Sentiment aware kernel mapping recommender system for Amazon product recommendations

• Published in: Pattern analysis and applications : PAA Vol. 28; no. 3
• Main Authors: Bukhari, Maryam, Maqsood, Muazzam, Ghazanfar, Mustansar Ali, Sattar, Asma
• Format: Journal Article
• Language: English
• Published: London Springer London 01.09.2025; Springer Nature B.V
• Subjects: Computer Science; Mapping; Original Article; Pattern Recognition; Probability density functions; Ratings; Recommender systems; Sentiment analysis; System reliability; Sentiment analysis; Kernel mapping algorithms; Textual feedback; Sentiment kernels; Natural Language processing; Recommender systems
• ISSN: 1433-7541; 1433-755X
• DOI: 10.1007/s10044-025-01521-x

Recommender systems help address information overload in domains such as e-commerce, entertainment, travel, education, and social media. Their major objective is to generate personalized recommendations that help users find relevant items. However, cold start and data sparsity issues hinder the model’s ability to understand user preferences accurately. Moreover, quantitative ratings alone do not capture the full spectrum of user sentiment, hence, to increase the reliability of recommender systems, analyzing user-generated textual reviews can provide deeper insights into user opinions. From this perspective, we have exploited user sentiment analysis in kernel mapping recommender systems referred to as KSR. Specifically, we propose a sentiment-based kernel mapping recommender that applies sentiment analysis to user reviews. Instead of traditional rating-based matrices, we construct user and item-level kernels over sentiment-enriched user-product matrices. The proposed KSR learns the multi-linear mapping between encoded vectors of user-product associations and the probability density function that determines the user’s sentiments toward products. In the second stage, we integrate quantitative ratings with sentiment information using both additive and multiplicative modeling approaches within the KSR. The proposed model is validated on three different product datasets from Amazon including Food, Software, and Music Reviews. The findings indicate that the proposed sentiment-based KSR model outperforms the original KMR method by achieving the best MAE error of 0.3306 and precision@20 score of 75.44% respectively.