Automating Document Processing in the Supply Chain with NER and Transformer-based Information Extraction methods

This thesis presents a solution to enhance supply chain management and optimize logistics by reducing the need for human intervention and monotonous tasks. The proposed solution is grounded on the extraction of information from diverse sources, such as PDF documents, images, and emails, employing advanced technologies, such as natural language processing, computer vision techniques, and optical character recognition methods. Additionally, the research integrates the use of transformers, AI document processing, named entity recognition, and data annotation to enhance the accuracy and efficacy of the information extraction process. The chief objective of this study is to aid companies in saving valuable time and cutting costs by deploying these models. The proposed solution automates the process of gathering and organizing information, freeing employees from tedious tasks and enabling them to concentrate on more valuable and crucial assignments. The results of this study reveal the feasibility of deploying these models to streamline the supply chain management process and boost the efficiency of logistics management. By saving time and reducing expenses, firms can boost their overall profitability and competitiveness in the market. This project was conceived as part of an internship at Wenda SRL, a burgeoning startup specializing in supply chain management based in Bologna. The study was conducted using the customary CRISP-DM methodology for data science projects, with a focus on improving operational efficiency in the SCM industry. In compliance with company policy, the dataset used in this dissertation is classified. The implementation phase leverages Hugging Face to prepare models for integration as supplementary products for Wenda SRL’s clientele.