Improving Recommendation Accuracy by Considering Electronic Word-of-Mouth and the Effects of Its Propagation Using Collective Matrix Factorization

In recent years, recommender systems have become an important tool for increasing sales and revenues for many online retailers, such as Amazon and eBay. Many of these recommender systems predict a user's interest in the items or the products by using the browsing/shopping history or item rating records of the user. However, many research studies show that, before making a purchase, people often read on-line reviews and exchange their opinion with friends in their social circles. The resulting electronic word-of-mouth (eWOM) has huge impact on customer's final purchase or decision. Nonetheless, most of the recommender systems in the current literature do not consider eWOM, let alone the effect of its propagation. Therefore, we propose a new recommendation model based on the collective matrix factorization technique for predicting customer's preferences in this paper. Our model not only considers customers' personal taste, their trust relationships, but also the effect of eWOM propagation in their social networks. We conduct a series of experiments using real-life data crawled from Epinions and Amazon. Experimental results show that our model significantly outperforms other closely related models that do not consider eWOM propagation effects by 5%-13% in terms of both RMSE and MAE.