on microtargeting socially divisive ads
play

On Microtargeting Socially Divisive Ads: Mahmoudreza Babaei A Case - PowerPoint PPT Presentation

Filipe N. Ribeiro, Koustuv Saha, Mahmoudreza Babaei, Lucas Henrique, Johnnatan Messias, Fabricio Benevenuto, Oana Goga, Krishna P. Gummadi, and Elissa M. Redmiles. 2019. On Microtargeting Socially Divisive Ads: A Case Study of Russia- Linked Ad


  1. Filipe N. Ribeiro, Koustuv Saha, Mahmoudreza Babaei, Lucas Henrique, Johnnatan Messias, Fabricio Benevenuto, Oana Goga, Krishna P. Gummadi, and Elissa M. Redmiles. 2019. On Microtargeting Socially Divisive Ads: A Case Study of Russia- Linked Ad Campaigns on Facebook. In Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT* '19). ACM, New York, NY, USA, 140-149. DOI: https://doi.org/10.1145/3287560.3287580 Filipe N. Ribeiro* Koustuv Saha * On Microtargeting Socially Divisive Ads: Mahmoudreza Babaei A Case Study of Russia-Linked Ad Lucas Henrique Campaigns on Facebook Johnnatan Messias Fabricio Benevenuto Oana Goga Krishna P. Gummadi Elissa M. Redmiles

  2. 2016 U 2016 U.S. P Pres esiden dential al E Elec ections ns 2

  3. Soci ocial Media “Targ rgeted” ” Adverti rtising Soci ocial Media Adverti rtising 3

  4. These ads… - exploited Facebook’s targeted advertising platform - mostly contained divisive or polarizing content # Views by topic (Source: USA Today)

  5. Ta Targeted - Facilitates selecting ad audience - These websites gather detailed demographic , Ad Advertising behavioral , location , and interest profiles of their users - Enables “microtargeting”: choose small number of users with very specific attributes 5

  6. Re Research Objectives - How divisive is the content of the IRA ads? - Were the targeting of the ads effectively done? - What features of the Facebook Ads platform were leveraged in targeting the audience of the ads? 6

  7. DATA DA Impressions 10 6 Finance 0.56 Clicks Cost (USD) 10 5 Ads 10 4 Amount 10 3 Healthcare 0.8 10 2 US Election 10 1 0 Fitness 1 Jun − 2015 Jul − 2015 Aug − 2015 Sep − 2015 Oct − 2015 Nov − 2015 Dec − 2015 Jan − 2016 Feb − 2016 Mar − 2016 Apr − 2016 May − 2016 Jun − 2016 Jul − 2016 Aug − 2016 Sep − 2016 Oct − 2016 Nov − 2016 Dec − 2016 Jan − 2017 Feb − 2017 Mar − 2017 Apr − 2017 May − 2017 Jun − 2017 Jul − 2017 Aug − 2017 - 3,517 Ads Retail 1.6 - Each with corresponding cost, impressions, clicks - Avg. IRA 10.8 - Cost per Ad: 34.5 USD, - Impressions: 11,536, 0 5 10 - Clicks: 1,062 Avg. CTR (%) of Facebook Ads 7

  8. Hig High h Im Impac pact t Ads ds - Union of top 10% ads in each of cost, clicks, impressions, and CTR - Run prior to the 2016 U.S. Elections - >80% of cost, clicks, impressions - 485 such ads

  9. RQ1: Divisiveness of the IRA Ads - U.S census-representative surveys (40% liberals, 40% conservatives, 20% moderate or neutral) - 15-30 responses per ad - Total 2,886 unique respondents If they would report Reaction: Approval / If they could the ad shown Disapproval identify false claims False Claims Reporting Approval 9

  10. Reporting For over 73% of the ads, at least 20% - Whether they would respondents reported the ad. “report / flag the content” of the ad shown. It’s sexually inappropriate It’s violent or prohibited content - If they would, why do they It’s offensive find the ad inappropriate It’s misleading or a scam I disagree with it It’s a false news story Incites racism Its a spam Something else Creates divide in 0 5 10 15 20 25 30 the society % Reported Responses

  11. RQ1: Divisiveness of the IRA Ads Ideological divisiveness False Claims Reporting Approval 11

  12. Reported by the Liberals Reported by the Conservatives Not Reported by the Conservatives Not Reported by the Liberals 12

  13. RQ2: How well was the targeting done? 0 . 8 1 . 5 0 . 45 Non-Targeted Non-Targeted Non-Targeted 0 . 7 0 . 40 Targeted Targeted Targeted 1 . 0 Proportion Reported 0 . 35 0 . 6 Approval Score Proportion FCs 0 . 30 0 . 5 0 . 5 0 . 25 0 . 4 0 . 20 0 . 0 0 . 3 0 . 15 0 . 2 0 . 10 − 0 . 5 0 . 1 0 . 05 0 . 0 0 . 00 − 1 . 0 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 Ad Ad Ad False Claims Reporting Approval 13

  14. RQ3: How the Facebook Ads Platform helped? Attribute-based targeting - Demographics (gender, age, location, language) - Interests (eg. interested in African-American civil rights movement) - Behaviors (eg. Individuals with ethnic affinity: African-American) Facebook attribute suggestion. (eg. similar pages with demographic skewness): 64% of the ads used this feature 14

  15. - Characterized the Russia-linked ad campaign on Im Implic plicatio tions ns Facebook - These ads were ideologically divisive - Ten times more effective than a typical Facebook ad - Biased especially in race and political leaning - Ads targeted the users who are less likely to identify the inappropriateness - Facebook Ads platform can likely be abused to create social discord 15

  16. Filipe N. Ribeiro, Koustuv Saha, Mahmoudreza Babaei, Lucas Henrique, Johnnatan Messias, Fabricio Benevenuto, Oana Goga, Krishna P. Gummadi, and Elissa M. Redmiles. 2019. On Microtargeting Socially Divisive Ads: A Case Study of Russia-Linked Ad Campaigns on Facebook. In Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT* '19). ACM, New York, NY, USA, 140-149. DOI: https://doi.org/10.1145/3287560.3287580 http://www.socially-divisive-ads.dcc.ufmg.br/ Thank You koustuv.saha@gatech.edu | koustuv.com 16

Recommend


More recommend