Scholarlink

<p>Large language models (LLMs) have shown exceptional capabilities in Natural Language Processing (NLP) tasks. However, their tendency to generate inaccurate or fabricated information (commonly referred to as hallucinations) poses serious challenges to reliability and user trust. Despite growing research, comprehensive reviews on LLM hallucinations are scarce, creating a critical gap in the literature. To address this, we conduct a systematic literature review (SLR) that offers a more focused and detailed analysis of LLM hallucinations than prior surveys. The SLR examined research published between 2020 and 2025, narrowing hundreds of initial papers down to 253 primary studies (PS) through a rigorous multi-stage screening process. The findings provide valuable insights into state-of-the-art methods for addressing hallucinations, focusing on affected NLP tasks and proposed mitigation strategies. This SLR offers a detailed overview of NLP tasks in which hallucinations are investigated. It stands out by comprehensively analysing techniques, datasets, and evaluation metrics used in experimenting with hallucinations. The findings of this SLR provide a roadmap for researchers and practitioners to enhance the reliability of LLMs and guide the development of robust, context-sensitive hallucination solutions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The rise of hallucination in large language models: systematic reviews, performance analysis and challenges

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Shamsu Abdullahi,
  • Kamaluddeen Usman Danyaro,
  • Haruna Chiroma

摘要

Large language models (LLMs) have shown exceptional capabilities in Natural Language Processing (NLP) tasks. However, their tendency to generate inaccurate or fabricated information (commonly referred to as hallucinations) poses serious challenges to reliability and user trust. Despite growing research, comprehensive reviews on LLM hallucinations are scarce, creating a critical gap in the literature. To address this, we conduct a systematic literature review (SLR) that offers a more focused and detailed analysis of LLM hallucinations than prior surveys. The SLR examined research published between 2020 and 2025, narrowing hundreds of initial papers down to 253 primary studies (PS) through a rigorous multi-stage screening process. The findings provide valuable insights into state-of-the-art methods for addressing hallucinations, focusing on affected NLP tasks and proposed mitigation strategies. This SLR offers a detailed overview of NLP tasks in which hallucinations are investigated. It stands out by comprehensively analysing techniques, datasets, and evaluation metrics used in experimenting with hallucinations. The findings of this SLR provide a roadmap for researchers and practitioners to enhance the reliability of LLMs and guide the development of robust, context-sensitive hallucination solutions.