Enhancing Movie Recommendation Systems Through Explainable Machine Learning Models in the Entertainment Industry

The emergence of digital entertainment systems and the rapid expansion of these systems has driven the necessity of developing reliable and clear movie recommending systems. Although Either content-based filtering or collaborative filtering are features of the conventional recommendation models, they are not always interpretable and do not utilize sentiment information provided in user reviews. This paper introduces a explainable sentiment analysis system that can be considered as an auxiliary element of film suggestion systems. Based on a balanced sample of the IMDB 50K dataset containing 25,000 labelled reviews, an XGBoost-based sentiment classifier will be trained to distinguish between positive and negative reviews. Tokenization, lemmatization, stop-word elimination, and count vectorization are examples of common NLP preprocessing techniques used in feature extraction. The model performance is evaluated using accuracy, precision, recall, F1-Score, and ROC-AUC; XGBoost achieves an F1-Score of 87.27. In order to discuss the transparency issue mentioned in the real-world recommendation environment, post-hoc explainability systems (SHAP and LIME) are used to interpret sentiment prediction on the global and local scales. Contrary to most of the previous IMDB sentiment systems, the innovation of this work is to integrate sentiment modeling and explanation technologies to help with sentiment-sensitive recommendation logic to enhance user trust and developer interpretability.