This document proposes methods for mining interesting locations and travel sequences from GPS trajectory data. It presents a system that first models user location histories through stay point detection and hierarchical clustering. It then infers location interest and user travel experience using a HITS-based model that considers the mutual reinforcement between users and locations. Experiments on a real GPS dataset show the proposed approach outperforms baselines in ranking interesting locations and classical travel sequences. The document also discusses related work and potential future directions.

1. ining Interesting Locations
and Travel Sequences from
GPS Trajectories
Yu Zheng, Lizhu Zhang, Xing Xie, Wei-Ying Ma
Microsoft Research Asia
M
Johnson Chin-Hui Chen
20090923 Seminar

4. Agenda
! Introduction
! Overview of the System
! Modeling Location History
! Location Interest Inference
! Experiments
! Related Work
! Future Work

5. INTRODUCTION
! GPS-enabled devices, like GPS-phones, are
changing the way people interact with the Web by
using locations as contexts.
! Users record their outdoor movements because…
! Travel experience sharing
! Life logging
! Sports activity analysis
! Multimedia content management

6. INTRODUCTION
! Websites or forums: geo-related Web communities
! Bikely
(www.bikely.com)
! GPS exchange
(www.gpsxchange.com)
! @trip
(www.a-trip.com)
!
(map.answerbox.net)

7. INTRODUCTION
! Although there are many raw GPS data…
! Without much understanding
! It’s impossible to browse each GPS trajectory one by one

8. INTRODUCTION
! Goal :
! Mine the top n interesting locations
! Mine the top m classical travel sequences
! Mine the most k experienced users in a geo-related
community
Culturally important places
(Statue of Liberty in NY)
Commonly frequented public areas
(shopping streets)

9. INTRODUCTION
! Difficulty :
! What is a location? (geographical scales)
! The interest level of a location
! not only frequency or counts
! but also lie in these users’ travel experiences
! How to determine a user’s travel experience?
! The location interest and user travel
! are region-related
(conditioned by the given geospatial region)
! are relative value (Ranking problem)
(not reasonable to judge whether or not a location is interesting)

10. Agenda
! Introduction
! Overview of the System
! Modeling Location History
! Location Interest Inference
! Experiments
! Related Work
! Future Work

11. OVERVIEW OF THE SYSTEM
! Preliminary
! Clarify some terms
! Architecture
! Application
! GeoLife 2.0 since Oct. 2007

12. OVERVIEW OF THE SYSTEM
Preliminary
! GPS logs P and GPS trajectory
! Stay points S = {s1, s2,…, sn}.
! P = {pm,pm+1,…,pn} is a group of consecutive GPS points
S.lat = avg lat of P S.arvT = pm.T
S.lngt = avg lngt of P S.levT = pn.T
p4
p3
p5
p6
p7
AStay Point S
p1
p2
Latitude, Longitude, Time
p1: Lat1, Lngt 1, T1
p2: Lat2, Lngt 2, T2
………...
pn: Latn, Lngtn, Tn

13. OVERVIEW OF THE SYSTEM
Preliminary
! Location history :
! represented by a sequence of stay points
! with transition intervals
! Tree-Based Hierarchy H :
! H = (C,L)
! L = {ʅ1 , ʅ2 , … , ʅn}
! C = {Cij| }
! jth cluster on level ʅi
! Ci : level ʅi clusters
𝐿�𝑜�𝑐�𝐻� = (𝑠�1
∆𝑡�1
𝑠�2
∆𝑡�2
,…,
∆𝑡�𝑛�− 1
ǦǦ 𝑠�𝑛� )
Stands for a stay point S
Stands for a stay point cluster cij
{C }
High
Low
Shared Hierarchical Framework
c10
c20 c21
c30 c31 c32 c33 c34

14. OVERVIEW OF THE SYSTEM
Preliminary
! Tree-Based Hierarchical Graph (TBHG)
! TBHG = (H,G)
! H = Tree-Based Hierarchy
! G={gi = (Ci,Ei), } l1
G3
G1
G2
c30
c31
c32
c33
c34
c20
c21 l2
l3
|L|i<1 ≤

15. OVERVIEW OF THE SYSTEM
! Preliminary
! Clarify some terms
! Architecture
! Application
! GeoLife 2.0 since Oct. 2007

16. OVERVIEW OF THE SYSTEM
Architecture
Offline
Offline

17. OVERVIEW OF THE SYSTEM
! Preliminary
! Clarify some terms
! Architecture
! Application
! GeoLife 2.0 since Oct. 2007

18. OVERVIEW OF THE SYSTEM
Application

19. OVERVIEW OF THE SYSTEM
Application

20. Agenda
! Introduction
! Overview of the System
! Modeling Location History
! Location Interest Inference
! Experiments
! Related Work
! Future Work

