An Item Similarity Prediction and Recommendation System using Aspect-based Sentiment Analysis and Graph Neural Networks

As IT devices become increasingly ubiquitous, the global IT device industry continues to grow, highlighting the critical role of consumer reviews. These reviews offer valuable insights into user experiences and feedback, which are essential for product improvement and personalized recommendations. In this study, we conducted aspect-based sentiment analysis (ABSA) on consumer reviews within the IT device sector to predict item similarities and implemented graph neural networks (GNNs) for advanced item recommendations. Using transformer-based models, we identified the optimal architecture for ABSA and utilized the resulting data to construct GNNs. The proposed method demonstrated strong performance in link prediction, achieving high accuracy and robust evaluation metrics. This approach effectively captures aspect-level similarities between items, enabling precise and consumer-focused recommendations. The findings highlight the potential of integrating deep learning-based sentiment analysis and graph learning to enhance recommendation systems in the IT device industry.