ProcureMail-IE Dataset

About Dataset

The ProcureExtract Dataset is a domain-specific dataset developed to train, fine-tune, and evaluate Artificial Intelligence (AI) and Natural Language Processing (NLP) models for automated procurement information extraction. The dataset focuses on converting unstructured procurement emails into structured machine-readable information that can be integrated with enterprise procurement systems, ERP platforms, analytics solutions, and business process automation workflows. The dataset consists of **1,304 procurement email samples** created from a combination of real procurement communications and augmented examples. Augmentation was performed to increase linguistic diversity and improve the model's ability to generalize across different procurement writing styles and formats. The dataset represents a variety of procurement scenarios involving industrial products, engineering materials, manufacturing supplies, and general business procurement requests. Each record in the dataset contains an input procurement email and a corresponding structured JSON output. Every product mentioned in an email is represented as an individual JSON object with five standardized attributes: **product_raw_text**, **product_brand**, **product_specifications**, **order_qty**, and **order_unit_of_measure**. Missing information is represented using empty strings to ensure annotations accurately reflect the source content without introducing unsupported values. The dataset contains both single-product and multi-product procurement requests, exposing AI models to realistic business communication patterns. Procurement emails exhibit significant variation in writing style, formatting, abbreviations, technical specifications, quantity expressions, unit representations, and product descriptions, making the dataset suitable for developing robust information extraction systems capable of handling real-world procurement documents. A standardized annotation methodology was followed to maintain consistency throughout the dataset. Each record was validated to ensure that the extracted fields correspond exactly to the information present in the source email. Quality assurance procedures included validation of JSON structure, verification of non-empty email content, confirmation of at least one annotated product per record, standardized field naming, and removal of malformed or incomplete records. These validation steps improve dataset reliability and support stable model training. The dataset is intended for multiple AI and enterprise applications, including procurement information extraction, purchase order automation, intelligent document processing, ERP data ingestion, enterprise workflow automation, supply chain digitization, procurement analytics, structured data generation, and information extraction research. It can also be used to benchmark sequence-to-sequence models, document understanding systems, and domain-specific NLP applications. Compared with generic NLP datasets, the ProcureExtract Dataset provides procurement-specific vocabulary, realistic business communication, standardized JSON annotations, multi-product extraction capability, and enterprise-oriented training data. These characteristics make it suitable for organizations developing AI-powered procurement automation solutions while reducing dependence on manually designed extraction rules. Although the current version primarily focuses on English procurement emails commonly used in Indian B2B environments, the dataset has been designed with extensibility in mind. Future versions may include multilingual procurement communications, purchase orders, quotations, invoices, supplier catalogues, and scanned procurement documents to support broader enterprise use cases. Overall, the ProcureExtract Dataset provides a high-quality foundation for developing AI systems capable of understanding procurement communications and transforming unstructured business emails into structured, enterprise-ready data. It supports digital procurement transformation by improving automation, reducing manual effort, enhancing data quality, and enabling scalable procurement intelligence across modern enterprise ecosystems.