Thomas J Walsh, Michael L. Littman
This work considers the problem of learning to perform efficient sequential queries from traces of user behavior. We define a restricted conceptual modeling language and outline how to learn models in this language from behavior traces. We then show how to plan instantiations of concepts in such models and discuss how this planning is affected by different optimality criteria. We examine simple example domains as well as a real world domain involving Amazon web services.
Subjects: 11.2 Ontologies; 1.11 Planning
Submitted: May 15, 2007