Quality Excellence Text Dataset

The deposited materials include: (1) results across validation settings, seeds, and folds; (2) sentence-level attention and SHAP outputs used for repository construction; (3) topic-assignment outputs underlying the BERTopic analyses; and (4) codes for preprocessing, model training, explanation extraction, and topic modelling. Raw annual reports were obtained from publicly available corporate disclosures of Chinese A-share listed firms. Because the original reports are third-party source documents, readers should access the original annual reports from the corresponding public disclosure platforms.