Department of Computer, Control and Management Engineering

In the design of a visualization system for exploratory data analysis, a designer faces several issues: (i) the recognition of the causes behind excessive latency experienced by end users, who become quickly disengaged in the exploration if the response time is below a desired threshold (i.e., 500ms); (ii) the discovery of portions of the visualization system that are poorly explored or may not work as intended; (iii) the lack of precise feedback from the end users who, struggling from excessive latency, become disinterested in the exploration and report high-level feedback that is too broad and generic for the designer to understand and transform into actionable changes to the design. To address these issues and provide more guidance to visualization system designers, we contributed a general framework to model and assess user interactions in big data visualization systems. It models the interaction space of the vi- sualization system with the concept of augmented statecharts that label interactions with their latency thresholds. It is implemented in a system, InterView (the name relates to the collaboration between visualization designers and end users), composed of two software components, one to automatically generate the interaction space of a visualization system using a statechart, and one to replay user traces, reproducing each in- teraction an end user performed in the interaction log. In this paper, we demonstrate the capabilities of InterView applying it to a well-known crossfilter interface, Falcon, to guide the visualization system designers in discovering the root causes behind excessive latency, coupled with a complete understanding of the interaction space of their visualization system. In such a way, designers can finally acknowledge the problems of their visualization system with higher granularity and precision, giving more context to the feedback received by the end users.