Upcycling Candidate Tokens of Large Language Models for Query Expansion
By: Jinseok Kim , Sukmin Cho , Soyeong Jeong and more
Potential Business Impact:
Finds better search results with less computer power.
Query Expansion (QE) improves retrieval performance by enriching queries with related terms. Recently, Large Language Models (LLMs) have been used for QE, but existing methods face a trade-off: generating diverse terms boosts performance but increases computational cost. To address this challenge, we propose Candidate Token Query Expansion (CTQE), which extracts diverse and relevant terms from a single LLM decoding pass by leveraging unselected candidate tokens. These tokens, though not part of the final output, are conditioned on the full query and capture useful information. By aggregating them, CTQE achieves both relevance and diversity without extra inference, reducing overhead and latency. Experiments show that CTQE delivers strong retrieval performance with significantly lower cost, outperforming or comparable to more expensive methods. Code is available at: https://github.com/bluejeans8/CTQE
Similar Papers
Query Expansion in the Age of Pre-trained and Large Language Models: A Comprehensive Survey
Information Retrieval
Helps computers find better answers to your questions.
Generative Query Expansion with Multilingual LLMs for Cross-Lingual Information Retrieval
Information Retrieval
Helps computers find information in different languages.
ThinkQE: Query Expansion via an Evolving Thinking Process
Information Retrieval
Finds better search results by thinking more.