Research On Personalized Recommendation Algorithm Based On Collaborative Filtering And Implementation Of System

With the rapid development of Internet technology, electronic commerce is gotten widely attention because of its convenience. But with the increase of the number of the commodity resources, it is hard for customers to find satisfied commodity in a relatively short period of time conveniently in the network shopping. In order to help the customers find the needed goods quickly, and bring higher profits to merchants at the same time, personalized recommendation technology and personalized recommendation system arises at the historic moment.At present, the collaborative filtering recommendation technology in personalized recommendation technology is applied widely, but some shortcomings exist, such as the matrix of score is too spars. In this paper, in order to improve the accuracy of recommendation, the traditional user-based collaborative filtering algorithm is improved. This paper proposes a new recommendation algorithm which is with the RFM model and calculating similarity with the shopping behavior of users, the beneficial exploration and research has been carried on in the key technologies, such as recommended strategy and recommendation algorithm. Compared with the traditional user-based collaborative filtering recommendation algorithm, the advantage of this algorithm mainly embodied in the following aspects:First, the improved algorithm is with RFM model which is used to select the original customer in some condition, making the recommended source of data more accurate and efficient; Second, in the improved algorithm the customer consumption history records is filled to the matrix to improve the consistency of the matrix of score; Third, the traditional Pearson similarity calculation formula is improved to make the search of target users of similar neighbor more accurate. Then the simulation experiment is carried on by using the improved algorithm. Comparing with the traditional algorithm by using Mean Absolute Error (MAE) as the measure indicator of quality of recommendation algorithm, it can be proved that the improved algorithm is better than the traditional one in accuracy, especially in the case of the matrix of score is sparse. At last, the improved algorithm is applied to a recommendation system with personalized recommendation function and finally a practical system prototype is achieved.