No Stupid Questions: An Analysis of Question Query Generation for Citation Recommendation

2025-06-09

AI Summary

Existing citation recommendation methods rely primarily on article content and metadata, lacking mechanisms to actively elicit novel insights from文献 excerpts through interrogative reasoning. Method: This paper introduces the first systematic use of generative questions as retrieval queries for citation recommendation. Leveraging GPT-4o-mini to emulate scholarly curiosity, it automatically generates insightful, context-aware questions from document excerpts. We propose MMR-RBO—a hybrid ranking strategy integrating Rank-Biased Overlap (RBO) and Maximal Marginal Relevance (MMR)—to efficiently select high-quality, diverse, and informationally complementary question subsets. Contribution/Results: Empirical evaluation demonstrates that question-based queries significantly outperform keyword-based queries (extracted by the same model) in specific scenarios. Crucially, results validate the “no stupid questions” hypothesis: distinct questions consistently retrieve unique and complementary citations, confirming their intrinsic semantic diversity and utility. This work establishes a question-driven paradigm for citation recommendation, advancing beyond conventional content- or metadata-centric approaches.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Natural Language Processing: Question Answering
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Web Mining and Content Analysis: Community question answering
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Existing techniques for citation recommendation are constrained by their adherence to article contents and metadata. We leverage GPT-4o-mini's latent expertise as an inquisitive assistant by instructing it to ask questions which, when answered, could expose new insights about an excerpt from a scientific article. We evaluate the utility of these questions as retrieval queries, measuring their effectiveness in retrieving and ranking masked target documents. In some cases, generated questions ended up being better queries than extractive keyword queries generated by the same model. We additionally propose MMR-RBO, a variation of Maximal Marginal Relevance (MMR) using Rank-Biased Overlap (RBO) to identify which questions will perform competitively with the keyword baseline. As all question queries yield unique result sets, we contend that there are no stupid questions.

Problem

• Improving citation recommendation via generated question queries
• Evaluating question queries against keyword-based retrieval
• Proposing MMR-RBO for competitive question selection

Innovation

• GPT-4o-mini generates insightful questions for retrieval
• MMR-RBO optimizes question selection for performance
• Question queries outperform keyword queries sometimes