Generated Query Expansion Still Helps Strong Sparse Retrieval: A Controlled Study with SPLADE-v3

2026-09-29

AI Summary

This study addresses the performance limitations in scientific retrieval caused by short queries mismatching domain-specific terminology, and investigates the underexplored effectiveness of generative query expansion under strong sparse retrievers. Building upon the SPLADE-v3 model, we systematically evaluate four LLM-generated query expansion formats—including term lists and pseudo-documents—across three datasets. By employing controlled variable analysis to isolate the impact of added content, rigorous evaluation is conducted using nDCG@10 with Holm correction. Our findings demonstrate that generated vocabulary can yield relative improvements of up to 9.47% even when paired with a strong sparse retriever. Furthermore, this work reveals that preserving original query weights is critical for sustaining these gains, and that lexical-level expansion outperforms concept graph structures.

Abstract

Scientific queries are often brief, while relevant papers use specialized vocabulary. Generated query expansion can bridge this mismatch, but earlier work suggests that its value shrinks as the underlying retriever becomes stronger. We test the four generated formats of term lists, a pseudo-document, multiple pseudo-references, and corpus-steered text all together with SPLADE-v3 on NFCorpus, TREC-COVID, and SciDocs. Every condition searches the same frozen document index and follows the same query-side integration rule and 256-dimension budget, isolating the effect of the added content. All twelve method-collection comparisons improve aggregate nDCG@10, with best relative gains of 4.81%, 8.92%, and 9.47%. Eleven remain significant after Holm correction. The gain persists in 103 of 114 interpolation settings, including every setting that assigns at least 30% of the mixture weight to the original query. Shuffled-text and non-contextual lexical-bag controls also remain above baseline in all 24 aggregate comparisons, showing that the added vocabulary carries most of the benefit. A corpus-induced typed concept graph, by contrast, produces no consistent gain, and its relation, depth, validation, random, and gating controls do not rescue it. Generated vocabulary can therefore complement a strong learned sparse retriever, provided that the original query remains strongly represented.

Problem

• query expansion
• sparse retrieval
• scientific search
• vocabulary mismatch
• SPLADE

Innovation

• Query Expansion
• Sparse Retrieval
• SPLADE-v3
• Scientific Information Retrieval
• Controlled Study