Discovering important locations from massive and low-quality cell phone trajectory data

Important locations mainly refer to the places where people spend much time in the daily life, including their home and working places. The development and popularization of smart cell phones bring great convenience to people's daily life. Besides making calls and surfing the Internet, the logs generated when visiting the base stations also contribute to users' pattern mining, such as important location discovery. However, it's challenging to deal with such kind of trajectory data, due to huge volume, data inaccuracy and diversity of cell phone users. In this research, a general framework is proposed to improve the usability of trajectory data. The framework includes a filter to improve data usability and a model to produce the mining results. Two concrete strategies, namely GPMA (grid-based parallel mining algorithm) and SPMA (station-based parallel mining algorithm), can be embedded into this framework separately. Moreover, three optimization techniques are developed for better performance: (1) a data fusion method, (2) an algorithm to find users who have no work places, and (3) an algorithm to find people who work at night and fix their important locations. Theoretical analysis and extensive experimental results on real datasets show that the proposed algorithms are efficient, scalable, and effective.