Intent Classification of Users Conversation using BERT for Conversational Dialogue System

Intent Analysis and detection are currently getting a lot of importance for their significance in both the Industry and Academia. The intent classification relies heavily on the rapidly expanding unstructured data of microblogging networks like Twitter and Facebook. However, the work is extremely difficult because of the social media data’s frequent noise and diversity. Furthermore, because labelled datasets are usually manually annotated, intent analysis frequently suffers from a lack of data. Modern language representation model BERT (Bidirectional Encoder Representation from Transformers), which accurately models’ language, has recently gained interest. In this paper, we developed an Intent Classification Model using BERT for the classification of Questions received from the Users or Humans to specific intents regarding the usage of specific features and components of the car. The dataset for the Classification was developed from the Owner’s Manual of the vehicles. The Classification Model was developed using BERT and produced promising results for the classification of Real Work Text into 199 different categories or Intents. The Model which was trained with 254,412 records plotted an overall categorical accuracy of 98.21%