Research On Collaborative Filtering Recommendation Algorithm And Its Optimization

With the rapid development of Internet technology,people have entered into the era of Web 2.0,the explosive growth of information makes it more difficult for users to get the information what they need,which undoubtedly caused a serious problem of information overload.In this case,personalized recommendation system came into being as an effective means to solve this problem at the historic moment.This technology is based on the information filtering technology and information retrieval technology,which can actively recommend useful information to users.Among many recommendation techniques,the collaborative filtering technology has become the most widely used method in recommender systems,but there are still some key problems which restrict its further development.In this paper,the recommendation system based on collaborative filtering technology is the research object.Aiming at the existing problems,in which the timeliness of user interest,data sparsity and system scalability,a series of research work has been carried out.The main contents of this paper are as follows:Firstly,aiming at the timeliness of user interest,this paper proposes a new algorithm based on tag and integrating the short-term and long-term interests of users,in which the short-term and long-term behavior of users achieves a good balance,so as to ensure the diversity and continuity of interest prediction.Firstly,from the perspective of user's tag,the characteristics of the short-term and long-term interests of users are constructed.Then,the feature is mapped to the user annotated resources to form a user-resource pseudo score matrix.Finally,the nearest neighbor set of user is calculated by the matrix,and the recommended results are produced.Then,aiming at the problem of data sparsity and system scalability in collaborative filtering recommendation system,this paper proposes an optimized recommendation algorithm based on KPCA dimensionality reduction and improved K-means clustering.Firstly,kernel principal component analysis(KPCA)is introduced to reduce the user-resource score matrix.Then,aiming at the disadvantage of the traditional Pearson similarity calculation method,correction factor are introduced to optimize the similarity calculation method.Finally,the improved similarity calculation method is applied to the K-means clustering to cluster the dimension reducted data of uers,the nearest neighbor set of user is searched in the cluster of this user,and the recommended results are produced.Finally,the standard data sets are used to evaluate the performance of the optimized collaborative filtering recommendation algorithms.By comparison with other recommendation algorithms,it can be found that the two kinds of optimization algorithm proposed in this paper have relieved the problem of timeliness of user interest,data sparsity and system scalability respectively,the quality of recommendation results has been improved.