Optimization Of Collaborative Filtering Recommendation Algorithm In The E-Commerce

In the open and developing quickly network environment, electronic commerce based on the browser/server mode, make the buyers and sellers not met for various business activities, more and more consumers tend to online shopping. With the advent of the information age and the rapid development of the Internet, e-commerce with its convenient online shopping, is in a period of rapid development and become a fashion and trend of people, showing the great modern commercial value. But more and more people are facing with network information overload problems with in a mass of product information. With the help of E-commerce recommendation system, users can quickly and accurately find useful information on the Internet. Collaborative filtering recommendation system stands out. But some flaws of the collaborative filtering recommendation system have been exposed, such as sparsity, cold start and real issues.Optimization method to the sparsity problem and real-time problem:The sparsity of user-item rating matrix has critical effect on the recommendation quality of the collaborative filtering(CF) algorithm, which is a key issue. Therefore, an item-based attributes similarity matrix filling method is proposed. The item-based similarity, which depends on the common attributes in each item, is adopted to fill the user-item rating matrix. Subsequently, the recommendation to target users is achieved by using the traditional user-based collaborative filtering approach. The effect of time factor on the recommendation quality is also considered in the final procedure. The proposed algorithm substantially improves the quality and accuracy of CF recommendation, which is verified by the experimental result.Optimization method to the cold start problem: The traditional collaborative filtering algorithms has the cold start problem, and reduces the recommended quality. In view of this situation, Ant colony algorithm was used to combine with the attribute information of the users or the items themselves, to solve the cold start problem. Firstly, the existing users and projects were clustered in accordance with the existing content attributes and similarity scores, and the most similar users or items were assigned to the same cluster. Secondly, when the new users or new items entered the recommendation system with no score, the special value of ? and ? of the ant colony algorithm are used to classify new users and new items to cluster similar attributes. Finally, new users or new items are recommended according to average points in the genus.Experimental results show that this algorithm resolves and improves the quality of recommendation.through analyzing experimental result and visual simulation.