A Novel Nearest Neighborhood Algorithm for Recommender Systems
Xiong Lei, Yang Xiang, Qi Zhang, Lili Lin · 2012
Traditional k-nearest neighborhood (KNN) model is being widely used in the recommender systems. However, it behaves badly without enough history records for new users, called the cold starting problem. Both time and space complexity are huge for computing all pair wise similarities among items or users. A mixed neighborhood algorithm is proposed for treating new users and old users separately. For new users, this paper takes into account users' characteristics. For old users, combined with Singular Value Decomposition (SVD), we reduce the time and space complexity efficiently. Experiment on Movie Lens dataset shows that the proposed model can solve the cold starting problem in effect and remarkably improve the accuracy of traditional model and lower time consuming level.