Intent recognition and classification from conversations

Abstract

In this digital world, an automated chatbot that could understand human intent and provide necessary information and advices will definitely reduce human workload. For training a chatbot, a large number of data is required. Along with it, intent words are also required that will teach the model how to predict the outputs correctly. Here, we have two conversational datasets, one about the awareness of COVID, and the other about legal domain. The COVID dataset consists of 36 conversations, with 713 utterances. The legal dataset consists of 235 conversations, with 3178 utterances. We have implemented different machine learning and deep learning methods. Random Forest and Support Vector Machine giving the best results of accuracy just over 77%.