Paper: Comparing User Simulation Models For Dialog Strategy Learning

ACL ID N07-2001
Title Comparing User Simulation Models For Dialog Strategy Learning
Venue Human Language Technologies
Session Short Paper
Year 2007
Authors

This paperexplores what kind ofuser sim- ulation model is suitable for developing a training corpus for using Markov Deci- sion Processes (MDPs) to automatically learn dialog strategies. Our results sug- gest that with sparse training data, a model that aims to randomly explore more dialog state spaces with certain constraints actu- ally performs at the same or better than a more complex model that simulates real- istic user behaviors in a statistical way.