基于轨迹活动语义挖掘的个体社会经济水平评估

Individual Socio-Economic Level Assessment Based on Trajectory Activity Semantics

个体社会经济水平评估对于商业决策、城市规划和公共卫生具有重要的应用价值。但现有方法多依赖定位数据和呼叫详单记录构建出行位置和手机业务特征集合,未充分考虑个体出行的语义上下文,难以从动机与需求层面理解出行行为,导致建模过程可解释性不足。为此,本文提出一种基于轨迹活动语义挖掘的个体社会经济水平评估方法,通过显式提取居住、购物、餐饮、娱乐、消费喜爱度与探索欲6类消费模式,从消费能力与意愿角度刻画个体社会经济水平,提高评估方法的可解释性。(1)通过网格化的语义地图为停留点赋予出行语义上下文,并划分居住、购物、餐饮、娱乐4类活动的停留点集合;(2)计算4类活动的时间熵、旋转半径和活动区域经济水平等时空语义特征,并通过结构方程模型筛选特征计算各类消费模式价值;(3)使用极端随机森林决策个体社会经济水平。本文基于深圳市635名个体2019年4—11月的私家车轨迹数据开展实验,通过核心商圈、劳动密集型工厂、高档住宅与城中村等典型场景筛选高低社会经济水平人群,验证了方法有效性;此外,对高低社会经济水平群体的出行时空分布和工作强度开展可视化分析,探讨了群体间的出行模式差异。本文方法可为人地交互视角下的人口统计属性建模提供参考。

Assessment of individual Socio-Economic Levels(SEL)is crucial for business decisions,urban planning,and public health.However,current methods highly rely on location data and call detail records to construct travel locations and mobile business features,which is inadequate to represent the semantic context of individual travel,and fail to understand the motivations and demands of travel activities.Consequently,it makes the modeling process lack interpretability.To address aforementioned issue,this paper proposes a novel assessment method of individual socioeconomic levels based on the analysis of trajectory activity semantics.It models individual socio-economic levels from the perspectives of consumption ability and willingness by explicitly extracting six consumption patterns including residence,shopping,dining,entertainment,consumption preferences and exploration,thereby enhancing the interpretability of the assessment method.Specifically,①Stay points extracted from trajectories are categorized into four types of activities,including residence,shopping,dining,and entertainment,by tagging semantic context through a grid-based semantic map;②Spatiotemporal and semantic features such as temporal entropy,gyration radius,and economic level of activity areas,are calculated for the four activities respectively.We then employ the structural equation model to select appropriate features for measuring the values of consumption patterns;③Extreme random forest is utilized to assess individual socio-economic levels using the values of six consumption patterns,which is calculated based on the economic levels of regions where an individual stays in the travel activities,as well as the preferences for visiting these regions.We use GPS trajectories of 635 anonymous private car drivers in Shenzhen city of China from April to November in 2019 as experimental data,and assess individual socio-economic levels for each driver.The effectiveness of the proposed method is validated by selecting representative individuals with high and low socio-economic levels from five typical scenarios i.e.,central business districts,labor-intensive factories,premium residences,and urban villages,which demonstrates alignment between the calculated socio-economic levels of individuals and the depicted value of the scenarios.Besides,we analyze the spatiotemporal distribution and work intensity of different socio-economic level groups,and explore their differences in travel patterns.The findings indicate that individuals with a higher socioeconomic level tend to have more flexible morning commutes,and exhibit a smoother travel distribution in the afternoon.It also presents a more concentrated spatial distribution in terms of their activity areas,which is consistent with the urban structures of Shenzhen.In summary,the proposed method can provide a reference for modeling demographic characteristics of individuals from the perspective of human-environment interaction.