Using Category and Keyword for Personalized Recommendation: A Scalable Collaborative Filtering Algorithm

Scalability is another major issue for recommender systems except data sparsity and prediction quality. However, it has still not been well solved while many social recommendation models have been propose to improve the latter two problems. In this paper, we propose a scalable collaborative filtering algorithm based matrix factorization that introduce two common context factors: category and keyword besides social information. In the proposed model, we make prediction together using two preference matrices:user-category and user-keyword instead of only using the user-item rating matrix. This has the advantage that for new items, our model can make use of the two factors to make prediction, although they do not exist in the rating matrix. Experimental results on real dataset show that our model has a good scalability for new items, while still performing better than other state-of-art models.