From rants to raves: unraveling movie critics’ reviews with explainable artificial intelligence

Abstract

Text classification and sentiment analysis are well-established methodologies, but the explainability of text classification needs to be adequately explored. There is a growing emphasis on making machine learning more interpretable and explainable. To address this, we used the Rotten Tomatoes movies and critic reviews dataset to explore the use of eXplainable Artificial Intelligence (XAI) methods in combination with various machine learning algorithms to identify words and features in text that can predict the label of the text which is related to sentiment of the text. We began by feature engineering through linguistic inquiry and word count to extract a series of features from the text. Then, we used classification-based machine learning algorithms to predict the label (i.e., fresh/rotten). We surveyed different algorithms to find the best-performing model based on performance metrics such as the Receiver Operating Characteristic (ROC) curve and confusion matrix. Finally, we applied global and local model-agnostic XAI methods to the best-performing algorithm to make the machine learning model interpretable and identify and explain which text features drove the prediction.