Heterogeneous Context-aware Recommendation Algorithm with Semi-supervised Tensor Factorization

Data sparsity is one of the most challenging problems in recommender systems. In this paper, we tackle the data sparsity problem by proposing a heterogeneous context-aware semi-supervised tensor factorization method named HASS. Firstly, heterogeneous context are classified and processed by different modeling approaches. We use a tensor factorization model to capture user-item interaction contexts and use a matrix factorization model to capture both user attributed contexts and item attributed contexts. Secondly, different context models are optimized with semi-supervised co-training approach. Finally, the two sub-models are combined effectively by an weight fusing method. As a result, the HASS method has several distinguished advantages for mitigating the data sparsity problem. One is that the method can well perceive diverse influences of heterogeneous contexts and another is that a large number of unlabeled samples can be utilized by the co-training stage to further alleviate the data sparsity problem. The proposed algorithm is evaluated on real-world datasets and the experimental results show that HASS model can significantly improve recommendation accuracy by comparing with the existing state-of-art recommendation algorithms.