Uncovering User Interest during Data Exploration via Unsupervised Clustering


The emergence and increasing usage of large and dense datasets pose a challenge to both the user and visualization designers. As the visualization becomes more congested with information, it grows difficult for the user to make observations due to information overload. One solution that has garnered immense interest in recent years is to create intelligent systems that learn the user’s interest and aid in data exploration. However, inferring high-level interest is still an open challenge. In this paper, we present a technique for uncovering potential data points of interest based on user interactions by learning natural groups in a given dataset via unsupervised clustering. We validate our technique’s ability to discover user interest with two crowd-sourced interaction datasets. We discuss and demonstrate how to incorporate our approach into an adaptive visual system that supports users during data exploration.