Generated Query Expansion Still Helps Strong Sparse Retrieval: A Controlled Study with SPLADE-v3
What it is
This paper investigates whether generated query expansion improves retrieval performance when using the strong sparse retriever SPLADE-v3. It compares four expansion formats (term lists, pseudo-documents, etc.) across three scientific datasets, finding that adding generated vocabulary consistently improves nDCG@10 scores, provided the original query remains strongly represented in the search mixture.
The problem it solves
Scientific queries are often brief and lack the specialized vocabulary used in relevant papers, leading to retrieval mismatches that shrink in value as retrievers become stronger.
What you could use it for
- Improving scientific literature search accuracy
- Enhancing sparse retrieval systems for domain-specific queries
- Designing query expansion strategies for LLM-augmented search engines
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.