The Flexible Socio Spatial Group Queries Bishwamittra Ghosh 1 , - - PowerPoint PPT Presentation

The Flexible Socio Spatial Group Queries

Bishwamittra Ghosh1, Mohammed Eunus Ali2, Farhana M. Choudhury3, Sajid Hasan Apon2, Timos Sellis4, Jianxin Li5 VLDB 2019

1National University of Singapore 2Bangladesh University of Engineering and Technology 3RMIT University and University of Melbourne, Australia 4Swinburne University of Technology, Australia 5The University of Western Australia Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 1

Socio-spatial Graph

spatial layer a b c d e social layer

• 2
• 1

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 2

Problem Formulation

Given

◮ Set of meeting points Q ◮ Socio-spatial graph G = (V , E)

Find top k groups such that score(Gi, qi) ≥ score(Gi+1, qi+1) where Gi is a subgraph of G, qi ∈ Q and 1 ≤ i ≤ k − 1

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 3

Constraints for a feasible group Gi = (V , E)

◮ minimum social connectivity constraint c

◮ degree(v) ≥ c, ∀v ∈ V

◮ maximum distance dmax

◮ dist(v, q) ≤ dmax, ∀v ∈ V

◮ minimum group size nmin, maximum group size nmax

◮ nmin ≤ |V | ≤ nmax Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 4

Score of group Gi = (V , E) w.r.t. meeting point q

scoresocial = 2|E| |V |(|V | − 1) scorespatial = 1 −

• v∈V dist(v, q)

dmax|V | scoresize = |V | nmax score = α · scoresocial + β · scorespatial + γ · scoresize

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 5

Literature review

There are existing works that address socio spatial group queries. The major gaps are

◮ specific group size6 vs variable group size ◮ finding only the best group6 vs top k groups ◮ fixed meeting point vs multiple meeting points7 ◮ average social connectivity constraint8 vs minimum social connectivity

constraint9

◮ ranking function combining social and spatial factors10 vs ranking

function combining social, spatial and group size factors

6[Fang17], [Shen16], [Zhu14],[Yang12] 7[Shen16] 8[Shen16], [Yang12] 9[Fang17],[Zhu14] 10[Armenatzoglou15] Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 6

Contribution

◮ Exact algorithm

◮ member ordering based on spatial distance ◮ optimistic assumption (maximum) on social connectivity of including

members

◮ early termination based on upper bound on spatial distance

Intermediate group nmax nmin

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 7

• Continued. . .

◮ Heuristic approximate approach

◮ member ordering based on spatial distance ◮ lower bound on social connectivity while including a member in the

intermediate group

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 8

• Continued. . .

◮ A fast approximate approach

◮ a tighter lower bound on social connectivity while including a member

in the intermediate group

◮ upper bound on spatial distance and lower bound on social connectivity

that improves the rank of current exploring group

◮ prune when including a member can not increase the score of

intermediate group

◮ Greedy approach

◮ avoid backtracking Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 9

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X select meeting point that has minimum spatial distance to first unexplored member

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X {a, b, c} is a result group

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X Advance termination based on upper bound on spatial distance

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X degree(c, {a}) < lower bound on social connectivity

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X degree(c, {b}) ≥ lower bound on social connectivity

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Simulation

◮ meeting point q1 ◮ distance ordered members

{a, b, c, d . . . } ∅ a a, b a, b, c a, d . . . b b, c . . . . . .

◮ meeting point q2 ◮ distance ordered members

{b, a, c, . . . } ∅ b b, a . . . b, c . . . X

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 10

Approximation ratio of fast approximate algorithm

approximation ratio = lowest scoring retrieved group best scoring group that may not be retrived Emphasis Weights Approximation ratio Social score α = 1, β = γ = 0

c nmax−1

Spatial score β = 1, α = γ = 0 1 Size score γ = 1, α = γ = 0

nmin nmax

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 11

Experimental Results

B = Baseline11, E = Exact, A = Approximate, FA = Fast approximate, GA = Greedy approximate

100 150 200 250 300 50 100 200 400 Time (ms) # of meeting points E A FA GA

(a) Brightkite

102 103 104 105 106 50 100 200 400 Time (ms) # of meeting points B E A FA GA

(b) Gowalla Figure: Computation time of different algorithm

11[YANG12] Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 12

Experimental Results

A = Approximate, FA = Fast approximate, GA = Greedy approximate

20 40 60 80 100 50 100 200 400 Accuracy (%) # of meeting points

k(A) 1.5k(A) 2k(A) k (FA) 1.5k (FA) 2k (FA) k (GA) 1.5k (GA) 2k (GA)

(a) Brightkite

20 40 60 80 100 50 100 200 400 Accuracy (%) # of meeting points

k(A) 1.5k(A) 2k(A) k (FA) 1.5k (FA) 2k (FA) k (GA) 1.5k (GA) 2k (GA)

(b) Gowalla Figure: Percentage of groups in top k of approximate algorithm that also appear in top k, top 1.5k, and top 2k of the exact algorithm

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 13

Conclusion

◮ we propose novel top k flexible social spatial group queries ◮ we devise a ranking function combining social closeness, spatial

distance, and group size

◮ we propose exact algorithm and efficient approximate algorithms ◮ Exact algorithm runs up to 10× faster than the baseline ◮ Fast approximate algorithm runs up to 100× faster than exact

algorithm and returns the same set of results in most cases

Thank You

Bishwamittra Ghosh The Flexible Socio Spatial Group Queries VLDB 2019 14