The Research Of Collaborative Filtering Recommendation Algorithm Based On SVD

With the rapid development of the Internet, the network data shows explosive growth and its structure is also becoming increasingly complex. The network is flooded with more and more information which makes people faced with "information overload". The personalized recommendation technology as the representative of recommended system can provide an effective mechanism that allows people to get the information they need efficiently.Collaborative filtering recommendation algorithm is currently applied most successful technology of the personalized recommendation technology, which utilizes the target user’s similar users’rate of an item to generate the prediction rate of the target user to the item. However, with the continuous expansion of information, the collaborative filtering technology is facing increasing challenges. Among them, data sparsity, scalability, cold start are the main problems faced by collaborative filtering.This paper does a deep research on SVD algorithm, latent factor model derived by SVD algorithm and traditional collaborative filtering recommendation algorithms, based on data sparsity and scalability problems in collaborative filtering recommendation technology.Firstly, the paper has a deep study about the background to the development and the architecture of the personalized recommendation technology, in which the similarity measure and the recommendation performance metrics are introduced in detail. It also has a deep study about traditional collaborative filtering recommendation algorithms:user-based collaborative filtering recommendation algorithm and item-based collaborative filtering recommendation algorithm. The two recommendation algorithms’principles are described in detail and their shortcomings are also analyzed.Secondly, the paper does a deep research on SVD recommendation algorithm which is based on matrix decomposition and focuses on SVD’s implementation steps and also analyzes its shortcomings. The latent factor model utilizes gradient descent algorithm to improve the SVD algorithm, solving the problems that SVD algorithm faced. Meanwhile, the paper also describes a number of other improvement SVD algorithms.Finally, due to the loss of some features’data in iterative learning processes of latent factor model, the paper uses KNN algorithm to correct the missing information and then proposes two hybrid recommendation algorithms based on KNN and latent factor model. In order to verify the validity of the proposed hybrid algorithm, the improved hybrid recommendation algorithms, latent factor model and traditional collaborative filtering algorithms are tested on a data set and their test results are analyzed.