Objective:

Leverage NLP based techniques for risk-based contract classification for a large number of contracts with a quick turnaround

Key Challenges:

• 100,000+ legal documents to be evaluated
• Lack of a standard format
• Foreign language usage in some of the contracts

Approach:

• Pre-processing engine for white-space removal, punctuation-removal, stop-words removal, etc.
• Term document matrix creation
• Text Classification and NLP Algorithms were leveraged to build the foundational ontology using feature extraction
• Use Expectation Maximization Algorithms to transfer the classification knowledge across languages, by translating the model features
• Use the extracted feature set, in conjunction with business rules to flag contracts into three risk categories – high, medium, and low
• Validate results against test set, and have incorporate feedback loop to continuously improve the model accuracy

Benefits:

• Identified fraudulent contracts with a precision of 98%
• Automated engine to efficiently parse 10000+ contracts per day