Exploring the universe of egregious conversations in chatbots

Abstract

As part of the raging commercial development of chatbot systems, the ability to improve the quality of conversations quickly and consistently is crucial. In this work, we focus on egregious conversations and the dialog failure points that lead to these extremely bad dialogs. We present a tool that helps chatbot designers exploring and prioritizing failure points that need to be handled. Each failure is accompanied with an explanation and a suggested remedy.