MRTensorCube: tensor factorization with data reduction for context-aware recommendations
- PDF / 1,449,022 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 22 Downloads / 191 Views
MRTensorCube: tensor factorization with data reduction for context-aware recommendations Svetlana Kim1 Yong-Ik Yoon1
· Suan Lee2 · Jinho Kim2 ·
© The Author(s) 2017. This article is an open access publication
Abstract Context information can be an important factor of user behavior modeling and various context recognition recommendations. However, state-of-the-art context modeling methods cannot deal with contexts of other dimensions such as those of users and items and cannot extract special semantics. On the other hand, some tasks for predicting multidimensional relationships can be used to recommend context recognition, but there is a problem with the generation recommendations based on a variety of context information. In this paper, we propose MRTensorCube, which is a largescale data cube calculation based on distributed parallel computing using MapReduce computation framework and supports efficient context recognition. The basic idea of MRTensorCube is the reduction of continuous data combined partial filter and slice when calculating using a four-way algorithm. From the experimental results, it is clear that MRTensor is superior to all other algorithms. Keywords Context awareness · Tensor data cube · MapReduce framework
B
Yong-Ik Yoon [email protected] Svetlana Kim [email protected]; [email protected] Suan Lee [email protected] Jinho Kim [email protected]
1
Department of Multimedia Science, Sookmyung Women’s University, Chungpa-Dong 2-Ga, Yongsan-Gu, Seoul 140-742, Korea
2
Department of Computer Science, Kangwon National University, Chuncheon, Kangwon, Korea
123
S. Kim et al.
1 Introduction Recommendation systems are powerful techniques by providing users behavior recommendation of various types of potentially interesting products and service. The system’s ability to gather information has been enhanced, and recommendation systems based on contextual modeling approach have become popular [1–4]. Most systems require to analyze and manage a large amount of large volume data such as Web and social contextual information over multidimensional. It is becoming increasingly important to make a rapid analysis of large context datasets to enable rapid decision making. Recent research has focused on integrating contextual information on user-itemcontext (User × Item × Context) and building multidimensional models based on tensor decomposition model [5,6]. The tensor factorization is a multidimensional array of conventional matrix factorization techniques by considering interactions between users, items, and context. Many recommendation systems based on tensor decomposition techniques have been proposed to better support situation recognition recommendations and other related analysis tasks [7–9]. However, since the previous research still operates in main memory, it cannot be extended to compute large tensor data. Each of the multidimensional data holds values aggregated by all possible combinations of dimensional attributes. It takes a lot of time to compute the data cube. In this paper, we propose MRTens
Data Loading...
