Collaborative filtering recommender system performs well when there are enough historical data of the users' online behavior, but it does not work on new users who have not rated any items, or new items that have not been rated by any users, which are called cold-start user and cold-start item, respectively. In order to alleviate the cold-start problem, additional information such as the attributes of users and items must be used. We propose a novel hybrid recommender system, which tries to construct the probabilistic relationship between user attributes and movie attributes using EM algorithm. It can make recommendation for both new users and new items. We evaluate our approach on MovieLens dataset and compare our method with the state-of-the-art approach. Experimental results show that the two approaches have almost the same performance, while our approach uses less time to train the model and make online recommendation.
展开▼
