Contract Information Extraction Using Watson NLP And Human-in-the-loop

Abstract

IT service providers sell high-valued service contracts to customers. Since the value of a contract can reach into millions of dollars, they need to be monitored continuously for spend leakage. Spend-leakage is a loss resulting from a mismatch in terms between contracts and invoices. Our work uses an NLP model to extract information from contracts as a step in spend-leakage triaging. The method utilizes a customer profile database, extraction tags, and template outline to accurately extract industry-specific information into in a template format. The extracted template information is checked using human-in-the-loop (HITL) and is then passed into a NLP model as training labels.