Automated urban location annotation on mobile records

Location information is becoming much more important than ever before, especially in mobile services. Being widespread, less cost of energy and almost free for collecting data make mobile phone a perfect location sensor probe. Meaningful location name rather than digital coordinates could provide much more valuable information. In this paper, we develop a location semantic predicting method referred to Location Annotation(LA) which can automatically annotate meaningful base stations of phone users with semantic tags such as “home”, “work place” and “club”. We extract several explicit features from phone records and spatial-temporal patterns of mobile phone users to build an annotation model based on Maximum Entropy Model. Then a machine learning method is presented to estimate the best configuration of parameters in the model. Finally, comprehensive experiments demonstrate good performance of our method. Overall accuracy is about 90% which outperforms simple and traditional classification methods by 10+%. Semantic location names are valuable to urban planning and optimization, transportation management and land use planning.