1 of 16

Accountability in Search Engine Manipulation:

A Case Study of the Iranian News Ecosystem

Peter Carragher

2 of 16

What is a Link Scheme?

  • Any site that creates “low-value content primarily for the purpose of manipulating linking and ranking signals”
  • Motives:
    • Promote an agenda (information operation)
    • Sell as a service (paid link scheme)
  • Can we tell if a link scheme is likely part of an information operation?

Peter Carragher, Evan M. Williams, Kathleen M. Carley. 2024. “Detection and Discovery of Misinformation Sources using Attributed Webgraphs” upcoming in ICWSM 2024

3 of 16

Previous work focused on unpaid SEO

3

3

Evan M. Willilams, Kathleen M. Carley. 2023. “Search engine manipulation to spread pro-Kremlin propaganda” in Harvard Kennedy School Misinformation Review, 2023

4 of 16

Paid Link Schemes serve disreputable customers across site categories

  • Linking to multiple site categories is rare
  • To multiple unreliable categories even less so
  • Sites that do this tend to be paid link schemes
  • Why? Most often as a paid SEO service
  • Why unreliable? Reputation is not valued

Peter Carragher, Evan M. Williams, Kathleen M. Carley. 2024. “Misinformation Resilient Search Rankings with Webgraph-based Interventions” ACM Transactions on Intelligent Systems and Technology.

5 of 16

5

6 of 16

Paid Link Scheming is (transparent) black-hat SEO

7 of 16

How do we know if a scheme is paid or unpaid?

8 of 16

How do we know if a scheme is paid or unpaid?

9 of 16

Case Study: SEO in the Iranian News Network

10 of 16

GPT can label the political bias of Iranian news sites

(in the absence of existing labeled datasets)

  • Of 31 labels from previous work, 27 are labeled correctly by GPT
  • Webgraph shows assortativity along political alignment given by GPT labels

Alignments:

  • Conservative (n=150): fundamentalist, pro-government
  • Neutral (n=137): highly self-censored news outlets
  • Reformist (n=29): “left-wing” movement
  • Anti-government (n=17): controversial and largely banned in Iran

Zanconato, A., Sabahi, F.: Iran-media landscape.

European Journalism Centre (2019)

11 of 16

Topics covered tend to have partisan framing (courtesy of google translate)

12 of 16

What do the paid vs clique link schemes link to?

13 of 16

Topics linked to by paid schemes (mixed)

and the link scheme clique (conservative)

14 of 16

Paid schemes link to Iranian News sites from every political alignment

15 of 16

Clique has unique link distributions

16 of 16

Conclusion: is the link scheme clique paid for or

is it part of an information operation?

  • Outlier Node/Edge Detection
  • Node Prediction
    • Add ‘owner’ / host nodes
  • Network-based Collusion Detection Models
    • Lack of ground truth for evaluation

Hosting

Camo

# Links

Topics

Webgraph

Outlink Types

Only paid links?

Paid

Joint

No

~10

Mixed

Mixed

Multi-category

Probably

Clique

Joint

Yes

~20M

Cons.

Cons.

News

Possibly

Hosting

Camo

# Links

Topics

Webgraph

Outlink Types

Paid

Joint

No

~10

Mixed

Mixed

Multi-category

Clique

Joint

Yes

~20M

Cons.

Cons.

News