Exploit the visual sentiment of the item images to fuse with textual sentiment in context aware collaborative filtering

• Published in: Expert systems with applications Vol. 265; p. 125970
• Main Author: Wu, Liang-Hong
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.03.2025
• Subjects: Context aware recommendation system (CARS); Factorization Machines; Multi-modal sentiment fusion; Tensor decomposition; Unsupervised deep learning; Visual sentiment analysis; Factorization Machines; Tensor decomposition; Unsupervised deep learning; Multi-modal sentiment fusion; Context aware recommendation system (CARS); Visual sentiment analysis
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2024.125970

Recommendation systems (RS) are widely used to predict users’ preferences for items in many research areas. Context-aware recommendation systems (CARS) exploit different contextual circumstances to predict changes in user preferences and improve RS performance. However, conventional CARS focuses on user context and neglects item context despite item images’ rich textual and visual sentiment information. We bridge the gap between sentiment analysis and conventional CARS by using sentiment as item context and applying Factorization Machines to fuse different modal information. Our ablation study shows the essential role of visual sentiment in CARS, particularly in scenarios with numerous item images, such as movie recommendations. With carefully selected baselines and datasets, our proposed IC-CARS demonstrates superior performance across selected metrics, particularly in movie recommendation scenarios. [Display omitted] •Integrated sentiment analysis with collaborative filtering CARS.•Treated textual and visual sentiment as context.•Proposed multi-modal sentiment fusion technique.•Developed sentiment-aware recommendation system model.