Dialogue Strategy Optimization with Reinforcement Learning in an AT&T Call Routing Application

Contents

Reinforcement Learning (RL) algorithms are particularly well suited to some of the challenges of spoken dialogue systems (SDS) design. RL provides an approach for automated learning that can adapt to new environments without supervision. SDS are constantly subjected to new environments in the form of new groups of users, and developer intervention is costly. In this paper, I will describe some results from experiments that used RL to select the optimal prompts to maximize the success rate in a call routing application. A simulation of the dialogue outcomes were used to experiment with different scenarios and demonstrate how RL can make these systems more robust.