Research On And Application Of Similarity Calculation And Users’ Interest Drifting Of Collaborative Filtering Algorithm

With the fusion of telecom network, Internet and broadcasting network, IPTV developed swiftly as a new generation of digital cable TV products. The number of IPTV users is growing rapidly in China nowadays due to its good interaction. IPTV content providers always supply digital images, videos, audio and so on. Because of the huge amount of information, it’s not easy for the users to find out which programs the users are interested in, and difficult to search for the program that they might be interested in despite complex operation.Currently, the recommendation algorithms such as collaborative filtering algorithm, content-based recommendation algorithm and hybrid recommendation algorithm are mainly for e-commerce recommendation systems, which is different from IPTV. Just for this reason, the accuracy of general recommendation algorithms is greatly reduced when applying in IPTV, so that the specialized algorithms for IPTV are in need.Collaborative filtering algorithm is the most widely used and the most successful recommendation technology of personalized recommendation system. To measure the similarity of the users is the core of collaborative filtering algorithm. The traditional similarity measurements cosine and Pearson similarity considered inadequate in terms of the vector length and the degree of overlap. In addition, the traditional ones have taken users’interests in different time into equal consideration.Considering the above problems, we analyses the difference between e-commerce and IPTV. and present an improved collaborative filtering algorithm TJacUOD and an improved content-based recommendation algorithm. The improved collaborative filtering algorithm T_JacUOD is based on JacUOD and time weight. The results of the experiment which use the dataset of MovieLens show that, comparing with tradition collaborative filtering algorithm and it’s improved algorithms, the MAE of T_JacUOD is lower, which means it’s prediction accuracy is improved to a certain extent. This paper uses the data sets of iTV movies-on-demand of Fujian Telecom system to validate the effectiveness of the improved content-based recommendation algorithm. The results of the experiment show that when compared with other algorithms such as label-based collaborative filtering recommendation the hit rate the proposed algorithm has increased, its recall rate is reduced, which means its prediction accuracy and performance are improved to a certain extent.Finally, the two algorithms proposed in this paper is used in the different occasions of Fujian Telecom iTV video recommendation system to achieve a combination of theory and practice.