Improving Hybrid Recommendations with VADER-Powered Sentiment Analysis^*

Abstract

Recommendation systems play a crucial role in helping users discover personalized and relevant content. In this study, we focus on optimizing hybrid recommendations by incorporating sentiment analysis. By analyzing the sentiments expressed in user-generated content, such as reviews and feedback, we gain valuable insights into user preferences. By leveraging sentiment analysis, we can enhance the effectiveness of recommendations, resulting in improved user experiences and increased engagement. Our proposed approach combines Singular Value Decomposition (SVD) for collaborative filtering, TF-IDF with Lasso for content-based filtering, and VADER for sentiment analysis. We transform continuous sentiment scores into discrete ratings and replace the initial ratings with sentiment-derived ratings from user reviews. This hybrid system demonstrates improved Root Mean Square Error (RMSE) and provides recommendations that are tailored to user preferences. This showcases the potential of sentiment-aware approaches in enhancing recommendation systems. Furthermore, we propose a hybrid recommendation system that utilizes Non-Negative Matrix Factorization (NMF) and DecisionTreeRegressor-based models. This hybrid model outperforms all other models, achieving an impressively low RMSE score of 0.092.