摘要：

Abstract:

【Purpose/significance】Since the current works of point of interest recommendation exist sparsity and low accuracy, this pa⁃

per proposed a point of interest recommendation model combining potential check-ins and multi-type contextual information.【Meth⁃

od/process】The model is divided into two stages of matrix filling and matrix decomposition. Firstly, established a matrix filling model

by using social and geographic information to learn a set of points of interest for each user. Secondly, the classification information is

embedded into the weighted matrix decomposition model to learn user preferences. Finally, the adaptive kernel density estimation is

used to model the geographical influence, and the GSC-WMF model is obtained by combining the results of matrix decomposition.

【Result/conclusion】The experimental results show the proposed model outperforms state-of-the-art recommendation models in terms

of precision and recall.【Innovation/limitation】Matrix filling model is proposed to explore the potential interest, and the category infor⁃

mation is effectively fused to address the problem of implicit information feedback. In the future research work, we will introduce the

deep learning to improve recommendation performance.