Building Adaptive Industry Cartridges Using a Semi-supervised Machine Learning Method

摘要

In the middle ground between research and industry applicability, there is optionality, although the first comes with proven results, the latter is challenged by scalability, constraints and assumptions when applied in real case scenarios. It is very common that promising research approaches or PoC (proof of concepts) encounter difficulties when applied in industry solutions, due to specific industry requirements, bias or constraints. The paper is to show how industry business knowledge can be incorporated into machine learning algorithms to help eliminate bias, that might have been overlooked, and build industry domains cartridges models to be used in future solutions. The industry models are currently explored by businesses' that want to enhance their portfolios with cognitive and AI capabilities and learn from transaction-based insights. With this research we aim to show how best machine learning models can learn from industry expertise and business use cases to create re-usable domain cartridges which can stand as core for: Bots, RPA (Robotic Processing Automation), industry patters, data insights discovery, control and compliance.