An Approach Toward Stateless Chatbots with the Benefit of Tensorflow Over Spacy Pipeline

Abstract

Conversational bots are assuming control over the business space. Consistently, individuals send in excess of a billion messages to organizations and associations through numerous messaging applications. The most usual method for building chatbots is the state machine approach which usually involves creating distinctive states and dependent on some logic invoking actions. But, with the increase in the number of states, lot of rules are required to be added, with additional logic, and hence create a delicate code which is difficult to keep up and maintain. In this work, we have shown how the stateless approach of building a chatbot using Rasa Core eliminated the need of complex state machine approach, as it makes use of machine learning-based dialog management. Along with this, we trained the model separately with the two pipelines “spacy” and “Tensorflow embedding”. We evaluated the two pipelines with cross-validation and without cross-validation for intent classification. The result showed that, with more training examples, Tensorflow embedding shows good accuracy for intent classification.