Creating an Artificial Intelligence for NDA Evaluation

Abstract

The non-disclosure agreement (NDA) or confidentiality agreement is a staple of the business community, in particular the high-tech and IT business. However, there is something of a gap in the market between what legal professionals need to review an NDA and what businesspersons are prepared to offer for such services. A machine learning system is constructed that can recommend whether a given NDA is acceptable or not from the perspective of the user. The system divides NDAs into sentences regarding different topics and uses a first support vector machine to assign topics. Sentences are then grouped by topic and qualified (e.g. as strict or relaxed, or as standard, broad and limited) by a set of second support vector machines. The system was trained on 80% of 304 source documents, and performance was enhanced using various techniques, resulting in a machine learning system that can make correct recommendations on NDAs with 87% accuracy (F1=0.86).

Keywords: artificial intelligence, Bayesian classifier, confidentiality agreement, non-disclosure agreement, support vector machine

JEL Classification: K12, M13