Informed Active Learning to Aid Domain Experts in Modeling Compliance
Sagar Sunkle, Deepali Kholkar, Vinay Kulkarni · 2016
Modern enterprises face an unprecedented regulatory regime. Traditional compliance practices in enterprises rely heavily on domain experts whose judgement determines what compliance means and how to reflect regulations onto the enterprise processes and data to make them compliant. These activities are mostly manual in nature. We present a machine learning approach to modeling compliance. Our key innovations are a) use of active learning- a semi-supervised system capable of learning interactively from the domain expert to identify regulations and b) informing the feature representation of the active learner based on domain- specific entities and relations to effectively build a domain model of regulations. Early results show that our system reduces the burden on domain experts to a large extent, enables latching domain expert's knowledge, and makes further steps in compliance easier by the use of models.