Online Program

Return to main conference page
Saturday, October 5
Sat, Oct 5, 7:30 AM - 8:30 AM
Evergreen Ballroom Prefuction
Continental Breakfast and Speed Poster 4

Machine Learning vs. Transparent Linguistic Models for Contract Analysis (306746)

*Diman Ghazi, IBM 

Keywords: ML, Explainable AI, NLP, Contract analysis, Semantic Linguistic Features

Arguably the most important type of business document, a contract, is one where machine learning (ML) falls the most short. Enterprise users can incorporate NLP models to automatically analyze contracts in their business processes, but struggle due to lack of explainability and consistency. Models also have to evolve over time while expected to behave consistently.

In this talk, we'll show in examples how we at IBM address this problem through building Semantic Language Features on top of Abstract Meaning Representation parsing and tense analysis. This method needs less labeled data compared to other supervised ML models, and is designed to continuously learn from an initial set of rules through direct feedback.