Generating Challenge Sets for Task Oriented Conversational Agents through Self-Play
With the Rise of Deep Learning Systems, there has been an expectation for a new generation of conversational agents. In this thesis we look at what are the current challenges faced by today’s conversational agents, what solutions are proposed in current literature and finally we propose our own solutions to these challenges. We then test our own deep learning system for a conversational scenario on a data set we create from scratch. We discuss the strengths and characteristics of our Data Set and how to interpret and use it for any future line of research.