GRADE: Generating multi-hop QA and fine-gRAined Difficulty matrix for RAG Evaluation

2025-08-23

AI Summary

Existing RAG evaluation benchmarks overlook the interaction between retrieval difficulty and reasoning depth, failing to capture real-world challenges in multi-hop reasoning and structural complexity. To address this, we propose GRADE—a novel framework that decouples retrieval difficulty from reasoning depth, formalizing a two-dimensional difficulty matrix parameterized by semantic distance and reasoning depth. GRADE synthesizes controllable-difficulty multi-hop QA datasets via knowledge graph extraction, semantic clustering-based completion, and multi-hop path generation. It further introduces dual-dimensional difficulty quantification—comprising generative and retrieval-oriented metrics—to enable fine-grained, interpretable performance diagnostics. Experiments demonstrate that our difficulty metrics strongly correlate with system error rates and significantly enhance cross-domain and cross-model discriminability and root-cause analysis capability for RAG systems.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Retrieval-Augmented Generation (RAG) systems are widely adopted in knowledge-intensive NLP tasks, but current evaluations often overlook the structural complexity and multi-step reasoning required in real-world scenarios. These benchmarks overlook key factors such as the interaction between retrieval difficulty and reasoning depth. To address this gap, we propose extsc{GRADE}, a novel evaluation framework that models task difficulty along two orthogonal dimensions: (1) reasoning depth, defined by the number of inference steps (hops), and (2) semantic distance between the query and its supporting evidence. We construct a synthetic multi-hop QA dataset from factual news articles by extracting knowledge graphs and augmenting them through semantic clustering to recover missing links, allowing us to generate diverse and difficulty-controlled queries. Central to our framework is a 2D difficulty matrix that combines generator-side and retriever-side difficulty. Experiments across multiple domains and models show that error rates strongly correlate with our difficulty measures, validating their diagnostic utility. extsc{GRADE} enables fine-grained analysis of RAG performance and provides a scalable foundation for evaluating and improving multi-hop reasoning in real-world applications.

Problem

• Evaluating RAG systems lacking multi-step reasoning complexity metrics
• Overlooking interaction between retrieval difficulty and reasoning depth
• Need fine-grained difficulty analysis for real-world RAG applications

Innovation

• Generating multi-hop QA dataset via knowledge graphs
• Creating 2D difficulty matrix combining retrieval and reasoning
• Using semantic clustering to recover missing knowledge links