SPLADE

Sparse Lexical and Expansion Model for first-stage ranking. Combines advantages of dense embeddings with inverted index efficiency using explicit sparsity regularization. Influential sparse neural retrieval model. From NAVER Labs Europe.