Int J Performability Eng ›› 2018, Vol. 14 ›› Issue (9): 2116-2126.doi: 10.23940/ijpe.18.09.p20.21162126

Previous Articles     Next Articles

An Improved Tensor Decomposition Model for Recommendation System

Wenqian Shang*, Kaixiang Wang, and Junjie Huang   

  1. School of Computer Science, Communication University of China, Beijing, 100024, China
  • Revised on ; Accepted on
  • Contact: * E-mail address:

Abstract: With the arrival of the large data age, the algorithm of traditional recommendation systems cannot fully excavate the context information of the user’s decision-making and cannot provide a satisfactory recommendation for users. With the development of the label system, it becomes a hot topic to use multidimensional context-aware data to provide an accurate recommendation for users. At present, the more advanced scheme is to use the recommendation algorithm based on tensor decomposition to excavate the three-element relationship group of user-item-label. This paper proposes K-Means and the Time-Context based Tensor Decomposition Model (KTTD). The initial clustering of datasets is carried out through K-Means to improve the data aggregation and algorithm efficiency. The time context of the situation recommendation is excavated, and the implicit feedback in the temporal context perception is used as a dimension of tensor to establish the tensor decomposition model to improve the efficiency and quality of the recommendation. At the end of the paper, we verified the model by experiments, and the results show that the improved algorithm is better than the traditional recommendation algorithm in the accuracy of the recommended system.

Key words: recommendation system, tensor decomposition, K-Means, time context, matrix decomposition