Twin Bridge Transfer Learning for Sparse Collaborative Filtering.

Collaborative filtering (CF) is widely applied in recommender systems. However, the sparsity issue is still a crucial bottleneck for most existing CF methods. Although target data are extremely sparse for a newly-built CF system, some dense auxiliary data may already exist in othermatured related domains. In this paper,we propose a novel approach, Twin Bridge Transfer Learning (TBT), to address the sparse collaborative filtering problem. TBT reduces the sparsity in target data by transferring knowledge from dense auxiliary data through two paths: 1) the latent factors of users and items learned from two dense auxiliary domains, and 2) the similarity graphs of users and items constructed from the learned latent factors. These two paths act as a twin bridge to allow more knowledge transferred across domains to reduce the sparsity of target data. Experiments on two benchmark datasets demonstrate that our TBT approach significantly outperforms state-of-the-art CF methods. © Springer-Verlag 2013.