This repository serves as a comprehensive survey of factuality in Large Language Models (LLMs), detailing knowledge representation, retrieval augmentation, and domain-specific challenges. It is intended for researchers and practitioners in NLP and AI who need a structured overview of LLM factuality issues, existing solutions, and evaluation benchmarks.

How It Works

The survey categorizes factuality issues into model-level (e.g., knowledge deficit, reasoning errors) and retrieval-level causes (e.g., distraction, misinterpretation). It then explores various enhancement methods, including continual pre-training, supervised fine-tuning, and model editing, often supported by external knowledge sources. The paper also provides an extensive review of relevant datasets and evaluation metrics used to assess LLM factuality across different domains.

Quick Start & Requirements

This repository is a collection of survey information and does not have a direct installation or execution command. The primary resource is the linked arXiv paper for detailed content.

Highlighted Details

• Comprehensive taxonomy of LLM factuality errors and their causes.
• Extensive catalog of LLM factuality evaluation benchmarks and metrics.
• Detailed review of enhancement methods for improving LLM factuality.
• Analysis of domain-specific LLMs and their factuality challenges (e.g., medicine, law, finance).

Maintenance & Community

The repository is associated with the survey paper "Survey on Factuality in Large Language Models: Knowledge, Retrieval and Domain-Specificity." Contributions via pull requests or issues are welcomed to improve the survey content.

Licensing & Compatibility

The repository itself does not specify a license. The survey paper is available on arXiv.

Limitations & Caveats

As a survey repository, it primarily aggregates and organizes information from other research papers. Real-time updates to the arXiv paper may not be reflected here.