Analysis of Multilingual Sentiment Analysis of OTT Reviews: A Comparison of CNN, GRU, and KNN-SVD Models

Abstract

The rapid development of Over-the-Top (OTT) platforms in recent years has resulted in a massive flow of multilingual user reviews. Effectively analysing these multilingual sentiments is essential for enhancing content recommendation systems, understanding user engagement and optimizing platform experience. This work presents a unified multilingual sentiment analysis framework designed to assess and compare the effectiveness of conventional AI and deep learning approaches in classifying user sentiments from OTT platform reviews. A curated dataset comprising reviews in four distinct languages like English, Spanish, French, and Hindi was collected from various OTT platforms, translated into English for uniformity and pre-processed using standard natural language processing (NLP) techniques including stop-word removal, tokenization, stemming, and vectorization. Three classification models were analysed and rigorously tested: (i) a traditional KNN algorithm combined with SVD for dimensionality reduction, (ii) a CNN for extracting local textual patterns, and (iii) a GRU model designed to capture long-term dependencies in text. Performance was assessed using standard metrics such as accuracy, precision, recall and F1-score. Experimental results revealed that the GRU model consistently outperformed the others with a classification accuracy of 89%, followed by CNN at 85%, and KNN-SVD at 82%. The GRU model demonstrated superior contextual understanding across languages, particularly in morphologically rich and syntactically complex languages such as Hindi and French. This analysis underscores the constraints of traditional AI models in interpreting semantic nuances in multilingual texts and highlights the advantages of sequential deep learning architectures like GRU. The framework presented in this work is highly scalable and can be integrated into real-time OTT sentiment monitoring systems for enhanced user experience and content personalization.