Systematic Review Dataset on Factuality and Faithfulness in LLM-Based Multi-Source News Summarization
Description
This dataset contains the structured data extracted from studies included in a systematic literature review on factuality and faithfulness in Large Language Model (LLM)-based multi-source news summarization. The dataset was developed to support the analysis of research trends, methodological approaches, datasets and experimental settings, evaluation methods, factuality and faithfulness errors, and research gaps in the existing literature. The dataset includes information on the characteristics of the reviewed studies, summarization methods and approaches, datasets and data sources, evaluation metrics, reported error types, and research gaps and future research directions. Particular attention is given to challenges specific to multi-source summarization, including source disagreement, source conflict, cross-source grounding, unsupported claims, hallucination, and source attribution. The dataset is intended to support systematic analysis, reproducibility, and further research on factuality and faithfulness evaluation in LLM-based news summarization. It may also serve as a reference for researchers developing multi-source news summarization benchmarks, evaluation frameworks, and source-aware LLM systems.