Towards Intelligent Recommendations: Exploring Sentiment and Context in Personalized Systems
Md Mahtab Alam, Mumtaz Ahmed · Procedia Computer Science · 2026
In recent years, recommendation systems have evolved from basic collaborative and content-based approaches to more intelligent models that incorporate user sentiment and contextual awareness. This paper explores the dual integration of sentiment analysis and context-sensitive computing to enhance the personalization and relevance of recommendations. By analyzing a spectrum of research contributions from 2019 to 2025, we observe a clear paradigm shift: from static user modeling towards a dynamic and fine-grained understanding of user preferences. Early works (2019-2021) focused primarily on augmenting collaborative filtering with basic contextual factors (e.g., time, location) or sentiment polarity derived from user reviews. However, these methods often suffer from cold start and data sparsity issues. More recent studies (2022–2025) have employed deep learning techniques, such as attention-based networks, graph neural networks, and hybrid fusion models, which demonstrate improved adaptability and accuracy. We conducted a comparative evaluation of key models over the past seven years, based on dimensions such as data modality, personalization depth, interpretability, scalability, and performance metrics, that is, Precision@K, Normalized Discounted Cumulative Gain (NDCG), and Recall. Our analysis reveals that hybrid models, which leverage both sentiment-aware embeddings and contextual information (e.g., user mood, device type, session data), outperform traditional recommendations by a significant margin in terms of user satisfaction and prediction accuracy. The paper concludes with a discussion on emerging challenges such as privacy, bias, and explainability, and outlines future directions for building brilliant and empathetic recommendation systems.