Review-Based Unify MF and GraphSAGE to Enhancing the Recommendations Accuracy

In the age of information overload, effective rec-ommendation systems (RS) are crucial for helping users find relevant content. This study investigates the integration of traditional matrix factorization (MF) techniques with GraphSAGE to improve link prediction in recommendation systems (RS). The proposed approach overcomes the limitations of rating-based recommendation methods by applying sentiment analysis to review data to generate initial embeddings. Experiments on the Amazon movie and music datasets reveal that this hybrid model significantly improves prediction accuracy, evidenced by much higher AUC values. The findings demonstrate that combining MF and graph-based techniques leverages their strengths, offering a robust and reliable framework for predicting user preferences.