Learning to Predict User Operations for Adaptive Scheduling

Mixed-initiativesystemspresent the challengeof finding an effective level of interaction betweenhumans and computers. Machinelearning presents a promising approachto this problemin the form of systems that automatically adapt their behavior to accommodate different users. In this paper, wepresent an empirical study of learning user modelsin an adaptive assistant for crisis scheduling. Wedescribe the problemdomain and the schedulingassistant, then present an initial formulation of the adaptiveassistant’s learning task andthe results of a baselinestudy. Afterthis, wereport the results of three subsequentexperiments that investigate the effects of problemreformulation and representation augmentation. Theresults suggest that problemreformulationleads to significantly better accuracywithoutsacrificing the usefulnessof the learned behavior. Thestudies also raise several interesting issues in adaptiveassistancefor scheduling.

Authors

• Melinda T. Gervasio Wayne Iba

Keywords

No keywords are indexed for this paper.