Learning Drifting User Interest Incrementally from Numerically Labeled Feedbacks

Incremental approaches learn drifting user interests mainly from user feedbacks. Most of those existing approaches assume that data instances in user feedbacks are binary labeled. This paper presents a novel incremental learning approach that learns drifting user interests from numerically labeled feedbacks instead of binary labeled ones. The approach models user interests as a set of probabilistic concepts, considers numerical instance labels as probabilities that the user likes those instances, and uses feedbacks to update user interest models incrementally. Experimental results on different learning tasks show that the approach outperforms existing approaches in numerically labeled feedback environment.