Annoyed Users: Ads and Ad-Block Usage in the Wild Enric Pujol - PowerPoint PPT Presentation

Annoyed Users: Ads and Ad-Block Usage in the Wild Enric Pujol Oliver Hohlfeld Anja Feldmann TU Berlin RWTH Aachen TU Berlin IMC’15 Tokyo, Japan

2

http://www.journalism.org/2015/04/29/digital-news-revenue-fact-sheet Page Fair and Adobe report on Ad Blocking 3

Mozilla Add-on statistics 4

Publishers reaction The Guardian 5

(Why can I not see BILD.de?) 6

Acceptable ads program (non-intrusive ads) White-listing within AdBlockPlus (ABP) 7

Our work Residential Broadband Network Online advertising Methodology based on AdBlockPlus Ad-blocker usage Characterize ad traffic 8

9

10

URL Content-Type Base URL ||ads.example.com^ $ script, image, domain=bar.com DOM Tree www.bar.com www.foo.com #document http://ads.example. HTML com/ad.gif HEAD BODY … IMG Content Policy Is this information available in HTTP header traces? 11

||ads.example.com ^ $ script, image, domain=foo.com 12

||ads.example.com ^ $ script, image, domain=foo.com HTTP logs Bro IDS 13

||ads.example.com ^ $ script, image, domain=foo.com HTTP logs Map Web Pages Bro IDS ReferrerMap 14

||ads.example.com ^ $ script, image, domain=foo.com HTTP logs Map Web Pages Bro IDS ReferrerMap Classify requests libadblockplus + filter lists 15

Limitations HTTPS traffic HTTP logs Content-Type Referrer Map Web page Hidden ads (CSS) Classification List sanity 16

Ad-blocker usage indicators Any ad-blocker AdBlock Plus Low ratio of ad Easylist updates requests Caveat: ad-free content Caveat: HTTPS + NATs How to find a suitable threshold? 17

Active measurements Browser Mode #HTTP ad-requests ratio Vanilla 57.8K 16 % Adblock Plus + all lists 48.5K <0.1 % Values for the complete set of Alexa top 1K 28.10.2015 18

When can we apply ad-ratio indicator? Vanilla: no plugin AdBP-Pa: adblockPlus + all lists At least 1K requests. Set threshold to 5% 28.10.2015 19

Data sets Trace RBN-1 RBN-2 Duration 4 days 15 and ½ hours Subscribers 7.5 K 19.7 K HTTP bytes 18.8 TB 11.4 TB HTTP requests 131.95 M 85.09 M Characterize Assess ad-blocker ad traffic prevalence 28.10.2015 20

How much ad-related traffic? 0 % 18 % 100 % Ad-related traffic Acceptable EasyPrivacy (35 %) EasyList and derivatives (55 %) ads (10 %) “Loose filters” @@||gstatic.com^$document Black-listed (57%) 28.10.2015 21

How many ABP-capable browsers? 500K pairs 1.6K unique 9.2 K users <IP, UserAgent> UserAgent strings Heavy hitters Known browsers Requests: 50 % (all) 72 % (ads) 33 % 24 % 15 % 21 % 7 % Firefox Chrome Safari Mobile IE 28.10.2015 22

Ad Ratio EasyList ✘ ✘ ✘ ✔ ✔ ✘ ✔ ✔ Browsers with no evidence of an adblocker: 46% of the ad traffic 28.10.2015 23

Ad Ratio EasyList ✘ ✘ ✘ ✔ ✔ ✘ ✔ ✔ Inconsistent indicators a) various users in the same household 28.10.2015 24

Ad Ratio EasyList ✘ ✘ ✘ ✔ ✔ ✘ ✔ ✔ Inconsistent indicators a) different plugin, or b) ad-free content 28.10.2015 25

Ad Ratio EasyList ✘ ✘ ✘ ✔ ✔ ✘ ✔ ✔ ? Evidence of Adblock Plus for 22% of the users Still generate ad traffic! 28.10.2015 26

How do users configure ABP? AGGREGATED 82% 11 % 7% EasyPrivacy Acceptable ads EasyList 99% 90% EasyPrivacy ABP users No ad blocker 87% 70% Acceptable ads Most AdBlockPlus users do not change default settings 27

Summary Can we use existing functionality to classify ad traffic? Yes, we can! (but limitations apply) 28.10.2015 28

Summary Can we use existing functionality to classify ad traffic? Yes, we can! (but limitations apply) Which traffic is potentially affected by Adblock Plus? ~18% requests and ~1% HTTP bytes Google, CDNs, Cloud Providers and dedicated ASes 10% of the ad-related traffic is white listed 28.10.2015 29

Summary Can we use existing functionality to classify ad traffic? Yes, we can! (but limitations apply) Which traffic is potentially affected by Adblock Plus? ~18% requests and ~1% HTTP bytes Google, CDNs, Cloud Providers and dedicated ASes 10% of the ad-related traffic conforms Acceptable Ads Can we measure how end users adopt ad-block plugins? Yes, in our traces ~20% of users Out-of-the-box i.e., acceptable ads / no EasyPrivacy 28.10.2015 30

Thank you! Can we use existing functionality to classify ad traffic? Yes, we can! (but limitations apply) Which traffic is potentially affected by Adblock Plus? ~18% requests and ~1% HTTP bytes Google, CDNs, Cloud Providers and dedicated ASes 10% of the ad-related traffic conforms Acceptable Ads Can we measure how end users adopt ad-block plugins? Yes, in our traces ~20% of users Out-of-the-box i.e., acceptable ads / no EasyPrivacy 28.10.2015 31

28.10.2015 32