Josselin Muller completed a 6-month internship at a digital marketing company in New York where he conducted user data analysis, log analysis, and developed a 3D graph visualization tool. He gained valuable international work experience, learned from his supervisor Jad Nehme, and is grateful to his colleagues for their support during the internship.
3. Internship Offer
• Selection: • Benefits:
– Test – $2 500 / month
– Interview – Apartment in Upper West
Side
• Location: New York – Airline tickets & Visa
– Transport from JFK
– Sports tickets
• 6 months – 6 days off
“ Welcome in Burgerland ! ”
10/07/2012 Josselin Muller - MTI 2012 3 | 28
4. Company
• 24/7 Real Media & The Media Innovation Group
– Digital Marketing
• Worldwide company based in New York City
– 12 countries
– 600+ employees
• WPP Group
– World’s 2nd largest advertising group
10/07/2012 Josselin Muller - MTI 2012 4 | 28
13. User Data Analysis
• Data warehouse:
• Questions
– “What is the probability for a visitor to come
back? When ?”
– “How long the User Data should be kept?”
10/07/2012 Josselin Muller - MTI 2012 13 | 28
14. User Data Analysis
• Subset of data: 3 months
• > 1 To
• User Data:
– Impressions (delivered ads)
– Clicks (clicks on delivered ads)
– Events (events on advertiser websites, like product
purchase)
10/07/2012 Josselin Muller - MTI 2012 14 | 28
15. User Data Analysis
• Global statistics
• Case study
• Theories
– Number of previous visits
– Last time seen
– Age of cookie
10/07/2012 Josselin Muller - MTI 2012 15 | 28
16. User Data Analysis
• Meeting to present the results
Number of previous visits
(absolute) Last time seen (absolute)
40000000 40
Millions
30000000 30
20000000 20
10000000 10
0 0
[1000,10…
0
1
[900,1000[
[100,200[
[200,300[
[300,400[
[400,500[
[500,600[
[600,700[
[700,800[
[800,900[
>= 10 000
[80,100[
[14,20[
[20,40[
[40,60[
[60,80[
[8,14[
[2,4[
[4,8[
Come back Don't come back
Come back Don't come back Summary
100.00%
50.00%
0.00%
[1h, [2h, [4h, [8h, [14h, [1d, [2d, [3d, [4d, [5d, [6d, [7d, [10d, [14d,
0 < 1h
2h[ 4h[ 8h[ 14h[ 24h[ 2d[ 3d[ 4d[ 5d[ 6d[ 7d[ 10d[ 14d[ 31d[
Delta 90. 66. 75. 79. 79. 81. 80. 83. 85. 86. 86. 86. 85. 86. 86. 93.
10/07/2012 Josselin Muller - MTI 2012 16 | 28
17. Log Analysis
• Safari Issue
– Problem: some « bad » users still come
– Solution: log analysis
0 1 2603648971646124 - C 0 imp b3
migAgencyId=335&migSource=b3&migTrackDataExt=korea;zapuser/8815;107108;95/450&migRandom
=7154171355&migTrackFmtExt=network;account/campaign;ad;page - 210.104.146.253 1330905600 ""
200 "Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 6.1; Trident/5.0; SLCC2; .NET CLR 2.0.50727;
.NET CLR 3.5.30729; .NET CLR 3.0.30729; Media Center PC 6.0; MASM; InfoPath.2; .NET4.0C)"
"http://adcreative.naver.com/ad3/1035/1035221/120227_HDC_03_475x100_W.swf"
UserID
Cookie state [C = cookied]
Event type [imp, event, click]
User Agent
10/07/2012 Josselin Muller - MTI 2012 17 | 28
18. Log Analysis
• 1st step:
– 28 « bad » users
– User Agent analysis
• Results: 3 categories
Category Description
Changed The UA contains a browser name, like Firefox ou Chrome, but not Safari
Truncated There is no browser name in the UA
Safari The UA contains Safari
10/07/2012 Josselin Muller - MTI 2012 18 | 28
19. Log Analysis
• 2nd step: Project Zebra
– Objective: count the number of bad users
– Solution: development of an optimized parser
• Bad users list (137 133 112 ids, > 2.3 Go)
• Log files (155 953 837 rows, > 81 Go)
• First version: C++
• Second version: full C (10x faster)
• Third version: flexible (options)
10/07/2012 Josselin Muller - MTI 2012 19 | 28
20. 3D Graph Visualization
• Existing project made by Jerome Champigny
• Framework: Three.js
10/07/2012 Josselin Muller - MTI 2012 20 | 28
21. 3D Graph Visualization
• 2D Text in 3D World
– Canvas for texture
– 3D Plan
– Material
10/07/2012 Josselin Muller - MTI 2012 21 | 28
22. 3D Graph Visualization
• Camera 2.0
– Smart Zoom (Google-map like)
– Translation & rotation with mouse
– Bounded to the grid
– JS Events:
• MouseUp
• MouseDown
• MouseMove,
• MouseWheel
10/07/2012 Josselin Muller - MTI 2012 22 | 28
23. 3D Graph Visualization
• “Terrain view”
– One point = one data
– All points linked with
the indexbuffer
– Special color for each “dataset”
10/07/2012 Josselin Muller - MTI 2012 23 | 28
25. 3D Graph Visualization
• Bar chart
– One data = One bar = 8 vertices
– Link vertices to make 5 faces (not the bottom one)
10/07/2012 Josselin Muller - MTI 2012 25 | 28