🤖 AI Summary
Existing dynamic RAG approaches exhibit critical blind spots in retrieval triggering timing and content validation, limiting their ability to mitigate LLM hallucination. To address this, we propose a dual-module framework integrating adaptive cognitive detection and context-aware retrieval optimization. Our method introduces the first cognitive-driven, representation-based dynamic retrieval decision mechanism—leveraging learnable retrieval gating, semantic consistency assessment, and context-aware re-ranking—to jointly optimize *when* and *what* to retrieve. Departing from static or heuristic triggering paradigms, it enables lightweight, real-time retrieval decisions grounded in the LLM’s internal representations. Evaluated across diverse multi-task benchmarks, our approach significantly outperforms state-of-the-art dynamic RAG methods: hallucination rates decrease by 23.6% on average, while answer accuracy and factual consistency both improve substantially.
📝 Abstract
Dynamic Retrieval-augmented Generation (RAG) has shown great success in mitigating hallucinations in large language models (LLMs) during generation. However, existing dynamic RAG methods face significant limitations in two key aspects: 1) Lack of an effective mechanism to control retrieval triggers, and 2) Lack of effective scrutiny of retrieval content. To address these limitations, we propose an innovative dynamic RAG method, DioR (Adaptive Cognitive Detection and Contextual Retrieval Optimization), which consists of two main components: adaptive cognitive detection and contextual retrieval optimization, specifically designed to determine when retrieval is needed and what to retrieve for LLMs is useful. Experimental results demonstrate that DioR achieves superior performance on all tasks, demonstrating the effectiveness of our work.