Plan recognition and information visualization for exploratory domains

In exploratory domains, agents’ actions map onto logs of behavior that include switching between activities, extraneous actions, and mistakes. These domains are becoming prevalent (e.g., flexible pedagogical software and IDEs, the one-laptop-per-child project) but present real challenges to inferring users' activities and to generating advice for the users. I will present several new algorithms for performing plan recognition in these domains,  including new heuristic methods that vary the extent to which they employ backtracking, as well as a reduction to constraint-satisfaction problems. The algorithms were empirically evaluated on people’s interaction with flexible, open-ended education software used in schools in several countries. I will show that the algorithms have excellent performance and generalizability ability when tested on real data collected from students using several types of  software to solve different problems. I will also present studies demonstrating the efficacy of these approaches when visualizing students' interactions to teachers.