Workflow-based Human-in-the-Loop Data Analytics
Juan Liu, Aaron Wilson, David Gunning · 2014
Exploration in data analytics system requires proficient skills in data science and a deep understanding of the domain, both of which are scarce and hard to find. Experienced analysts leverage knowledge from past experience and solve a problem within minutes, yet the same problem could cost novice analysts days. In this paper we advocate a human-in-the-loop data analytics system which logs the workflow of users within the data analytics system and learns common workflow patterns using graph analysis, information scent, and instance-based learning techniques. Based on the learned common patterns, the system automates laborious processing steps to improve analyst productivity and provides recommendations based on the workflow of expert users. This will narrow the productivity gap and lead to improved usability.