A Comparative Analysis Between Deep Learning and Machine Learning Algorithms Based on User Review Sentiment Analysis from Various OTT Applications

The golden period of television has moved to the screen in our hands, as streaming in the era of OTT(over-the-top) platforms. However, among OTT users, disengagement and platform churn are becoming more frequent. This study examines sentiment analysis in the context of OTT content through a comparative analysis of machine learning (ML) and deep learning (DL) models on user reviews from the Google Play Store and Apple App Store. We analyzed 56,351 user reviews from ten popular OTT apps, classifying them as positive (21,446), neutral (19,120), and negative (15,785). Preprocessed and feature-extracted data was fed to both ML models (Logistic Regression, XGBoost, etc.) and DL models (BiLSTM, LSTM, CNN). All of these are capable of extracting textual characteristics and insights from datasets and are also used to capture complex sentiment nuances in user reviews. BiLSTM, a deep learning algorithm, surpassed all other models, achieving an astounding 92% accuracy compared to Logistic regression, which achieved 66.62% accuracy and was the best-performing machine learning model in capturing user sentiment.