GeoLife: A Collaborative Social Networking Service Yu Zheng, Xing Xie and Wei-Ying Ma

1. GeoLife: A Collaborative Social Networking Service Yu Zheng, Xing Xie and Wei-Ying Ma IEEE Data Eng. Bull. (DEBU) 33(2):32-39 (2010) Paper presentation by Sowmya Ramesh 02/16/2010

2. GeoLife • GeoLife is a GPS-data-driven social networking service where people can share life experiences and connect to each other with their location histories

3. Why GeoLife? • Existing applications allow users to upload, share and browse GPS track logs and associated multimedia content over Web maps enabling user to share their life experience among people. So GPS data is being used directly without much understanding • GeoLife focuses on visualization, organization, fast retrieval, and effective understanding of GPS track logs for both personal and public use. It helps user to understand a person’s past experience from GPS data • A step towards integrating social networking into GIS

4. Architecture of GeoLife

5. Key Applications • Sharing life experiences based on GPS trajectories • Generic travel recommendations • Top interesting locations • Travel sequences among locations and • Travel experts in a given region • Collaborative location and activity recommendation • Personalized friend and location recommendation

6. Search Trajectories by Location • If a user is planning a trip to multiple places of interest in an unfamiliar city then GeoLife can provide user with similar routes traveled by other people • GeoLife uses trajectory query called the k Best-Connected Trajectory (k-BCT) query for searching trajectories by multiple geographical locations • Similarity function • Incremental k-Nearest Neighbor Algorithm (IKNN) using R-tree index : get the shortest distance from a query location to the trajectories

7. Search Trajectories by Location • Similarity Function: It reflects how close a trajectory is to the given locations, and we call the most similar trajectory the best-connected trajectory • Step 1. find out the closest trajectory point on R to each location qi • Step 2. sum up the contribution of each matched pair. (unordered query)

8. Search Trajectories by Location • Incremental k-NN Algorithm (IKNN) : Retrieves the nearest trajectory points with regard to each query location incrementally and examines the k-BCT from the trajectory points discovered so far • k-Best Connected Trajectory (k-BCT) query: Given a set of trajectories T = {R1, R2, … , Rn}, a set of query locations Q = {q1, q2, … ,qm}, and the similarity function Sim(Q, R), the k-BCT query is to find the k trajectories among T that have the highest similarity

9. Search Trajectories by Spatio-Temporal Queries • People are interested in some GPS trajectories showing the travel experiences within a particular geo-region and in a specific time interval E.g. : Where would be interesting in the downtown Beijing during Christmas? • Compressed Start-End tree: spatio-temporal index scheme which uses B+ tree index for frequently updated groups and sorted dynamic array for rarely updated ones. This requires less index space and less update cost

10. Learning transportation modes based on GPS Data • Classification of GPS trajectories by transportation modes so that smart route recommendations can be performed for a person based on his needs • Supervised learning to automatically infer user’s transportation modes • Approach consists of three parts : • Change point-based segmentation method • An inference model and • Graph-based post-processing algorithm

11. Learning transportation modes based on GPS Data

12. Change point-based segmentation method • Change point stands for a place where a user changes their transportation mode in a trajectory • From each segment, identify a set of sophisticated features and are fed to a generative inference model to classify the segments of different modes • Conduct graph-based post-processing to further improve the inference performance

13. Generic Travel Recommendation • This recommender provides a user with the top n experienced users (experts), interesting locations and the classical travel sequences among these locations • Interesting location : The interest level of a location can be calculated by the experiences of the users who have accessed this location • Experts : User’s travel experience can be represented by the interest levels of the visited locations

14. Tree-based hierarchical graph • Detect stay points • Formulate a tree-based Hierarchy using density-based clustering algorithm • Build graphs on each level

15. Hypertext Induced Topic Search • HITS-based model to rank users’ travel experiences and interest of a location within a region • Hub is a user who has accessed many places • An authority is a location which has been visited by many users • Users’ travel experiences (hub scores) and the interests of locations (authority scores) have a mutual reinforcement relation • Using user’s travel experiences and the interests of locations, calculate a classical score for each location sequence within the given geospatial region to identify the classical travel sequence

16. Collaborative Location and Activity Recommendation • Location recommendation given some activity query • Activity recommendation given some location query System Architecture

17. Collaborative Location and Activity Recommendation • Location-activity matrix • Rows as locations and columns as activities • Constructed from comments or tips added by users to a point location in a trajectory • Entry in the matrix denotes the frequency for the users to perform some activity on some location • Matrix is incomplete and very sparse

18. Location-Activity Extraction • Location-activity matrix GPS: “39.903, 116.391, 14/9/2009 15:25” Stay Region: “39.910, 116.400 (Forbidden City)” Tourism Food … Forbidden City “We took a tour bus to see around along the forbidden city moat …” Zhongguancun Activity: tourism … Location-Activity Matrix User comments are few -> this matrix is sparse! Our objective: to fill this matrix.

19. Collaborative Location and Activity Recommendation • Location-feature matrix • exploit the location features with the help of POI category database. Database is based on the city yellow pages • each entry of the matrix denotes some feature value on that location • Activity-activity matrix • exploit theWorldWideWeb to get the knowledge about the activity correlations • each entry of the matrix denotes the correlation between a pair of activities

20. Collaborative location-activity learning model • Fill those missing entries in the location-activity matrix with the information learned from the other two matrices • Collaborative filtering (CF) approach based on collective matrix factorization used • Based on the filled location-activity matrix, we can rank and retrieve the top k locations/activities as recommendations to the users

21. Location-activity relation 5 3 2 5 An entry denotes how popular an activity is performed at a location Ranking along the Columns or rows 4 1 Example: Tourism Exhibition Shopping Location recommendation Tourism: Forbidden City > Bird’s Nest > Zhongguancun Forbidden City Bird’s Nest Activity recommendation Forbidden City: Tourism > Exhibition > Shopping Zhongguancun

22. Personalized Friend & Location Recommendation • People who have similar location histories might share similar interests and preferences • Based on users’ GPS trajectories provide personalized friend & location recommendation

23. Conclusion • User-generated GPS trajectories do not only connect locations in the physical world but also bridge the gap between people and locations • GeoLife, aims to understand trajectories, users, and locations in a collaborative manner • Overall, user, location and trajectory have a collaborative and mutual reinforcement relationship among each other.

24. References • Yu Zheng, Like Liu, Longhao Wang, Xing Xie. Learning Transportation Modes from Raw GPS Data for Geographic Application on the Web, In WWW 2008 • Yu Zheng, Lizhu Zhang, Xing Xie, Wei-Ying Ma. Mining interesting locations and travel sequences from GPS trajectories. In WWW 2009 • Yu Zheng, Lizhu Zhang, Xing Xie. Recommending friends and locations based on individual location history. To appear in ACM Transaction on the Web, 2010 • Yu Zheng, Yukun Chen, Xing Xie, Wei-Ying Ma. GeoLife2.0: A Location-Based Social Networking Service In MDM 2009 • Zaiben Chen, Heng Tao Shen, Xiaofang Zhou, Yu Zheng, Xing Xie. Searching Trajectories by Locations -An Efficiency Study, In ACM SIGMOD 2010

25. Questions