Generalized Pseudo-Relevance Feedback

2025-10-29

AI Summary

Traditional pseudo-relevance feedback (PRF) and its vectorized extension (VPRF) rely on strong assumptions: the relevance assumption (top-ranked documents are relevant) and the model-specific assumption (query rewriting must be tailored to a particular retriever). While generative relevance feedback (GRF) alleviates the latter, it remains vulnerable to LLM hallucinations and still inherits the relevance assumption. This paper proposes Generalized Pseudo-Relevance Feedback (GPRF), a model-agnostic, weakly relevance-dependent natural language query rewriting framework. Its core is a utility-driven reinforcement learning training pipeline that leverages LLMs to generate robust rewrites while explicitly suppressing hallucination and reducing sensitivity to top-document quality. Extensive experiments across multiple benchmarks and diverse retrievers demonstrate that GPRF consistently and significantly outperforms PRF, VPRF, and various GRF baselines. To our knowledge, GPRF is the first framework achieving universal, noise-resilient, and model-agnostic query rewriting without requiring retriever-specific customization.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Learning Preferences or Rankings
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Abstract

Query rewriting is a fundamental technique in information retrieval (IR). It typically employs the retrieval result as relevance feedback to refine the query and thereby addresses the vocabulary mismatch between user queries and relevant documents. Traditional pseudo-relevance feedback (PRF) and its vector-based extension (VPRF) improve retrieval performance by leveraging top-retrieved documents as relevance feedback. However, they are constructed based on two major hypotheses: the relevance assumption (top documents are relevant) and the model assumption (rewriting methods need to be designed specifically for particular model architectures). While recent large language models (LLMs)-based generative relevance feedback (GRF) enables model-free query reformulation, it either suffers from severe LLM hallucination or, again, relies on the relevance assumption to guarantee the effectiveness of rewriting quality. To overcome these limitations, we introduce an assumption-relaxed framework: extit{Generalized Pseudo Relevance Feedback} (GPRF), which performs model-free, natural language rewriting based on retrieved documents, not only eliminating the model assumption but also reducing dependence on the relevance assumption. Specifically, we design a utility-oriented training pipeline with reinforcement learning to ensure robustness against noisy feedback. Extensive experiments across multiple benchmarks and retrievers demonstrate that GPRF consistently outperforms strong baselines, establishing it as an effective and generalizable framework for query rewriting.

Problem

• Addresses vocabulary mismatch between queries and documents
• Reduces reliance on relevance and model architecture assumptions
• Provides robust query rewriting using reinforcement learning

Innovation

• Assumption-relaxed framework for model-free query rewriting
• Utility-oriented training with reinforcement learning
• Robustness against noisy feedback documents