The Evolution of Retrieval Systems: From BM25 to Agentic Retrieval

Retrieval in modern search and recommendation systems has evolved significantly, moving beyond simple index lookups. This evolution spans from Lexical and Behavioral approaches to Learned Sparse, Dense, Hybrid, and Multimodal methods. The latest advancements include Generative and Agentic Retrieval, indicating a continuous progression in how information is retrieved and processed. The discussion invites further exploration into specific retrieval approaches or production tradeoffs.

This report uniquely traces the evolution of retrieval systems from BM25 to Agentic Retrieval, unlike other discussions that focus on individual methods.