To improve the performance of the commodity recommendation system, this paper proposes two methods: (1) recommend commodity in real-time; (2) understand customers’ purchase intentions. In order to achieve the above goals, we summarize commodity recommendation location, and the label of commodity. Based on the above, we propose a scheme to improve the performance of the commodity recommendation system. We conduct an in-depth analysis on existing work, and give necessary theoretical analysis and performance evaluation for the proposed commodity recommend scheme. From the technical point of view, this method improves the above commodity recommend scheme.