Invoice Notes Alert System
Project description
In the field of technology expense management, we handle tens of thousands of invoices, along with numerous invoice notes indicating rate changes and disconnection notifications. The challenge is that these notes are often lengthy and go unread, leading to unexpected financial and operational disruptions like unplanned rate hikes or circuit disconnections. The project aims to develop a practical solution for efficiently extracting and communicating key information from these notes to our clients, saving both time and resources.
To address this issue, I utilized Natural Language Processing (NLP) techniques to swiftly identify essential details from invoice notes. I developed two NLP models for this purpose. The first model employs a multi-class classification approach, segmenting each note into individual sentences and categorizing them into classes such as 'Rate Increase.' This was done using a pre-trained transformer model and an initial labeled dataset. The second model utilizes Named Entity Recognition (NER) to extract key terms like 'Charge Type' and the associated rates. This model was built using the spaCy Python package, a tool commonly used for NER tasks. As a result, we've achieved a high accuracy rate in predicting not only the categories of invoice notes but also identifying specific charge types and rates when applicable.
