· Xiaojing Yang · Retrieval & Knowledge Systems · 1 min read
Retrieval Is Not Just Vector Search
A retrieval-systems note on sparse, dense, hybrid, reranking, graph expansion, long documents, and evidence grounding.
中文导读
很多 RAG 问题不是生成模型的问题，而是 retrieval 没有找到、没有排序好、没有保留证据结构，或者没有正确处理长文档。
Working outline
- Retrieval before generation
- Sparse retrieval
- Dense retrieval
- Hybrid retrieval and reranking
- Long documents and structure
- Evidence grounding and evaluation
Reference materials to digest
- Benchmarking Large Language Models in Retrieval-Augmented Generation
- Hugging Face Course: semantic search with FAISS
- Google ML guides
My angle
This track can cover RAG, annual reports, financial QA, GraphRAG, information retrieval, and broader knowledge-system design.