Apr 17, 2018
Jolley Hall, Room 309
"Inferring Intent from Interaction"
Thesis Adviser: Alvitta Ottley
Today's state-of-the-art analysis tools combine the human visual system and domain knowledge, with the machine's computational power. The human performs the reasoning, deduction, hypothesis generation, and judgment. The entire burden of learning from the data usually rests squarely on the human user's shoulders. This model, while successful in simple scenarios, is neither scalable nor generalizable. In this thesis, we propose a system that integrates advancements from artificial intelligence within a visualization system to detect the user's goals. At a high level, we use hidden unobservable states to represent goals/intentions of users. We automatically infer these goals from passive observations of the user's actions (e.g., mouse clicks), thereby allowing accurate predictions of future clicks. We evaluate this technique with a crime map and demonstrate that, depending on the type of task, users' clicks appear in our prediction set 79% -- 97% of the time. Further analysis shows that we can achieve high prediction accuracy after only a short period (typically after three clicks). Altogether, we show that passive observations of interaction data can reveal valuable information about users' high-level goals, laying the foundation for next-generation visual analytics systems that can automatically learn users' intentions and support the analysis process proactively.