Collaborative Filtering Recommendation (CFR) is the earliest proposed and widest used method in recommendation system. It can not only find out what people are interested in at present but also mine out the implicit interest information. As time goes by and data volumes grow recommendations effect can be improved significantly. Therefore the collaborative filtering recommendation is one of the most popular recommendation technologies in the electronic commerce recommendation system. But the algorithm also faces some problems such as cold start sparse real-time and scalability etc. problems affecting the quality of recommendation. An improved collaborative filtering algorithm based on properties is presented in this paper to improve the utilization of data resources and decrease the sparse degree of matrix. Meanwhile in order to reduce the similarity calculation quantity of work and filter dirty data effectively the collaborative filtering algorithm based on clustering and singular value decomposition is put forwards to improve the effect of collaborative filtering algorithm. At last the effectiveness of the improved collaborative algorithm in this paper is verified through experiments.