Self-service workflows for recommendation systems using online machine learning services

This paper details the design and development of a helper software to ease the process of setting up a machine learning system for recommendation systems. This helper automates common tasks of setting up online machine learning systems through a workflow and guides the user through the process in stages. It uses a custom wrapper to interface with the online machine learning service BigML for easy setup of a machine learning system. Optional means of improving the accuracy of a machine learning system are also built into the helper and automated unit tests checks for data integrity. To evaluate whether the artifact created is fit-for-purpose, a user evaluation, specification-based evaluation and demonstration to client/experts were conducted. Overall, the helper reduced the time taken to create a recommendation system from 15 hours to 45 minutes, reducing user interaction (mouse click and keyboard entries) with BigML by 77% and the artifact was found to be fit-for-purpose.