TOPTRAC: Topical Trajectory Pattern Mining.

Area of Science:

• Data Science
• Computational Social Science
• Geographic Information Science

Background:

• Geo-tagging of media like photos and micro-blogging messages is increasingly popular due to GPS-enabled mobile phones.
• Existing methods for analyzing GPS data focus on movement patterns but often underutilize the latent topics within associated text messages.
• Understanding location-based user behaviors requires integrating both spatial and textual information.

Purpose of the Study:

• To develop a latent topic-based clustering algorithm for discovering patterns in geo-tagged text message trajectories.
• To propose a probabilistic model that captures semantic regions and movement patterns between them.
• To effectively utilize the latent topics in text messages alongside location data for behavior analysis.

Main Methods:

• A novel probabilistic model was developed to represent semantic regions and movement patterns.
• An efficient inference algorithm was created to estimate model parameters.
• A clustering algorithm was devised to identify frequent and significant movement patterns based on the probabilistic model.

Main Results:

• The proposed algorithm successfully discovers diverse and interesting trajectory patterns from real-life datasets.
• Semantic regions are identified with finer granularity compared to traditional geographical clustering methods.
• The integration of latent topics enhances the analysis of location-based user behaviors.

Conclusions:

• The developed latent topic-based clustering algorithm effectively analyzes geo-tagged text message trajectories.
• This approach offers a more nuanced understanding of user behavior by combining spatial and semantic information.
• The method provides a significant advancement in discovering complex movement patterns and defining semantic regions.