Behavioral Knowledge from GPS Traces for Categorizing Mobile Users - PowerPoint PPT Presentation

Modeling of Human Movement Behavioral Knowledge from GPS Traces for Categorizing Mobile Users S O U R S E : W W W 2 0 1 7 A D V I S O R : J I A - L I N G KO H S P E A K E R : H S I U - Y I , C H U D AT E : 2 0 1 7 / 1 1 / 7

Outline  Introduction  Method  Experiment  Conclusion

Introduction  Question Can we map the knowledge of one known region to another unknown(target) region and use this knowledge to categorize the users in the target region?

Introduction  Goal Labeled Unlabeled Knowledge GPS log GPS log User category label

Outline  Introduction  Method  Experiment  Conclusion

Method

Method  User Trajectory Segment  <S[], W[], Traj_Win[]>  S[]: list of stay points, s = <lat, lon, Geo tagg >  W[]: list of waiting points, w = <lat,lon>  Traj_Win[]: {S 1 , (x 1 ,x 2 ), (x 2 ,y 2 ), … ,S 2 }

Method  User Trajectory Segment

Method  Semantic Stay Point Taxonomy(SSP Taxonomy )  SSP Taxonomy :<N, N c , W>  N: place type of the Taxonomy  N c : associated code of the node place  W: aggregated footprints of user

Method  Semantic Stay Point Taxonomy(SSP Taxonomy )

Method  User-Trace Summary(UTS)  N B = <G,Θ>,G=<V,E>  v 1i : (N c ,t i )  N c : associated code of the node place  t i : temporal value of the node  e i : dependences between the vertices

Method  User-Trace Summary(UTS)  Bayesian Network

Method  Θ x5|Pax5 = 0.46

Method  Θ x4|Pax4 * Θ x5|Pax5 * Θ x2|Pax2 = 0.6*0.46*0.98=0.27048

Method  Temporal Common Sub-sequence, (TempCS)clustering algorithm  Similarity measure(Bhattacharyya distance): D B (X 4 , X 5 )=-ln{[X 4 (0)X 5 (0)] 1/2 +[X 4 (1)X 5 (1)] 1/2 }= -ln{[0.4*(0.4*0.32+0.6*0.54)] 1/2 +[0.6*(0.4*0.68+0.6*0.46)] 1/2 }

Method  Temporal Common Sub-sequence, (TempCS)clustering algorithm  N B1 :X 4 X 3 X 5 X 1 X 2  N B2 :X 4 X 1 X 6 X 5 X 2  Common stay points(L c ):X 4 X 5 X 2 X 1  Common Sub-sequence(L s ):X 4 X 5 X 2

Method  Similarity between N B1 and N B2 :  Sim Sequence (N B1 ,N B2 )= 3/4[D B (X 4 , X 5 )+ D B (X 5 , X 2 )]

Method  User Categorization  Classification task:  PV u = {p 1 ,p 2 ,…,p i }  i: user-category  p i : probability of the user u in category i

Method  User Categorization  Feature  f1: visit in types of places  f2: Speed of movement or transportation mode  f3: User Movement

Method  User Categorization  Bayesian network  When independent  Weighting each of feature

Method  Transfer Learning

Method  Transfer Learning  Extract the parent’s code c p of a node c.  Node c has n sibling, append n+1 along with the parent’s code c p n+1.  Check whether the same place-type in and assign the same code if present.  Generate  Get the common taxonomy

Outline  Introduction  Method  Experiment  Conclusion

Experiment  Dataset

Experiment  Accuracy of User-Classification

Experiment

Outline  Introduction  Method  Experiment  Conclusion

Conclusion  Address the user categorization problem from the GPS traces of the users.  Propose a framework to model individual’s movement patterns.  Transfer knowledge base from one city domain to another unknown city.