Rapid urbanization and modern civilization require sound integration with public transportation systems. In the same time, the volume and complexity of public transportation network are increasing, making it harder to understand the public transportation dynamics. As a first step, understanding the similarity among subway stations is imperative. In this paper, we proposed a semantic framework inspired from natural language processing (NLP) to interpret subway stations as compound words. Specifically, we transplanted context and literal meaning of compound words into mobility and location attributes of stations. Using smart card data, we trained stacked autoencoders (SAE) with designed flow matrices as an embedding method to learn the mobility attributes. Subsequently, to discover the location attributes, we have applied affinity propagation clustering to classify 9 point of interest (POI) categories. Combined with urban planning knowledge, we manage to comprehend the land use meanings of 9 POI clusters. The location semantics is chosen from those categories reflecting its urban land use pattern. By choose meaningful combination of mobility and location semantics for stations’ similarity case studies, we summarized potential applications of this semantic framework.

快速的城市化和现代文明要求与公共交通系统进行良好的整合。同时，公共交通网络的数量和复杂性也在增加，这使得人们难以理解公共交通的动态。首先，必须了解地铁站之间的相似性。在本文中，我们提出了一个受自然语言处理（NLP）启发的语义框架，用于将地铁车站解释为复合词。具体来说，我们将复合词的上下文和字面意思移植到站点的移动性和位置属性中。使用智能卡数据，我们使用设计的流量矩阵作为一种学习移动性属性的嵌入方法来训练堆叠式自动编码器（SAE）。随后，要发现位置属性，我们已应用亲和力传播聚类对9个兴趣点（POI）类别进行分类。结合城市规划知识，我们设法了解了9个POI集群的土地使用含义。从反映城市土地使用模式的那些类别中选择位置语义。通过为站点的相似性案例研究选择移动性和位置语义的有意义的组合，我们总结了此语义框架的潜在应用。