Abstract
We propose a novel recommendation method based on tree. With user behaviordata, the tree based model can capture user interests from coarse to fine, bytraversing nodes top down and make decisions whether to pick up each node touser. Compared to traditional model-based methods like matrix factorization(MF), our tree based model does not have to fetch and estimate each item in theentire set. Instead, candidates are drawn from subsets corresponding to user'shigh-level interests, which is defined by the tree structure. Meanwhile,finding candidates from the entire corpus brings more novelty thancontent-based approaches like item-based collaborative filtering.Moreover, inthis paper, we show that the tree structure can also act to refine userinterests distribution, to benefit both training and prediction. Theexperimental results in both open dataset and Taobao display advertisingdataset indicate that the proposed method outperforms existing methods.