Presentation is loading. Please wait.

Presentation is loading. Please wait.

Landmark-Based User Location Inference in Social Media YUTO YAMAGUCHI †, TOSHIYUKI AMAGASA † AND HIROYUKI KITAGAWA † †UNIVERSITY OF TSUKUBA 13/10/08 COSN.

Similar presentations


Presentation on theme: "Landmark-Based User Location Inference in Social Media YUTO YAMAGUCHI †, TOSHIYUKI AMAGASA † AND HIROYUKI KITAGAWA † †UNIVERSITY OF TSUKUBA 13/10/08 COSN."— Presentation transcript:

1 Landmark-Based User Location Inference in Social Media YUTO YAMAGUCHI †, TOSHIYUKI AMAGASA † AND HIROYUKI KITAGAWA † †UNIVERSITY OF TSUKUBA 13/10/08 COSN Yuto Yamaguchi 1

2 LOCATION-RELATED INFORMATION 13/10/08 COSN Yuto Yamaguchi 2 Eating seafood !!! I’m at Logan airport Profile Residence: Tokyo, Japan northeastern

3 APPLICATIONS Various Researches using Home Locations Outbreak Modeling [Poul+, ICWSM’12] Real-World Event Detection [Sakaki+, WWW’12] Analyzing Disasters [Mandel+, LSM’12] Other Useful Applications Location-aware Recommender [Levandoski+, ICDE’12] Merketing, Ads Disaster Warning 13/10/08 COSN Yuto Yamaguchi 3

4 OUR PROBLEM Location profiles are not available for … 76% of Twitter users[Cheng et al., CIKM’10] 94% of Facebook users[Backstrom et al., WWW’10] This reduces opportunities of location information  User Home Location Inference 13/10/08 COSN Yuto Yamaguchi 4

5 USER HOME LOCATION INFERENCE Content-Based Approaches [Cheng et al., CIKM’10] [Kinsella et al., SMUC’11] [Chandra et al., SocialCom’11] Graph-Based Approaches [Backstrom et al., WWW’10] [Sadilek et al., WSDM’12] [Jurgens, ICWSM’13] 13/10/08 COSN Yuto Yamaguchi 5 Our focus

6 GRAPH-BASED APPROACH (1/2) Basic Idea 13/10/08 COSN Yuto Yamaguchi 6 Boston Chicago New York Boston? friends

7 GRAPH-BASED APPROACH (2/2) Closeness Assumption 13/10/08 COSN Yuto Yamaguchi 7 Friends Not friends Spatially close Spatially distant Really close? 60% are 100km distant

8 CONCENTRATION ASSUMPTION 13/10/08 COSN Yuto Yamaguchi 8 Boston Boston? LANDMARK Unknown NYChicago

9 LANDMARKS 13/10/08 9 COSN Yuto Yamaguchi

10 REQUIREMENTS Small Dispersion Large Centrality 13/10/08 COSN Yuto Yamaguchi 10

11 EXAMPLES IN TWITTER 13/10/08 COSN Yuto Yamaguchi 11

12 LANDMARKS MAPPING 13/10/08 COSN Yuto Yamaguchi 12 Red: all users Blue: landmarks

13 PROPOSED METHOD 13/10/08 13 COSN Yuto Yamaguchi

14 OVERVIEW Probabilistic Model Modeling 13/10/08 COSN Yuto Yamaguchi 14 Each user has his/her location distribution Location inference = Selecting the location with the largest probability density location set  LANDMARK MIXTURE MODEL

15 DOMINANCE DISTRIBUTION Spatial distribution of followers’ home locations Modeled as Gaussian Landmarks have small covariances  many followers at the center 13/10/08 COSN Yuto Yamaguchi 15 latitude longitude many followers few followers

16 LANDMARK MIXTURE MODEL (LMM) 13/10/08 COSN Yuto Yamaguchi 16 Inference target user follow Landmark Non-landmark Dominance distribution Mixture weight Large weight for landmark

17 MIXTURE WEIGHTS 13/10/08 COSN Yuto Yamaguchi 17 Proportional to centrality LandmarkNon-landmark Large mixture weightSmall mixture weight

18 CONFIDENCE CONSTRAINT If the distribution does not have a clear peak, we should not infer the location of that user 13/10/08 COSN Yuto Yamaguchi 18  High precision but low recall

19 CENTRALITY CONSTRAINT We can reduce the cost by ignoring non-landmarks 13/10/08 COSN Yuto Yamaguchi 19  low cost but low recall Inference target user follow Landmark Non-landmark

20 EXPERIMENTS 13/10/08 20 COSN Yuto Yamaguchi

21 DATASET Twitter dataset provided by [Li et al., KDD’12] 3M users in the U.S. 285M follow edges Geocode their location profiles for ground truth 465K users (15%)  labeled users Test set 46K users (10% of labeled users) 13/10/08 COSN Yuto Yamaguchi 21

22 PERFORMANCE COMPARISON 13/10/08 COSN Yuto Yamaguchi 22 Compared three methods LMM: our method UDI: [Li+, KDD’12] Naïve:Spatial median

23 EFFECT OF CONFIDENCE CONSTRAINT 13/10/08 COSN Yuto Yamaguchi 23 p0 We can adjust the trade-off between precision and recall

24 EFFECT OF CENTRALITY CONSTRAINT 13/10/08 COSN Yuto Yamaguchi 24 c0 We can adjust the trade-off between cost and recall

25 CONCLUSION Introduced the concentration assumption instead of widely-used closeness assumption There exist landmarks Proposed landmark mixture model Outperforms the state-of-the-art method Confidence / Centrality constraint Future work Other application of landmarks Recommending landmarks or their tweets 13/10/08 COSN Yuto Yamaguchi 25


Download ppt "Landmark-Based User Location Inference in Social Media YUTO YAMAGUCHI †, TOSHIYUKI AMAGASA † AND HIROYUKI KITAGAWA † †UNIVERSITY OF TSUKUBA 13/10/08 COSN."

Similar presentations


Ads by Google