Title: MRMRP: Multi-source Review-based Model for Rating Prediction
Authors: Xiaochen Wang, Tingsong Xiao, Jie Tan, Deqiang Ouyang and Jie Shao
Abstract: Reviews written by users often contain rich semantic information which can reflect users' preferences for different attributes of items. For the past few years, many studies in recommender systems take user reviews into consideration and achieve promising performance. However, in daily life, most consumers are used to leaving no comments for products purchased and most reviews written by consumers are short, which leads to the performance degradation of most existing review-based methods. In order to alleviate the data sparsity problem of user reviews, in this paper, we propose a novel review-based model MRMRP, which stands for Multi-source Review-based Model for Rating Prediction. In this model, to build multi-source user reviews, we collect supplementary reviews from similar users for each user, where similar users refer to users who have similar consuming behaviors and historical rating records. MRMRP is capable of extracting useful features from supplementary reviews to further improve recommendation performance by applying a deep learning based method. Moreover, the supplementary reviews can be incorporated into different neural models to boost rating prediction accuracy. Experiments are conducted on four real-world datasets and the results demonstrate that MRMRP achieves better rating prediction accuracy than the state-of-the-art methods.