21. MODELING LOCATION HISTORY

22. MODELING LOCATION HISTORY
GPS Logs of
User 1
GPS Logs of
User 2
GPS Logs of
User n
GPS Logs of
Useri
GPS Logs of
User i+1
GPS Logs of
Usern-1
Stands for a stay point S
Stands for a stay point cluster cij
{C }
High
Low
c10
c20 c21
c30 c31 c32 c33 c34
1. Stay point detection
2. Hierarchical clustering
l1
G3
G1
G2
c30
c31
c32
c33
c34
c20
c21 l2
l3
3.Graph Building

23. GPS Logs of
User 1
GPS Logs of
User 2
GPS Logs of
User n
GPS Logs of
Useri
GPS Logs of
User i+1
GPS Logs of
Usern-1
Stands for a stay point S
Stands for a stay point cluster cij
{C }
High
Low
c10
c20 c21
c30 c31 c32 c33 c34
1. Stay point detection
2. Hierarchical clustering
l1
G3
G1
G2
c30
c31
c32
c33
c34
c20
c21 l2
l3
3.Graph Building

24. MODELING LOCATION HISTORY
𝐿�𝑜�𝑐�𝐻� = (𝑠�1
∆𝑡�1
𝑠�2
∆𝑡�2
,…,
∆𝑡�𝑛�− 1
ǦǦ 𝑠�𝑛� )

25. Agenda
! Introduction
! Overview of the System
! Modeling Location History
! Location Interest Inference
! Experiments
! Related Work
! Future Work

26. LOCATION INTEREST INFERENCE

27. LOCATION INTEREST INFERENCE
! 1. Basic concepts of HITS
! 2. HITS-Based Inference Model
! 3. Mining Classical Travel Sequences

28. LOCATION INTEREST INFERENCE
Basic concepts of HITS
! A search-query-dependent ranking algorithm.
! query -> a list -> Hub/Authority ranking

29. LOCATION INTEREST INFERENCE
! 1. Basic concepts of HITS
! 2. HITS-Based Inference Model
! 3. Mining Classical Travel Sequences

30. LOCATION INTEREST INFERENCE
HITS-Based Inference Model
l1
G3
G1
G2
c30
c31
c32
c33
c34
c20
c21 l2
l3
Users:
Hub nodes
Locations:
Authority nodes
Mutual reinforcement relationship
A user with rich travel knowledge are more likely to visit
more interesting locations.
A interesting location would be accessed by many users with
rich travel knowledge.

31. LOCATION INTEREST INFERENCE
HITS-Based Inference Model
! Difficulty : region-related
! aligned with the query-dependent property of HITS
! But online selection is time consuming …
! Using the regions specified by their ascendant clusters
! a location have multiple authority scores based on the different
region scales it falls in.
! a user have multiple hub scores conditioned by the regions of
different clusters.
l1
G3
G1
G2
c30
c31
c32
c33
c34
c20
c21 l2
l3

32. LOCATION INTEREST INFERENCE
HITS-Based Inference Model
! Location Interest :
! Authority scores (cij)
: the auth scores of cij based on the region specified by
its ascendant nodes on level ,where
! User Travel Experience :
! Hub scores ( )
Stands for a stay point S
Stands for a stay point cluster cij
{C }
High
Low
Shared Hierarchical Framework
c10
c20 c21
c30 c31 c32 c33 c34

33. LOCATION INTEREST INFERENCE
HITS-Based Inference Model
33
{C }
Ascendant
Stands for a stay point cluster cij
{C }
Descendant
A region specified by a user
Stands for a cluster that covers the region specified by the user
c35c31 c32 c33 c34 c35c31 c32 c33 c34
A) A region covering locations
from single parent cluster
B) A region covering locations
from multiple parent clusters
c11
c22c21
c11
c22c21{C }
Ascendant
Stands for a stay point cluster cij
{C }
Descendant
A region specified by a user
Stands for a cluster that covers the region specified by the user
c35c31 c32 c33 c34 c35c31 c32 c33 c34
A) A region covering locations
from single parent cluster
B) A region covering locations
from multiple parent clusters
c11
c22c21
c11
c22c21

34. LOCATION INTEREST INFERENCE
HITS-Based Inference Model
! Inference :
! Build adjacent matrix M
!
! : uk has visited cluster cij

35. LOCATION INTEREST INFERENCE
HITS-Based Inference Model
! Mutual reinforcement relationship (matrix form):
! Conditioned by the region of cluster C11

36. LOCATION INTEREST INFERENCE
! 1. Basic concepts of HITS
! 2. HITS-Based Inference Model
! 3. Mining Classical Travel Sequences

37. LOCATION INTEREST INFERENCE
Mining Classical Travel Sequences
37
• Three factors determining the classical score :
– Travel experiences (hub scores) of the users taking the sequence
– The location interests (authority scores) weighted by
– The probability that people would take a specific sequence
: Authority score of location A
: Authority score of location C
: User k’s hub score

