Trajectory Pattern Mining Fosca Giannotti Mirco Nanni Dino

Trajectory Pattern Mining Fosca Giannotti, Mirco Nanni, Dino Pedreschi, Fabio Pinelli KDD Lab (ISTI-CNR & Univ. Pisa) Presented by: Qiming Zou

Overview o o o Motivations Trajectory T-Pattern Regions of Interest Future Work Q&A

Motivations o Large number of mobile devices, mobile services available

Motivations o o o It is possible to collect position traces from such devices We can extract information and patterns from these data to describe mobility behaviors Use this information for fields such as urban planning

Trajectory o Trajectories are sequences that contain the spatial and temporal information about movements

Trajectory o o o Trajectories are usually given as spatiotemporal (ST) sequences: <(x 0, y 0, t 0), . . . , (xn, yn, tn)> xi, yi is the position coordinate relative to the origin ti is the time stamp for the position information

Trajectory o 2 D and 3 D representation of a trajectory:

T-Pattern o A Trajectory Pattern (T-Pattern) is a couple (s, α), where: n n s = <(x 0, y 0), . . . , (xn, yn)> is a sequence of n+1 locations α= <α 1, . . . , αn> are the transition times such that αi = Δti = ti – ti-1

T-Pattern o A T-Pattern Tp occurs in a trajectory if it contains a subsequence S such that: n n each (xi, yi) in Tp matches a point (xi’, yi’) in S the transition times in Tp are similar to those in S

T-Pattern o The same exact spatial location (x, y) usually never occurs n o Yet, close locations often represent the same place The same exact transition times usually do not occur often n However, close times often indicate similar behavior

T-Pattern o To solve the problem, we introduce the notions of: n n Spatial neighborhood: Two points match if one falls within a spatial neighborhood N() of the other Temporal tolerance: Two transition times match if their temporal difference is ≤ τ

T-Pattern o Example:

Regions of Interest o o It is too computational intensive and yield little practical use to generate all T-Patterns Solution: Use a Regions of Interest approach, only use these regions as nodes of the T-Patterns

Regions of Interest o Given a set of Regions of Interest R, define the neighborhood of (x, y) as: n n Neighbors = belong to the same region Points in no region have no neighbors

Regions of Interest S=<(x 0, y 1, t 1), . . . , (x 4, y 4, t 4)> => <(R 4, t 0), (R 3, t 2), (R 3, t 3), (R 1, t 4)>

Regions of Interest o o What if the Regions of Interests are not known before hand? Define heuristics for automatic Regions of Interest extraction from data: n n n Geography-based (crossroads) Usage-based (popular places) Mixed (popular squares)

Future Work o o Application-oriented tests on large, real datasets Study relations with n n n Geographic background knowledge Privacy issues Reasoning on trajectories and patterns

Trajectory Pattern Mining Questions?