Chatbot via Machine Learning and Deep Learning Hybrid

Abstract

In present-day world, Conversational AI has become a part of our daily life while shopping, banking, or querying any customer service. These are commonly called as Chatbot nowadays, which are mostly used in web or android apps like Siri, Google Assistant, Alexa, etc. This paper introduces a chatbot that can be designed for any specific domain. In other words, it can handle multiple use cases according to the business requirements. The architecture of this chatbot is a multi-level model, consists of a machine learning classifier e.g. SVM, sentence similarities like cosine similarity, and the deep learning seq2seq model. Each level is trained on different datasets except the sentence similarity. We achieved a better result in terms of qualitative and quantitative terms i.e. classifier accuracy of 0.83 and bleu score of 0.31 compared to the state-of-the-art.