A Multi-feature Fusion Based Rating Prediction Model by Considering Timeliness of Reviews and Influence of Trusted Neighbor Ratings

In recommendation systems, the integration of multiple user-item features derived from diverse datasets, such as rating data and review text, is essential for predicting user ratings. However, prevailing rating prediction models overlook the inclusion of timeliness of reviews and influence of trusted neighbor ratings in the fused features. Consequently, these models fail to capture important information, such as users’ tendency to prioritize reviews closer to the present and their inclination towards friends’ recommendations when selecting items. To address this gap, this paper presents a novel multi-feature fusion based rating prediction model for recommendation systems. Specifically, this paper incorporates timeliness of reviews and influence of trusted neighbor ratings into our model. Unlike existing approaches that solely rely on review text and rating matrices, this paper devise a time-aware attention mechanism to measure the importance of reviews, taking into account their timeliness. Additionally, this paper construct a trust model to obtain a trust vector for user ratings, which helps measure the influence of trusted neighbor ratings. As a result, our model can generate more informative user and item features, enhancing the accuracy of rating predictions. To assess the effectiveness of our model, this paper conduct comprehensive evaluations using five publicly available datasets. The results demonstrate its superiority over baseline models in terms of prediction accuracy.