Mitigating Hallucination in Large Language Models (LLMs): An Application-Oriented Survey on RAG, Reasoning, and Agentic Systems

TL;DR

The paper tackles the persistent challenge of hallucinations in large language models by adopting an application-oriented, capability-enhancement lens. It systematically analyzes how Retrieval-Augmented Generation (RAG), reasoning enhancements (including Chain-of-Thought, tool-augmented, and symbolic reasoning), and their integration in Agentic Systems mitigate both knowledge-based and logic-based hallucinations. A unified framework is proposed, with a detailed taxonomy, a thorough review of RAG pipelines (pre-retrieval, retrieval, post-retrieval), precise and broad retrieval strategies (graph-Augmented RAG, KG-RAG, web/multi-modal retrieval), and reasoning paradigms, complemented by representative benchmarks spanning knowledge, logic, and composite hallucinations. The work also discusses practical applications across healthcare, law, finance, and education, and outlines challenges and future directions, emphasizing efficiency, cross-domain generalization, and multi-modal reliability. Overall, it provides a comprehensive, integrated reference for developing reliable, interpretable, and scalable LLMs through capability enhancement rather than purely model-centric suppression of hallucinations.

Abstract

Hallucination remains one of the key obstacles to the reliable deployment of large language models (LLMs), particularly in real-world applications. Among various mitigation strategies, Retrieval-Augmented Generation (RAG) and reasoning enhancement have emerged as two of the most effective and widely adopted approaches, marking a shift from merely suppressing hallucinations to balancing creativity and reliability. However, their synergistic potential and underlying mechanisms for hallucination mitigation have not yet been systematically examined. This survey adopts an application-oriented perspective of capability enhancement to analyze how RAG, reasoning enhancement, and their integration in Agentic Systems mitigate hallucinations. We propose a taxonomy distinguishing knowledge-based and logic-based hallucinations, systematically examine how RAG and reasoning address each, and present a unified framework supported by real-world applications, evaluations, and benchmarks.

Figures (7)

• Figure 1: Two types of hallucinations in LLM responses
• Figure 2: Overview of hallucination mitigation and evaluation methods with representative models and benchmarks
• Figure 3: Overview of the RAG pipeline
• Figure 4: Illustrative examples of four methods for enhancing intent understanding
• Figure 5: Demonstration of a typical broad retrieval process