Location: PSH (Professor Stuart Hall Building) - 302,
Goldsmiths, University of London, Building 2
Campus Map Contributors:
- Tao Cheng, SpaceTimeLab, University College London
- Juntao Lai, University College London
- Jianan Shen, University College London
Background:
Profiling place is a picture of the social, economic and environmental facets of a place, which describe how people see, hear and more importantly interact with the spaces they inhabit. For business intelligence, profiling places is extremely useful for understanding market potential for all products and services and finding the ideal locations for new stores, franchises and service centres.
Objective:
Most existing methods profile places based upon on a static and/or aggregated view of the place, ignoring its detailed and dynamic nature. For example, the areas around London tube stations are typical considered a place for transport. However, Victoria tube station of London is a traffic hub in the morning peak and evening peak, it is also a stations with several musical shows around, so the area is also a musical and culture place in addition to transport. King’s Cross mainly serves the transport for rush peaks, while Leicester Square is quiet in the morning, but busy in the evening since people go there for dinner and theatre activities. Therefore, to fully grasp the uniqueness of a place, the dynamic nature of the place should be considered and used in profiling places so that the subtle difference between places could be appreciated. The objective of this paper is to develop an innovative method to profile places based upon social media data.
Methods:
A space-time approach will be developed to profile places. The space-time profiles of places will be extracted based upon geotagged Tweets in those places. Here we use the areas around London tube stations as the case study. First, topics of key interests of people around the tube stations are extracted based upon LDA topic modelling. Then, the space-time profile of the tube station areas are built as composition of the multiple topics in different time periods. Last, two clustering techniques (K-menas and hierarchical clustering) are applied to group the tube stations based upon the space-time profiles. Each clustered group represents a unique profile of several tube station areas, which has been validated with the ground truth.
Results:
The comparison between the cluster results and ground truths shows that the stations were allocated into groups reasonably. Stations having strong and specific characteristics could be easily noticed, which is one of the advantages of clustering. For example, the area of stations close to 8 football clubs (stadiums) are all clustered as one group, so are London Heathrow Terminals stations.
Future Work:
As many places may have multiple functions changing with time, the approach developed here could identify locations with similar characteristics on the basis of topic distributions by various time periods, and give more accurate interpretations of the places. It could be helpful to understand and manage places in groups instead of individuals, benefiting for a variety of applications, such as advertising and retailing. We will conduct the work for city-size area with longer period of data in order to test the scalability of the method.
