A Super Linear Collaborative Filtering Algorithm Based on Hybrid Similarity

摘要

The personalized recommendation technology of electronic commerce has become a widely concerned subject in recent years, while the challenge of enhancing recommendation accuracy with sparse data is facing us. In this research we reduce the dimensions of historical rating matrix with hybrid similarity. Then, we replace the null data in nearest-neighbor matrix with values which are computed off line with item-item similarity to alleviate sparsity problem effectively. Finally, we predict the rating value with user-user similarity in a super linear way. The experimental results show that the given algorithm in this research has a better performance in e-commerce recommendation than the traditional algorithms.