Spatio-temporal Analysis of Reverted Wikipedia Edits Johannes Kiesel - - PowerPoint PPT Presentation
Spatio-temporal Analysis of Reverted Wikipedia Edits Johannes Kiesel , Martin Potthast, Matthias Hagen, Benno Stein < first name > . < last name > @uni-weimar.de Bauhaus-Universitt Weimar www.webis.de ICWSM-17, May 18 th 2017 1
Johannes Kiesel, Martin Potthast, Matthias Hagen, Benno Stein <first name>.<last name>@uni-weimar.de Bauhaus-Universität Weimar www.webis.de ICWSM-17, May 18th 2017
1 @KieselJohannes
2 @KieselJohannes
3 @KieselJohannes
❑ 470 million article edits to the English Wikipedia in 2003 – 2016 ❑ 40 million (9.5%) are vandalism
→ a vandalism case every 10s
4 @KieselJohannes
❑ 470 million article edits to the English Wikipedia in 2003 – 2016 ❑ 40 million (9.5%) are vandalism
→ a vandalism case every 10s Countermeasure: Bots that detect and revert vandalism Problem: False positives of the bots discourage editors
5 @KieselJohannes
How to avoid vandalism in the first place? → Understand why people vandalize Wikipedia. → Analyze when people vandalize. → Analyze where these people are.
6 @KieselJohannes
How to avoid vandalism in the first place? → Understand why people vandalize Wikipedia. → Analyze when people vandalize. → Analyze where these people are. We analyzed all 1.2 billion edits to the 7 most-edited Wikipedias
❑ Large-scale mining of vandalism using reverted edits ❑ Historical geolocation of anonymous editors by cross-checking several geolocation sources ❑ Spatio-temporal analysis revealing when anonymous editors vandalize from where
7 @KieselJohannes
8 @KieselJohannes
9 @KieselJohannes
10 @KieselJohannes
11 @KieselJohannes
❑ Not all reverted edits are vandalism ❑ Relying on non-obligatory revert comments underestimates vandalism
Identified 7 patterns of non-vandalism or ambiguous reverts
Revert to blank page
Empty revert
+
Self-revert
Revert correction (enlargement)
Reverted revert
Interleaved reverts (edit war)
Revert reverting more than one editor
Filter 67% of reverted edits Vandalism detection with precision 82.8%, recall 84.7%
12 @KieselJohannes
Hour of day 6 8 10 12 14 16 18 20 22 2 4
vandalism edits edits
1 2 3 4 5 6 7 Edits (in millions)
13 @KieselJohannes
Hour of day 6 8 10 12 14 16 18 20 22 2 4 0.0 0.1 0.2 0.3 0.4 0.5 Vandalism ratio
vandalism edits edits vandalism ratio =
14 @KieselJohannes
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4 English Wikipedia from United States Monday - Thursday Friday Saturday Sunday
Estimates from less than 1000 vandalism edits are shown as dotted lines
15 @KieselJohannes
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4
English Wikipedia from United States
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4
English Wikipedia from Canada
Monday - Thursday Friday Saturday Sunday
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4
English Wikipedia from Australia
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4
English Wikipedia from United Kingdom
Estimates from less than 1000 vandalism edits are shown as dotted lines
16 @KieselJohannes
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4 Japanese Wikipedia from Japan Monday - Thursday Friday Saturday Sunday
Estimates from less than 1000 vandalism edits are shown as dotted lines
17 @KieselJohannes
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4 French Wikipedia from France
Wednesday
Monday - Thursday Friday Saturday Sunday
Estimates from less than 1000 vandalism edits are shown as dotted lines
18 @KieselJohannes
0.12 0.16 0.2 0.24 0.28 Vandalism ratio
Country estimates from less than 1000 vandalism edits are not colored
19 @KieselJohannes
0.12 0.16 0.2 0.24 0.28 Vandalism ratio
Country estimates from less than 1000 vandalism edits are not colored
20 @KieselJohannes
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4
English Wikipedia from Germany
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4
German Wikipedia from Germany Monday - Thursday Friday Saturday Sunday
Estimates from less than 1000 vandalism edits are shown as dotted lines
21 @KieselJohannes
Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4 0.0 0.1
English Wikipedia from Germany
0.0 0.1 0.2 0.3 0.4 0.5 Hour of day Vandalism ratio 6 8 10 12 14 16 18 20 22 2 4
German Wikipedia from Germany Monday - Thursday Friday Saturday Sunday
Estimates from less than 1000 vandalism edits are shown as dotted lines
22 @KieselJohannes
Future Work
❑ Identify different types of vandalism ❑ Identify changes in vandalism behavior over the years
Resources
❑ Interactive tool for exploring the vandalism ratio graphs
webis16.medien.uni-weimar.de/wikipedia-vandalism
❑ Supplementary material (∼50 pages of tables and graphs)
github.com/webis-de/ICWSM-17/raw/master/supplementary-material.pdf
❑ Code for historical geolocation
github.com/webis-de/aitools4-aq-geolocation
❑ Code for reproducing experiments
github.com/webis-de/ICWSM-17
23 @KieselJohannes