Respecting Privacy with Look alike data sets Tim Garnsey - Director - - PowerPoint PPT Presentation

respecting privacy with look alike data sets
SMART_READER_LITE
LIVE PREVIEW

Respecting Privacy with Look alike data sets Tim Garnsey - Director - - PowerPoint PPT Presentation

YOW! Data Respecting Privacy with Look alike data sets Tim Garnsey - Director - Verge Labs Data is the new oil Clive Humby What makes refining hard? 1. Corporate competitive advantage 2. Peoples privacy Competitive


slide-1
SLIDE 1

YOW! Data

Respecting Privacy with ‘Look alike’ data sets

Tim Garnsey - Director - Verge Labs

slide-2
SLIDE 2

“Data is the new oil”

Clive Humby

slide-3
SLIDE 3

What makes refining hard?

  • 1. Corporate competitive advantage
  • 2. People’s privacy
slide-4
SLIDE 4

Competitive advantage

User looks for content Generating content in the process Making content attractive

slide-5
SLIDE 5

People’s privacy

slide-6
SLIDE 6

Analysing data

  • 1. Care about relationships within subjects (in aggregate)
  • 2. Care about relationships across subjects (in aggregate)
  • 3. Don’t care about subjects
slide-7
SLIDE 7

Analysing data

Date Name Business * Lat Long City 2017-06-15 Nicole Thelma's Filipino Restaurant 5 36.0631

  • 115.038

Henderson NV 2015-09-23 Matt Cabo's Mexican Cuisine and Cantina 1 35.0854

  • 80.8476

Charlotte NC 2016-05-03 Maddie The Parlor 3 33.5094

  • 112.04

Phoenix AZ 2017-08-08 Nick DC Steak House 1 33.3024

  • 111.842

Chandler AZ 2017-07-16 Keith Kneaders Bakery & Cafe 1 33.6266

  • 111.896

Scottsdale AZ 2013-02-10 Mike Hickory Tavern 4 35.1019

  • 80.9912

Charlotte NC 2016-08-23 Michael Wendy's Noodle Cafe 3 36.1275

  • 115.225

Las Vegas NV 2017-02-10 Petrina Clones 4 Patients ONLY 1 36.2858

  • 115.285

Las Vegas NV 2016-01-03 Maddie Chicks With Spiritual Gifts 5 33.4689

  • 112.07

Phoenix AZ 2016-10-19 Sunggin Cholla Prime Steakhouse & Lounge 4 33.454

  • 111.886

Scottsdale AZ 2012-01-30 Maribeth Urban Cookies Bakeshop 5 33.4742

  • 112.065

Glendale AZ 2016-11-04 Jaime In-N-Out Burger 5 33.508

  • 112.266

Phoenix AZ

slide-8
SLIDE 8

Analysing data

Date Name Business * Lat Long City 2017-11-22 Angela Fat Tuesday 5 36.1095

  • 115.171

Las Vegas NV 2013-09-18 Kenny Yanni's Gyros 5 36.0119

  • 115.136

Las Vegas NV 2012-12-07 Joyce Panera Bread 4 33.5805

  • 112.122

Glendale AZ 2017-12-04 Kenny Scallywags 5 43.6878

  • 79.3945

Toronto ON 2012-11-15 Abby Tip of the Tail Grooming 1 41.4317

  • 81.8867

North Olmsted OH 2011-02-26 Laura Thai Express 1 43.6132

  • 79.5558

Etobicoke ON 2017-10-11 Liane Armando's Mexican Food 4 33.6842

  • 112.107

Phoenix AZ 2017-05-17 Laura Momo Hair Salon 5 43.7039

  • 79.3979

Toronto ON 2014-01-02 Caitlin Chopstix Express 3 36.1425

  • 115.209

Las Vegas NV 2017-03-04 Han Take Over Lease 5 40.4403

  • 79.9863

Pittsburgh PA 2016-10-21 Will Betty's Flower Shop 1 36.2384

  • 115.155

North Las Vegas NV 2013-08-19 Steve Rise Biscuits & Donuts 3 33.4923

  • 111.924

Scottsdale AZ

slide-9
SLIDE 9

Analysing data

  • 1. Care about relationships within subjects (in aggregate)
  • 2. Care about relationships across subjects (in aggregate)
  • 3. Don’t care about subjects

… really need a “very statistically similar” data set

slide-10
SLIDE 10

Copying machines

slide-11
SLIDE 11

Use GANs …

0.3 0.6 0.1

slide-12
SLIDE 12

Use differential privacy …

slide-13
SLIDE 13

… and join them together

0.3 0.6 0.1

slide-14
SLIDE 14

Check-in data

slide-15
SLIDE 15

Check-in data

slide-16
SLIDE 16

Zoom

slide-17
SLIDE 17

Tips and tricks

There is a universe here, but getting started is easy:

  • 1. Blogs, SO and wikipedia
  • 2. Convert to floating point (cities, text)
  • 3. Start small (there is always something that is valuable

enough)

slide-18
SLIDE 18

Questions?

tim@vergelabs.ai @TimGarnsey