A case study for Beijing point of interest data using group linked Cox process

We develop in this article a group linked Cox process model for analyzing point of interest (POI) data. We focus on a Beijing POI dataset, which contains more than 22 thousand POIs in Beijing urban area. These POIs have been divided into many small categories (e.g., restaurants, movie theaters, hospitals, universities and subway stations) by the digital map maker (e.g., Baidu Map). Empirical analysis provides substantial evidence that POIs across different categories could be highly correlated so that those small categories can be further grouped. To this end, we develop here a group linked Cox process model. Specifically, within each group, we model POI locations by a standard Cox process so that the POI clustering effect can be well described. Furthermore, the idea of bivariate linked Cox process is borrowed and further extended to its multivariate counterpart. Consequently, a more significant number of POI categories can be accommodated within each group. To estimate the model, a minimum contrast type method is developed, and an automatically grouping method is provided. Simulation studies are conducted to validate the proposed methodology. At last, we apply our method to the aforementioned real dataset, and a total of 4 groups are uncovered. This leads to the discovery of some urban-planning-related features.

Keywords

Cox process, group linked Cox process, location based service, point of interest

Full Text (PDF format)

The research of Rui Pan is supported by National Nature Science Foundation of China (NSFC, 11601539, 11631003, 71771224), the Fundamental Research Funds for the Central Universities (QL18010), and China’s National Key Research Special Program Grant 2016YFC0207704.

The research of Rong Guan is supported in part by National Natural Science Foundation of China (NSFC, 71401192, 71873012, 71771204), the Fundamental Research Funds for the Central Universities (QL18009), and the Program for Innovation Research in Central University of Finance and Economics (011650317002).

Hansheng Wang’s research is partially supported by National Natural Science Foundation of China (No. 11831008, 11525101, 71532001, 71332006). It is also supported in part by China’s National Key Research Special Program (No. 2016YFC0207704).

Received 31 October 2017

Published 11 March 2019