Personalized Service Research

Personalized services are highly demanded along with the explosion of information on Internet. Because of the diversity of users' interests and behavior, it is a great challenge to provide appropriate services for different users. Collaborative filtering has many problems to be solved, which is the most successful personalized recommendation technology in research fields and business applications. This thesis is to solve those problems.The main content of this thesis can be summarized as follows.First, this paper discusses the major problems of personalized services, such as representation of user profile, representation of resources, and recommendation technologies . Those are the base of understanding of the development of personalized services technologies, existent problem and the precondition of research.Secondly, this paper analyses collaborative filtering in details. However, existing collaborative filtering algorithms only consider users' mark information but do not consider users' social information, and do not consider the creditability of users' mark. To solve this problem, this paper puts forward the calculation of social information similarity, and credit evaluation of users' marks, so we can find more similar neighbors and more credible marks to calculate.Finally, the predicted value produced by collaborative filtering algorithm is always a decimal fraction, and needs to be judged as an integer correspond to some grade. However, existing collaborative filtering algorithms round predict value and get the judgment value simply without consideration of users' grade trend. To solve this problem, the paper describes a judgment algorithm for predict value based on users' grade trend. The algorithm considers the distance between the predicted value and the grade value, and users' grade trend, and then gets the judgment value. Experimental results show that our proposed algorithm outperforms traditional collaborative filtering algorithm.