An Enhanced Sentiment Score Prediction in the Collaborative Filtering Algorithm through a Hybrid-Dynamic-Topic-Sentiment User-based KNN (HDTS-uKNN)

Currently, social media generates a large amount of content every day, with relevant content including the experience of product performance, price, and branding, which can assist retailers in understanding the change of consumers’ sentiment toward a certain type of product and provide recommendations for consumers to purchase related products. Although different existing recommendation systems have recommended products through users’ ratings, the recommendation systems generally neglect the time factor, contain sparse data, and lack the ability for semantic understanding. The current study proposed an enhanced collaborative filtering algorithm, namely the Hybrid-Dynamic-Topic-Sentiment user-based K-Nearest Neighbors (HDTS-uKNN), based on the topic model, sentiment analysis, and time decay function. The experimental results revealed that the proposed algorithm demonstrated enhanced performance compared to other baseline algorithms on the Twitter dataset, with the lowest mean absolute error (MAE) score of 0.1003. The proposed collaborative filtering algorithm not only aided in resolving the data sparsity issue but also utilised the dynamic hybrid similarity method to contribute to more refined product and service recommendations.