POI Type Matching based on Culturally Different Datasets

The development of mobile social media networks has changed our daily life. More and more people tend to share their locations, emotions and activities with their friends, which are called ‘check-in records’. These geo-tagged data create an unprecedented opportunity for researchers to reveal spatio-temporal activity patterns of citizens [1,2], capture usages of urban public facilities [3], and understand the interactions between citizens and urban built environment. Points of interest (POI) data play an important role when analyzing check-in records, because they contain categories like restaurant, school to indicate people’s activities. Considering the existing various social media data sources, POI type matching is a basic problem if we intent to do research on two or more check-in datasets with different type schemata, which is also a major challenge for POI data conflation in existing map navigation products like Apple Maps. Currently, typical approaches POI type matching focuses on comparing types through schema. For example Schema.org is often used as an organized vocabulary representing a hierarchy of place types (e.g., a Restaurant is a type of Food Establishment). Much of the ongoing work in assessing the similarity of POI types happens at a top-down schema level rather than a bottom-up data-driven approach. This work is very much focused on a bottom-up approach to assessing the similarity between POI types in two culturally different datasets adhering to two different schemata. Much existing literature has found that the temporal population variation, named ‘temporal signature’, of POIs has strong relationship with the category of POIs [3], but there are few attempts to take the temporal signatures of POIs into consideration when making POI type matching. In this work, we try to match points of interest between culturally (and linguistically) different datasets based on the temporal signatures. We conducted an experiment on two different location-based social networking check-in datasets from different countries. Both linguistic meaning and temporal signatures of POI types were adopted to address the POI type matching problem. We summarized four kinds of POI type matching patterns and measured the similarity between matching POI types, which supports the aforementioned research hypothesis.