Doc2Query++: Topic-Coverage based Document Expansion and its Application to Dense Retrieval via Dual-Index Fusion

TL;DR

Doc2Query++ tackles vocabulary mismatch by guiding document expansion with topic coverage inferred via unsupervised topic modeling, merging topic-derived and document-derived keywords, and generating multiple topic-aware queries. It also addresses dense retrieval noise through Dual-Index Fusion, which separately encodes original text and generated queries and fuses their signals at inference with a tunable weight. Across BEIR subsets, Doc2Query++ yields consistent gains in MAP, nDCG@10, and Recall@100 for both sparse and dense retrieval, demonstrating strong cross-domain generalization. The work highlights the importance of structured, coverage-driven expansion and offers a practical, domain-agnostic approach with potential for adaptive topic control and multilingual extensions.

Abstract

Document expansion (DE) via query generation tackles vocabulary mismatch in sparse retrieval, yet faces limitations: uncontrolled generation producing hallucinated or redundant queries with low diversity; poor generalization from in-domain training (e.g., MS MARCO) to out-of-domain data like BEIR; and noise from concatenation harming dense retrieval. While Large Language Models (LLMs) enable cross-domain query generation, basic prompting lacks control, and taxonomy-based methods rely on domain-specific structures, limiting applicability. To address these challenges, we introduce Doc2Query++, a DE framework that structures query generation by first inferring a document's latent topics via unsupervised topic modeling for cross-domain applicability, then using hybrid keyword selection to create a diverse and relevant keyword set per document. This guides LLM not only to leverage keywords, which ensure comprehensive topic representation, but also to reduce redundancy through diverse, relevant terms. To prevent noise from query appending in dense retrieval, we propose Dual-Index Fusion strategy that isolates text and query signals, boosting performance in dense settings. Extensive experiments show Doc2Query++ significantly outperforms state-of-the-art baselines, achieving substantial gains in MAP, nDCG@10 and Recall@100 across diverse datasets on both sparse and dense retrieval.

Figures (4)

• Figure 1: Overview of the Doc2Query$++$ framework. The model infers latent topics via BERTopic, extracts topic- and document-level keywords, and uses them to guide LLM-based query generation. Generated queries are appended for sparse retrieval and fused through Dual-Index Fusion for dense retrieval.
• Figure 2: Pearson Correlation between topic recall and Retrieval Gain
• Figure 3: Retrieval performance (nDCG@10 and Recall@100) on FiQA-2018 with varying numbers of generated queries using Doc2Query. Performance peaks around 100 queries and degrades afterward, while our coverage-aware method (30 queries) outperforms even the best Doc2Query setting.
• Figure 4: Impact of fusion weight $\alpha$ on retrieval performance across different document expansion methods.