With the fast development of e-commerce, most people have begun to buy things online, and the recommendation system equipped by many online stores has greatly promoted the sale of commodities. However, currently most recommendation systems are based on the collaborative filtering algorithm, which needs a lot of user ratings to work and can hardly get accurate recommendation for new users who has rated few items. To solve this problem, this paper proposes a recommendation method based on analyzing customer comments, where a user who has commented briefly on only one product will receive accurate recommendation. At the core of the proposed method is a missing value complement mechanism based on Bayesian Network, where each known value corresponds to a user’s sentiment regarding a product feature. The proposed method is tested against plenty of real comment data collected from an online smart phone store in China and has achieved satisfactory result compared with alternative methods.