A New Collaborative Filtering Algorithm Based on the Improved Similarity

In the memory-based collaborative filtering, the calculation of the similarity is very important, which affects both the neighbors to choose and the weight of rating prediction. The accuracy of the similarity directly affects the performance of the recommendation algorithm. However, the traditional similarity calculation, some only consider the value of common ratings, some only consider the number of common ratings, instead of considering these two aspects. To solve this problem, we first propose a similarity calculation method based on shrinkage factor, which is better than the existing shrunk correlation coefficient; then put forward a similarity calculation method which combines the implicit feedback; finally we replace the basic value in the rating prediction formula with the basic value which is derived from training through the latent factor model like SVD, thus we get a new rating prediction model. Through the experiments on MovieLens dataset, we verify the effectiveness of the algorithm.