2026-10-14T21:07:47Z https://www.scidb.cn/oai

10.57760/sciencedb.41824 2026-10-14T14:13:29Z

2026-10-14 A Structured Excel Dataset for International Trade Invoice Information Extraction and Anomaly Detection doi:10.57760/sciencedb.41824 en This article constructs a structured Excel dataset for international trade invoice information extraction and anomaly recognition. On the premise of protecting privacy, provide research samples with diverse formats, complete field hierarchy, and traceable task labeling, including standard invoices, JSON files of invoice recognition results, question and answer labeling, enhanced invoices, and error labeling in various data forms. It can support multiple research tasks such as complex table understanding, fine-tuning of big language models, structured generation, anomaly detection, and intelligent question and answer, providing a reusable data foundation for intelligent processing of cross-border logistics documents. The dataset is based on 99 Excel invoice templates from real business, combined with a virtual entity database covering 30 countries and regions (3022 enterprise records, 3028 address records) and a product database containing 2915 real HS Code constraints. Two batches of 9900 standard Excel invoices were generated through a regularized program. The dataset consists of four subdirectories as a whole: （1）clean_batch_001_cn_shipper_en_consignee_en： Standard data, both the sender and receiver are in English, suitable for all English document creation scenarios. （2）clean_batch_002_cn_shipper_zh_consignee_en： Standard data, sender in Chinese and receiver in English, suitable for mixed Chinese English scenarios. (3) Augmented_batch_001 and Augmented_batch_002: correspond to the enhanced versions of the two batches of standard data mentioned above, and are paired with the standard samples one by one. Version 2 retains the 99 templates, two data batches, and file organization of Version 1. Each batch contains 4,950 standard invoices and their corresponding augmented invoices. This update focuses on data quality: the standard samples and target answers were regenerated using corrected source templates and field mappings to address inconsistencies between invoice contents and answers. The augmented samples were also regenerated with clearer criteria for five anomaly types, so valid formatting variations are not labeled as errors. In addition, the augmented question–answer files were updated to reflect missing fields, duplicate manifest records were removed, and missing paired files were restored. A complete file inventory and validation results are provided to support checks of file integrity and sample pairing. Excel; international trade; invoice Wan Fucheng Wu Chengxue PUBLIC https://creativecommons.org/licenses/by/4.0/ dataset Science Data Bank