Enhanced review-based rating prediction by exploiting aside information and user influence

User-generated reviews greatly supplement the descriptions of items and thereby play an important role in decision making. Researchers have been exploiting these invaluable resources to discover the users’ preferences, model the items’ properties and further provide an explainable recommendation. Legacy strategies seek to quantify the reviews by directly processing the text. However, not all reviews are equally reliable or influential, as the reviews might be generated by different users under various conditions, purposes and habits. Besides, not all reviews given by the users equally contribute to reflecting the users’ preference for the target item since users care about different aspects of different items. In this paper, we propose a novel end-to-end model, named Enhanced Review-based Rating Prediction by Exploiting Aside Information and User Influence (ERP), which differentiates the influence of reviews generated by different users and learns the item-aware user preference with aside information along with their own reviews. On benchmark datasets, our model achieves 1.32% improvements on average in terms of MSE compared to the best result among baselines.