Interpretable movie review analysis using machine learning and transformer models leveraging XAI

Abstract

Text classification has been a common topic of interest for many years. A lot of advanced models has been developed so far in this area. But it is very difficult to understand how the models behave while predicting the class of the text. In our work, we utilized some models to classify the sentiment of movie reviews from text data and observed how the models behaved using Explainable Artificial Intelligence (XAI). At first dataset was collected and pre-processed. Then the processed dataset was separated into different train and test sets. The train set was used to classify using different different machine learning and neural network based models. The test set was used after training to evaluate the trained classifiers. Finally, the performance of the classifiers were compared and evaluated. After different variations of pre-processing and training steps, the best accuracy score of 91% was obtained using Roberta LSTM model. In the trained models, we sent texts that are correctly classified by RoBERTa models but misclassified by other models. Finally we figured out the reasons of misclassification with the help of LIME Algorithm.