Research Area:  Machine Learning
Traditional recommendation techniques are hindered by the simplicity and sparsity of user-item interaction data and can be improved by introducing auxiliary information related to users and/or items. However, most studies have focused on a single typed external relationship and not fully utilized the latent relationships among users and items. In this paper, we propose a heterogeneous network embedding-based recommendation method called HetNERec. Specifically, we first construct the co-occurrence networks by extracting multiple co-occurrence relationships from a recommendation-oriented heterogeneous network. We then propose an integration function to integrate multiple network embedded representations into a single representation to enhance the recommendation performance. Finally, the matrix factorization is extended by integrating the embedded representations and considering the latent relationships among users and items. The experimental results on real-world datasets demonstrate that the proposed HetNERec outperforms several state-of-the-art recommendation methods.
Author(s) Name:  Zhongying Zhao,Xuejian Zhang,Hui Zhou,Chao Li,Maoguo Gong,Yongqing Wang
Journal name:  Knowledge-Based Systems
Publisher name:  Elsevier
Volume Information:  Volume 204, 27 September 2020, 106218
Paper Link:   https://www.sciencedirect.com/science/article/abs/pii/S0950705120304317