间不一致会导致检索完全失效
95%Unverified现代Embedding模型通常基于Transformer架构，通过对比学习在大规模文本对上训练，向量维度通常在768到1536之间
90%