Intent Sets: Architectural Choices for Building Practical Chatbots

Abstract

"Chatbot" is a colloquial term used to refer to software components that possess the ability to interact with the end-user using natural language phrases. Many commercial platforms are offering sophisticated dashboards to build these chatbots with no or minimal coding. However, the job of composing the chatbot from real-world scenarios is not a trivial activity and requires a significant understanding of the problem as well as the domain. In this work, we present the concept of Intent Sets - an Architectural choice, that impacts the overall accuracy of the chatbot. We show that the same chatbot can be built choosing one out of many possible Intent Sets. We also present our observations collected through a set of experiments while building the same chatbot over three commercial platforms - Google Dialogflow, IBM Watson Assistant and Amazon Lex.