Supervised-LDA: A probabilistic topic model for collaborative filtering

Collaborative filtering, which identifies and recommends interest items to users based on the interest groups of other users, has received significant interest recently. In this paper, we propose a supervised LDA model (Supervised-LDA) for collaborative filtering. Supervised-LDA can deal with document collections where each document is accompanied by a ratting variable. By modeling the relationship among words in a document and the rating for the document directly, Supervised-LDA can generate an item list with the highest ratings for each latent topic. Moreover, Supervised-LDA can obtain the contributions of words in vocabulary, which can be used to predict the ratings of unseen items. Experimental results on real world data set show that the proposed model can address the collaborative filtering task effectively.