Applying HOSVD to Alleviate the Sparsity Problem in Context-aware Recommender Systems

It is quite a great challenge for Contextaware recommender systems (CARS) to generate accurate recommendations with only a few available none-zero contextual user preferences. This paper presents a new approach to alleviate this high sparsity problem by applying the Higher order singular value decomposition (HOSVD) technique. Firstly, it constructs an N-order tensor to represent multidimensional contextual user preferences and decompose it into (N-2) 3-order tensors according to different types of context (such as time, location and activity). Secondly, it introduces HOSVD to automatically discover the latent associations among these different 3dimensional objects and predicts unknown unidimensional contextual user preferences. Finally, it calculates every contextual influence coefficient that each type of context factor influences user preferences and then constructs a new N-order tensor using weighted linearization method in order to provide recommendations. Experimental evaluation on a simulated personalized mobile services environment demonstrates the efficacy of our approach against the other baseline ones.