Intent classification: a comparative analysis

Abstract

Understanding the user's purpose is a critical stage in any human-computer interaction in dialogue systems. Typically, different classifiers are used to classify natural language sentences into specified intent groups. Intent Classification is the task of comprehending and classifying the intents based on the text presented. The goal of this thesis is to classify the intentions based on the provided texts. We initially trained the texts in the datasets using a variety of methods, including Classical Text Classification models like Linear Support Vector Machine, Linear Regression, Random Forest, and Multinomial Naïve Bayes and deep learning methods like Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and the state-of-the-art model, BERT (Bidirectional Encoder Representations from Transformers) and compared the results. In this thesis, we have used three different datasets which are SNIPS Natural Language Understanding Benchmark (SNIPS), Banking77, and ATIS Airline Travel Information System. We have individually trained and tested all three of the datasets on the above-given models and we have got results based on it.