An Improved Personalized Filtering Recommendation Algorithm

In the development of Web collection recommendation algorithms surrender quality to the huge and sparse dataset. A memory-based collaborative filtering method increases computational complexity. Obviously sparsity and expensive complexity of computation are trade-offs. In order to settle this problem we propose an improved recommendation algorithm based on collaborative tagging called personalized filtering (PF). PF defines and weighs the feature of tags using 4-D dataset, which can show latent personal interests and long-term personal interests. To decrease the computational complexity, PF constructs a top-N tags set to filter out the undersized dataset. To track the changes of personal interests, PF proposed a novel interest changing algorithm on the 4-D dataset. Some empirical experiments were done and the results shown that the sparsity level of PF is much lower and the computing speed is faster than traditional algorithms.