Quantifying the potential of ride-sharing using Call

Description Records (CDRs)

Blerim Cici*, Athina Markopoulou*, Enrique Frías-Martínez**, Nikolaos Laoutaris**

*University of California, Irvine

**Telefonica Research

Outline

• Introduction • Mobility Data • Algorithms and Results

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 2

Outline

• Introduction • Mobility Data • Algorithms and Results

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 3

What is Ride-Sharing ?  

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 4

Benefits

5  B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 5

Ride-Sharing: An old idea

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 6

Ride-Sharing in the past 1.  Difficult to set up

2.  Few opportunities

3.  Inflexible

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 7

Ride-Sharing now

1.  Difficult to set up

2.  Few opportunities 3.  Inflexible

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 8

Ride-Sharing now

1.  Difficult to set up

2.  Few opportunities 3.  Inflexible

But, why it’s not mainstream yet ?

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 9

Introduction •  We want to investigate if ride-

sharing possible.

•  Considered quantifiable parameters: 1.  Distance tolerance 2.  Distribution of departure times 3.  Time tolerance

•  An upper bound due to simplifications

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 10

Parameters - Space

d : distance tolerance

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 11

Parameters - Time

Time     8  am   5  pm  

•  σ : standard deviation of Home/Work departure times

•  τ : time tolerance

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 12

Contributions

•  We used real location information to validate its potential

•  We formulated ride-sharing as a facility location problem.

•  We developed scalable and efficient algorithms to match the users.

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 13

Outline

• Introduction • Mobility Data • Algorithms and Results

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 14

Data

•  Call Description Records (CDRs): – every phone call a new entry – the location of closest tower is recorded

•  Our CDR dataset: – September – December 2009 – 5M users in Madrid

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 15

Determining Home/Work

•  Use existing methodology: – S. Isaacman, R. Becker, R. Caceres, S.

Kobourov, M. Martonosi, J. Rowland, and A. Varshavsky, “Identifying Important Places in People’s Lives from Cellular Network Data”, Pervasive 2011

•  Home/Work locations of 272K users

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 16

Outline

• Introduction • Mobility Data • Algorithms and Results

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 17

Formulation

•  Create Ride-Sharing groups: – All cars have capacity of 4 – Matching users who live and work close

by – Goal: is to minimize the number of cars

used

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 18

Formulation  •  Capacitated Facility location with

Unsplittable Demands: – Facilities : Drivers – Clients : Passengers

•  We choose as driver the user, who will minimize the distance traveled by his passengers.

•  Large cost for every new driver

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 19

EndPoints RS •  NP  hard  !  

•  Inspired  by:    – M.  Korupolu,  C.  Plaxton,  and  R.  Rajaraman.  “Analysis  of  a  local  search  heurisFc  for  facility”,  ACM-­‐SIAM  1998  

•  EndPoints RS: – Efficient heuristic – Scalable

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 20

EndPoints RS

•  EndPoints RS: – Start with an initial “smart” solution – Iterative improvements by local search

in solution space

•  Scalability – Fixed local search steps – Fix numbers of iterations

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 21

Results for EndPoint RS

0.2 0.4 0.6 0.8 10

20

40

60

80

d (km)

% o

f car

s re

mov

ed

Success of end−point ride−sharing

Absolute upper boundTighter upper boundTime indifferento = 10, m = 10o = 10, m = 20

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 22

EnRoute RS •  Find Home/Work path

through Google Maps

•  EnRoute RS: –  Iterative algorithm –  Fill empty seats by

pick-ups

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 23

Results for EnRoute RS

0.2 0.4 0.6 0.8 10

20

40

60

80

km

% o

f car

s re

mov

ed

Success of en−route ride−sharing

Absolute upper boundTime indifferento = 10, m = 10o = 10 , m = 20

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 24

Summary

•  We evaluated the potential or ride-sharing in the city of Madrid.

•  We used mobility data from CDRs of a major European Telco.

•  There seems to be great potential for ride-sharing

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 25

Limitations

•  We assumed that people are willing to share a ride with strangers

•  This is a strong assumptions. Our work shows only an upper bound

•  In future work we plan to use social filtering

B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS 26

Thank You

27  B. Cici et al. Quantifying the potential of Ride-Sharing using CDRS