广东省商业业态时空演变特征及影响因素研究——基于POI大数据的可解释机器学习模型

Research on the Spatiotemporal Evolution Characteristics and Influencing Factors of Commercial Formats in Guangdong Province—An Interpretable Machine Learning Model Based on POI Big Data

摘要: 本文基于2019、2021、2023年的POI数据，以广东省21个城市为研究对象，采用空间核密度估计法、Theil指数、可解释机器学习模型等研究方法，探讨2019~2023年期间广东省商业业态空间分布发展格局变化，分析不同商业业态空间分布的影响因素。研究表明：1) 广东省商业空间布局呈现珠三角地区多中心连绵集聚、外围县域中心点状集聚的多中心空间发展格局。2019~2023年期间，广东省商业中心等级体系呈扁平化发展趋势；2) 2019~2023年期间，广东省不同商业业态网点规模结构发生动态调整，其中购物服务、餐饮服务网点规模先升后降，休闲娱乐网点规模逐步增长的态势；3) 2019~2023年期间，广东不同类型商业业态的时空演变趋势存在差异，购物服务网点空间集聚程度明显下降，餐饮服务网点集聚程度呈现先上升后下降，休闲娱乐网点空间集聚程度则不断提升；4) Theil指数分析表明，广东省商业总体及三大细分业态网点数量的区域差异呈现缩小态势，组间差异明显大于组内差异；5) 基于SHAP方法的可解释机器学习模型分析发现：人均GDP、人口规模、人均消费支出、工业总产值、快递规模、人口密度、旅游人数等变量对商业及三类细分业态均有正向促进作用，但不同因素对不同商业业态的影响程度存在异质性和非线性特征，人均GDP、人均消费支出、人口规模等变量与对应SHAP值呈现出较为近似的线性正相关关系，其他变量的非线性特征较为明显。

Abstract: Based on POI data from 2019, 2021, and 2023, this paper takes 21 cities in Guangdong Province as the research object and employs research methods such as spatial kernel density estimation, the Theil index, and interpretable machine learning models to explore the changes in the spatial distribution development pattern of commercial formats in Guangdong Province from 2019 to 2023, and analyze the influencing factors of the spatial distribution of different commercial formats. The research findings are as follows: 1) The spatial layout of commerce in Guangdong Province exhibits a multi-center spatial development pattern characterized by multi-center contiguous agglomeration in the Pearl River Delta region and point agglomeration of county-level centers in the periphery. From 2019 to 2023, the hierarchy of commercial centers in Guangdong Province showed a trend of flattening development; 2) During this period, the scale structure of outlets for different commercial formats in Guangdong Province underwent dynamic adjustments. Specifically, the scale of shopping service and catering service outlets first increased and then decreased, while the scale of leisure and entertainment outlets gradually grew; 3) The spatiotemporal evolution trends of different types of commercial formats in Guangdong Province varied during 2019~2023. The degree of spatial agglomeration of shopping service outlets decreased significantly, that of catering service outlets first increased and then decreased, and the spatial agglomeration of leisure and entertainment outlets continued to rise; 4) Theil index analysis indicated that the regional disparities in the number of outlets of overall commerce and the three major sub-sectors (shopping, catering, and leisure/entertainment) in Guangdong Province showed a narrowing trend, with inter-group disparities significantly larger than intra-group disparities; 5) Analysis using an interpretable machine learning model based on the SHAP method revealed that variables such as per capita GDP, population size, per capita consumption expenditure, gross industrial output value, number of express parcels, population density, and number of tourists had positive promotional effects on both overall commerce and the three sub-sectors. However, the influence of different factors on various commercial formats exhibited heterogeneity and non-linear characteristics. Variables such as per capita GDP, per capita consumption expenditure, and population size showed an approximately linear positive correlation with their corresponding SHAP values, while non-linear characteristics were more pronounced for other variables.