1. TOPTRAC: Topical Trajectory Pattern Mining
Source: KDD 2015
Advisor: Jia-Ling Koh
Speaker: Hsiu-Yi,Chu
Date: 2018/1/22

2. Outline
Introduction
Method
Experient
conclusion

3. Introduction

4. Introduction Goal Topical trajectory mining problem:
Given a collection of geo-tagged message trajectories, it’s to find topical transition pattern and the top-k transition snippets which best represent each transition pattern

5. Introduction
Transition pattern Transition snippet

6. Introduction Definition Trajectory(st) geo-tagged message (mt,i)
Geo-tag Gt,i : 2-dim vector(Gt,i,x,Gt,i,y)
Bag-of-word wt,i : N words{wt,i,1,…, wt,i,n}

7. Introduction Definition Latent semantic region:
a geographical location where messages are posted with the same topic preference
Topical transition pattern:
a movement from one semantic region to another frequently

8. Outline
Introduction
Method
Experience
conclusion

9. Method Generative Model Assume there are M latent semantic regions
K hidden topics
in the collection of geo-tagged messages
How to generate each sequence
st = (mt,1, mt,2 , … , mt,n )

10. Method Generative process Ex:Θ1=(topic1,…topic k )
Ex:Φk=(word1,…,word v)
0.3
0.2
0.6 2 3

11. Method λt : Bernoulli distribution(0~1) St,i = {0,1}:
mt,1
mt,2
λt : Bernoulli distribution(0~1)
St,i = {0,1}:
Whether mt,i is in the local context

12. Method Case1: St,i = 1 Case2: i =1and St,i = 1 or
mt,1
mt,2
Case1: St,i = 1
Select Rt,i = Uniform(1/M)
Generate Gt,i = Uniform(f0)
Case2: i =1and St,i = 1 or
i >= 2 and St,i = 1 and St,i-1 = 0
Select Rt,I = Categorical(δ0)
Generate Gt,I = f(Gt,I)

13. Method Case3: else Select Rt,I = Categorical(δr(t,i-1),z(t,i-1))
mt,1
mt,3
Method
Case3: else
Select Rt,I = Categorical(δr(t,i-1),z(t,i-1))
Generate Gt,I = f(Gt,I)

14. Method Select Topic Select a message Zt,i = Categorical(θRt,i)
wt,I = Multinomial(ΦZt,i)

15. Method
Likelihood

16. Method Variational EM Algorithm Maximum likelihood estimation
θR, Φk, λt
St,i, Rt,i, Zt,I
μr, Σr

17. Method Finding the Most Likely Sequence Notations:
: maximum probability to generate the subsequence when St,i=0
: : maximum probability to generate the subsequence when St,i=1

18. Method
Compute :
Compute :
case1: St,i-1 = 0 ; case2 : St,i-1 = 1

19. Method Finding Frequent Transition Patterns
st’ = {(st,1, rt,1, zt,1),…,(st,n, rt,n, zt,n)}
Transition Patterns = {( r1, z1)(r2, z2)}
Start with (1, r1, z1) and ends with (1, r2, z2)
τ : minimum support

20. Method Example Top-k transition snippets k largest probabilities of
s1’={(0,1,1)(1,1,2)(1,2,1)}, s2’={(1,1,2)(0,2,1)(1,2,1)} with τ = 2 → {(1,2)(2,1)} is a transition pattern
Top-k transition snippets
k largest probabilities of

21. Outline
Introduction
Method
Experience
conclusion

22. Experience Data sets NYC SANF
9070 trajectories, geo-tagged messages
M = 30, K = 30, τ = 100
SANF
809 trajectories,19664 geo-tagged messages
M = 20, K = 20, τ = 10

23. Experience Baseline LGTA NAÏVE
Run the inference algorithm and find frequent trajectory patterns similar in page15,16
NAÏVE
First groups messages using EM clustering
Cluster the messages in each group with LDA

24. Experience

25. Experience

26. Experience

27. Experience

28. Outline
Introduction
Method
Experience
conclusion

29. Conclusion
Propose a trajectory pattern mining algorithm, called TOPTRAC, using probabilistic model to capture the spatial and topical patterns of users.
Developed an efficient inference algorithm for our model and also devised algorithms to find frequent transition patterns as well as the best representative snippets of each pattern.