A Chinese Tourism Service Defect Quintuple Dataset for Large Language Models

Based on 23,697 real negative tourist reviews, the dataset contains 78,429 fine-grained features annotated via a human-in-the-loop approach.

It pioneers a novel quintuple annotation scheme comprising "service dimension, specific entity, defect manifestation, severity, and textual evidence." Divided into a silver-standard training set (73,810 instances) for instruction tuning and a gold-standard test set (4,619 instances) for precise evaluation, this dataset fills the gap in fine-grained instruction tuning data for the Chinese tourism domain. Furthermore, it provides essential data support and a benchmark platform for developing intelligent and interpretable tourism quality monitoring systems.