Sparse Retrieval

• What is the difference between sparse and dense retrieval?

Sparse retrieval methods, like TF-IDF or BM25, represent text as high-dimensional vectors where most dimensions are zero, encoding the presence or absence of specific words.

• Their main goal is to do Keyword Search or something between keyword and semantic search i.e. sparse retrieval

Source: ChatGPT¹

• Traditional Sparse Retrieval
• Learned Sparse Retrieval

• Keyword search is all you need: Achieving RAG-Level Performance without vector databases using agentic tool use
  • Paper by AWS

ChatGPT¹ (Note that BGE-M3 is not in the table):