Novel Self Learning Agent for Chatbot Applications

Abstract

With over 5 billion users using chat applications on a daily basis. This has spiked an interest of large organizations to leverage the power and strength of chat based applications. With the market being more inclined to using chat applications over other modes of communication applications, there has been an increased demand for creating chatbots to provide support over, purchasing, selling etc kinds of tasks. Creating a chatbot restricted to a set of queries within limited context and answers is an easy task but it gets more challenging when you require the bot to keep a conversation going without any human intervening in the process. In this paper we explain the approach through which we train a neural network model to learn and answer to questions on its own while keeping a track of variable context. Using bAbI dataset from Facebook we were able to successfully train our model to do the same with 89% accuracy on an average over 20 tasks provide in the bAbI dataset.