In recent years, recommender systems are widely applied in e-commerce system to help users locating their interested information. However, the 'all good reputation' problem brings down the accuracy of recommender systems. In addition, users' social network can benefit the recommendation especially when dealing with cold-start scenarios. In this paper, a novel trust-aware recommendation approach for e-commerce is proposed, which unearths the hint from ordinary rating and trust network by users' instant interactions in e-commerce system. More precisely, a rating revamping algorithm is designed to extract semantic ratings from feedback comments, and further construct fine grained rating score for the next process. Then, the recommendation scheme is studied through analyzing the users' trust network and their own behavior in e-commerce system. Finally, evaluations conducted based on a real dataset 'Douban' to demonstrate the effectiveness of the proposed method.