Location-based topic evolution

  • Authors:

  • Haiqin Yang;Shouyuan Chen;Michael R. Lyu;Irwin King

  • Affiliations:

  • The Chinese University of Hong Kong, Hong Kong, Hong Kong;The Chinese University of Hong Kong, Hong Kong, Hong Kong;The Chinese University of Hong Kong, Hong Kong, Hong Kong;The Chinese University of Hong Kong, Hong Kong, Hong Kong & AT&T Labs Research, San Francisco, CA, USA

  • Venue:
  • Proceedings of the 1st international workshop on Mobile location-based service
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

As the advance of mobile technologies, geographical records can be easily embedded in the data to form the location-associated documents. For example, in Twitter, the location of tweets can be identified by the GPS locations or IP addresses from smart phones. In Flickr, photos may be tagged and recorded with GPS locations. With the geographical information, it is more likely to model users' interests in different regions so as to determine the corresponding marketing strategy. Due to its potential in providing personalized and context-aware services, several pieces of work have started to explore in this area. One stream of work tries to discover users' interest topics from location-associated documents. These models work under the assumption that words close in geographical positions are likely to be clustered into the same geographical topic. However, they attain this in a static mode. That is, they do not consider the evolution of the topics. In addition, they have to specify the total number of topics for the corpus in advance. In order to utilize the geographical information and to model the change of topics, we propose a location-based topic evolution (LBTE) model to tackle the above issues. Main advantages of our model lie that it can reveal the appearance and disappearance of the topics in different regions. Moreover, topics can be automatically determined based on the location-associated documents and its total number is not restricted to a preset value. Finally, we conduct a series of experiments on both synthetic and real-world datasets to demonstrate the merits of our proposed LBTE model in capturing users' interest topics.