Modeling of Human Movement Behavioral Knowledge from GPS

Modeling of Human Movement Behavioral Knowledge from GPS Traces for Categorizing Mobile Users SOURSE: WWW 2017 ADVISOR: JIA-LING KOH SPEAKER: HSIU-YI, CHU DATE: 2017/11/7

Outline l. Introduction l. Method l. Experiment l. Conclusion

Introduction l. 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 l. Goal Labeled GPS log Knowledge Unlabeled GPS log User category label

Outline l. Introduction l. Method l. Experiment l. Conclusion

Method

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

Method l. User Trajectory Segment

Method l. Semantic Stay Point Taxonomy(SSPTaxonomy) l. SSPTaxonomy: <N, Nc, W> l. N: place type of the Taxonomy l. Nc: associated code of the node place l. W: aggregated footprints of user

Method l. Semantic Stay Point Taxonomy(SSPTaxonomy)

Method l. User-Trace Summary(UTS) l. NB = <G, Θ>, G=<V, E> lv 1 i: (Nc, ti) l. Nc: associated code of the node place lti: temporal value of the node lei: dependences between the vertices

Method l. User-Trace Summary(UTS) l. Bayesian Network

Method lΘx 5|Pax 5 = 0. 46

Method lΘx 4|Pax 4*Θx 5|Pax 5*Θx 2|Pax 2 = 0. 6*0. 46*0. 98=0. 27048

Method l. Temporal Common Sub-sequence, (Temp. CS)clustering algorithm l. Similarity measure(Bhattacharyya distance): DB(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 l. Temporal Common Sub-sequence, (Temp. CS)clustering algorithm l. NB 1: X 4 X 3 X 5 X 1 X 2 l. NB 2: X 4 X 1 X 6 X 5 X 2 l. Common stay points(Lc): X 4 X 5 X 2 X 1 l. Common Sub-sequence(Ls): X 4 X 5 X 2

Method l. Similarity between NB 1 and NB 2: l. Sim. Sequence(NB 1, NB 2)= 3/4[DB(X 4, X 5)+ DB(X 5, X 2)]

Method l. User Categorization l. Classification task: l. PVu = {p 1, p 2, …, pi} li: user-category lpi: probability of the user u in category i

Method l. User Categorization l. Feature lf 1: visit in types of places lf 2: Speed of movement or transportation mode lf 3: User Movement

Method l. User Categorization l. Bayesian network l. When independent l. Weighting each of feature

Method l. Transfer Learning

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

Outline l. Introduction l. Method l. Experiment l. Conclusion

Experiment l. Dataset

Experiment l. Accuracy of User-Classification

Experiment

Outline l. Introduction l. Method l. Experiment l. Conclusion

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