38. LOCATION INTEREST INFERENCE
Mining Classical Travel Sequences
: Authority score of location A
: Authority score of location C
: User k’s hub score
𝑆�𝐴�𝐶� = ‫ﻃ‬ (𝑎�𝐴� · 𝑂�𝑢�𝑡�𝐴�𝐶� + 𝑎�𝐶� · 𝐼�𝑛�𝐴�𝐶� + ℎ𝑘�
𝑢�𝑘�∈𝑈�𝐴�𝐶�
)
A
B
C D
E
2 3
4
45
6
3
2 1
The classical score of sequence A!C:

39. Agenda
! Introduction
! Overview of the System
! Modeling Location History
! Location Interest Inference
! Experiments
! Related Work
! Future Work

40. EXPERIMENTS
! Settings
! Evaluation Approaches
! Result
! Discussions

41. EXPERIMENTS
Setting
! GPS Devices
! Coordinates every two seconds.
! 107 users (M:F = 58:49) from May 2007 to Oct 2008.

42. EXPERIMENTS
Setting
! GPS Data – most parts were created in Beijing
! 166,372 km
! 5,081,369 GPS points

43. EXPERIMENTS
Setting
! Parameter Selection
! Stay point detection :
! Tthreh = 20 mins
! Dthreh = 200m
! Extract 10,354 stay points
! Clustering :
! Use OPTICS (Ordering Points To Identify the Clustering
Structure)
Capable of detecting irregular structures

44. EXPERIMENTS
Evaluation Approaches
! User study : 29 subjects (M:F = 15:14) , who have
been in Beijing for more than 6 years
! Location : the fourth ring road of Beijing ( )

45. EXPERIMENTS
Evaluation Approaches
! 2 aspects of evaluations
! Presentation (ability of the retrieved interesting locations)
! Representative : How many locations in this retrieved set are
representative of the given region (0-10) ?
! Comprehensive : Do these locations offer a comprehensive view
of the given region (1-5) ?
! Novelty : How many locations in this retrieved set have
interested you even though they only appeared recently(0-10) ?
! Rank (ranking performance)

46. EXPERIMENTS
Evaluation Approaches
! Select the mode of ratings

47. EXPERIMENTS
Evaluation Approaches
! Baselines :
! Mining interesting locations :
! Rank-by-count
! Rank-by-frequency
! Mining classical travel sequences :
! Rank-by-count
! Rank-by-interests
Consider interests of the locations in a sequence
! Rank-by-experience
Consider experiences of the users who have taken this sequence

48. EXPERIMENTS
Evaluation Approaches

49. EXPERIMENTS
Result
! Results Related to Interesting Locations
! Presentation ability
only 2.4>2.2 doesn’t pass T-test (p>0.2).
! Ranking ability
! There are 60% overlaping (ours vs rank-by-count) , but show
effectively ranking.

50. EXPERIMENTS
Result
! Results Related to Classical Sequences
! Classical rate : the ratio of sequences with a score of
2 in the set.
! Combine …
! user’s travel experiences + rank-by-counts : improved
! locations interests + rank-by-counts : improved

51. EXPERIMENTS
Discussions
! About Interesting Locations
! Why Rank-by-count is bad ?
! Why Rank-by-frequency is bad ?
! About Classical Sequences
! Only Rank-by-counts ?
! Only individuals’ travel experiences ?
! Only location interest ?

52. Agenda
! Introduction
! Overview of the System
! Modeling Location History
! Location Interest Inference
! Experiments
! Related Work
! Future Work

53. RELATED WORK
! Mining Location History
! Mining individual location history
! Mining multiple users’ location histories
1. Detecting significant locations of a user. [2004]
2. Predicting the user’s movement among these
locations. [2005]
3. Recognizing user-specific activities at each
location. [2003]
1. Mining similar sequences from users’ moving
trajectories. [2007]
2. Propose a framework for retrieving maximum
periodic patterns. [2004]
3. Predict where a driver may be going as a trip
progresses. [2003]
4. Recognizing the social pattern in daily user activity.
[2005]

54. RELATED WORK
! Location Recommenders
! Recommenders based on real-time location
! Recommender based on location history
1. Problem: Without understanding the individual and
the nearby locations.
2. Filter away from the returned results the invisible
entities occluded by building. [2007]
1. Recommend geographic locations like shops to
users. [2006]
2. Proposed an enhanced collaborative filtering
solution. [2006]

55. Agenda
! Introduction
! Overview of the System
! Modeling Location History
! Location Interest Inference
! Experiments
! Related Work
! Future Work

56. FUTURE WORK
! Grouping users based on their histories.
! Clustering locations in terms of people’s visits.