IEEE Project Abstract

With living data growing and evolving rapidly, traditional machine learning algorithms arehard to update models when dealing with new training data. When new data arrives, traditional collaborativefiltering methods have to train their model from scratch. It is expensive for them to retrain a model andupdate their parameters. Comparing with traditional collaborative filtering, the online collaborative filteringis effective to update the models instantly when new data arrives. But the cold start and data sparsityremain major problems. In this paper, we try to utilize convolutional neural network to extract user / itemfeatures from user / item side information to address these problems. And we makes use of one-by-oneupdating mechanism of online collaborative filtering to update model instantly. First, we proposed a deepbias probabilistic matrix factorization model (DBPMF) by utilizing convolutional neural network to extractlatent user / item features and adding the bias into probabilistic matrix factorization to track user ratingbehavior and item popularity. Second, we constrain user-specific and item-specific feature vectors to furtherimprove the performance of DBPMF. Third, we update two models by online learning algorithm. Extensiveexperiments for three datasets (MovieLens100K, MovieLens1M and HetRec2011) show that our methodshave better performance than baseline approaches.