“…To better represent QoS data, Guo et al [19] proposed three product recommendation models based on implicit user feedback combined with social trust, and introduced a matrix factorization technique to restore user preference between rated items and unrated items according to user-user and item-item similarity. Zhou et al [20] integrated the user's network location information into the MF model. Although many studies have improved the prediction accuracy from a series of aspects, the existing methods have severe limitations and can only extract or learn shallow features.…”