mining social contact with active rfid - ccc event blog · mining social contact with active rfid...
TRANSCRIPT
25C3, Berlin, December 30th, 2008
Ciro CattutoInstitute for Scientific Interchange Foundation, Torino, Italy
http://isiosf.isi.it/~cattuto
mining social contact with active RFID
‣ detecting face-to-face social contacts‣ distributed contact detection‣ a case study at a conference‣ contact networks, contact duration‣ outlook
in collaboration withWouter Van den Broeck, Alain Barrat, Jean-François Pinton,
Vittoria Colizza, Alessandro Vespignani,OpenBeacon/OpenAMD team @ 25C3
contact detection: motivations
★ fundamental knowledge on human contact★ epidemiology★ design of p2p mobile sharing schemes★ integration with on-line information★ enhanced user experience
R. Munroe, http://xkcd.com/403/
contact detection: requirements
‣ spatial resolution ~ 1 meter‣ anisotropy‣ temporal resolution ~ 5-20 seconds‣ unobtrusive‣ scalable‣ very low cost‣ open (hardware and software)
RX
channel B
TX
power 0
channel A
TX
power 1
channel A
TX
power 3
channel A
sleep TX
power 2
channel A
sleep sleep
TX
power X
channel Y
TX
power X
channel Y
sleep sleepRX
channel B
TX
power 0
channel A
TX
power 3
channel A
TX
power 1
channel B
2 out of 8 1 out of 8
TX
power 0
channel B
5 out of 8
TX
power 1
channel A
TX
power 2
channel A
, , , }{
cyclic
CONTACT REPORTING
active RFID tags & OpenBeacon‣ beacon RFID tags
• active RFID• RF broadcast of tag ID• very low power consumption• long range• possibility of triangulation
‣ OpenBeacon• open hardware and software design• 2.4 GHz microwave ISM band• PIC microcontroller + RF chip• easily reprogrammable• 128 bytes RAM, 2 kB flash program memory• low power-consumption• inexpensive
courtesy M. Meriac
LAN
active RFID device, 2.4 Ghz
data collection pipeline
42
“tag 42 at power 1”
= detector
location
101102
103104
202
201
203
204
205206
301 302
303304
experiment layout (may ‘08)
5m
co-presence at stations
see also: V. Kostakos et al., “Brief encounter networks”, http://arxiv.org/abs/0709.0223
RFID-based contact detection: strategy
✓ unobtrusive✓ scalable✓ very low cost✓ open (hw and sw)❖ spatial resolution ~ 1 meter❖ anisotropy❖ temporal resolution ~ 5-20 seconds
RX
channel B
TX
power 0
channel A
TX
power 1
channel A
TX
power 3
channel A
sleep TX
power 2
channel A
sleep sleep
TX
power X
channel Y
TX
power X
channel Y
sleep sleepRX
channel B
TX
power 0
channel A
TX
power 3
channel A
TX
power 1
channel B
2 out of 8 1 out of 8
TX
power 0
channel B
5 out of 8
TX
power 1
channel A
TX
power 2
channel A
, , , }{
cyclic
CONTACT REPORTING
direct contact detection
data collection pipeline
LAN
42
13“42 saw 13 at power 0”
2 tags on desk 1, 3 tags on desk 2 2 tags on desk 1, 2 tags on desk 2,
1 tag (4200) on desk 3
4 tags on desk 1,
1 tag (4200) on desk 2
5 tags on a single desk 5 tags arranged in a row,
spaced by about 60-70 cm from one another. tag 4200 is at one side.
At increasing distances there are 4150, 4212, 4221, 4170.
- a thick link (weight = 1) indicates that at least 1 low-power packet was exchanged during the last 20 seconds- weaker weights measure the rate of exchange of next-to-lowest power packets
direct contact detection
fine-grained spatial (~ 1m) and temporal (~ 10s) resolution
on-line post-processing @ 25C3readers
Python/Twistedaggregator
Python/Twistedbackend engine:real-time views
UDP
TCP
AMF socket
web-based Flash viz
lobby
cafeteria
bar
conf.
room
experiment layout (oct ‘08)
7:30am 9am 11am 1pm 3pm 5pm 7pm 8:50pmtime of the day
10
20
30
40#
pers
ons
conference roombarcafeteria
presence of attendees
instantaneous contact network
9am 11am 1pm 3pm 5pm 6pm4pm2pm10am 12pmtime of the day
10
20
30
40#
pers
ons
0
1
2
3
4
〈 k 〉
number of active contacts
9am 11am 1pm 3pm 5pm 6pm4pm2pm10am 12pmtime of the day
0
10
20
30
40#
pers
ons
0
5
10
15
20
25
# cl
ique
s
2-cliques3-cliques4-cliques
number of cliques
contact duration
101 102 103
Δt (s)10-4
10-3
10-2
10-1
100
P(Δ
t)
P (!t) ! (!t)−2
x y
epidemics on real contact patterns
cumulative contact networks
PajekPajek
ItalyUSThe NetherlandsFrance
SwedenUK
Israel
AustriaPortugalSwitzerland
PolandAlgeria
Germany
Senior researcher
Junior researcher
Student
outlook
‣ large-scale experiments (> 1,000 persons) in early 2009‣ several medium scale experiments planned
‣ variety of contexts• conference• office• hospital• museum
‣ models of human dynamics‣ longitudinal contact network analysis‣ epidemiology‣ organizational studies‣ integration with background knowledge‣ social networking services
“social serendipity”
off-line
on-line
finding surprising connections
www.sociopatterns.org
acknowledgements:• aestetix• Janette Lehmann• Brita Meriac• Milosch Meriac• Aurora Mazzone• Eric• Marco Perosa• Tomasz Rybak
• YOU