Incremental Learning for Dynamic Collaborative Filtering

Collaborative Filtering (CF) is one of the widely used methods for recommendation problem. The key idea is to predict further the interests of a user (ratings) based on the available rating information from many users. Recently, matrix factorization (MF) based approaches, one branch of collaborative filtering, have proven successful for the rating prediction issues. However, most of the state-of-the-art MF models share the same drawback that the established models are static. They are only capable of handling CF systems with static settings, but never practical for a real-world system, which involves dynamic scenarios like new user signing in, new item being added and new rating being given now and then. For conventional MF models, they have to conduct repetitive learning every time dynamic scenario occurs. It is computational expensive and hard to meet the real-time demand. Therefore, an incremental learning framework based on Weighted NMF is proposed. To reduce the computational cost, it utilizes partially the optimization information from the original system, and stores some corresponding information for the subsequent incremental model. Our empirical studies show that the IWNMF scheme for different dynamic scenarios greatly lower the computational cost without degrading the prediction accuracy.