Do Search Engines Suppress Controversy?
(How would we know?)

• The Web has inherent biases
• Controversy does matter in web content
  
• Controversy suppression can be experimentally measured
• 5 case studies tell us  numerically and by causes
• There could be a more Objective Web
• Bottom Line: Search, you get the "sunny side", but why...?

Inherent Biases of the Web

• Search engines don't cover the whole web
• Crawl sites to  index pages (estimated 20B)
• Each index only part of  "visible web" (<25%)
  
• Index different parts of web (30% overlap)
• Omit "Invisible web" dynamic and databases (5-100x larger)
• Rank by proprietary algorithms (mostly  links)
  
• Exist off advertising, referrals
• Indexical bias abounds!
  
  
• Web is open, cheap, but not fair
  
• Publishing != Retrievability != Visibility
• Big, traditional organizations dominate web traffic
• "Googlearchy" = search engines (media owners) rule by links and size
  June 2003 FCC ruling assumed  enough Web outlets for opinion, "little guy". NOT!
  
  
• The Web exhibits "power laws"  (like physics, ecology)
• A few sites get most  links & traffic (90% to top 10), most sites get 1
• Also "rich get richer", sites with many links get increasingly more over time
• Links are the currency ($$) of the Web!
• Is there a more "Objective Web"?
  
  
• Background Studies
• NEC Research Lab: size of web, linking patterns, modeling the web
  
• Xerox PARC/HP, Notre Dame: power laws, surfing patterns, information ecology
• Educational/Political Studies: Wellesley, Nissenbaum/Latrona, Hindman (Googlearchy)
  
• CCNY: Mowshowitz benchmarking bias
• Consumer WebWatch, Webmaster World, Search Engine Watch: marketing via search engines
  
  
• Summary: Interlocking communities determine web behavior and biases
  
• Page authors: keywords, structure content, bestow links to other sites
• Search engines: crawl links, rank pages, collect revenues
• Searchers/Surfers: favor engine, choose paid or top links, click through/stay

Why Controversy Matters

1. Life-critical, business-critical decisions might miss facts or disputes
• Drug or treatment side effects
• Educational or career path
  
2. Controversies expand the richness and depth of topics, dramatize change
   
• Human dimensions of technical topics
• Shifts in roles in society, Political undercurrents
• History gets revised
  
3. Professionals need to report, investigate, and analyze controversies
   
• Journalist looks for different sides to story, expose differences
• Scientist finds good problems in controversy, must address contradictions
• Librarians select fairly, inform openly
• Intelligence analysts mine disagreements, uncertainty
  

Measuring Controversy - Experimental Methodology

1. Choose topics: broad, controversial subtopic, factual subtopic
2. Search using multiple engines (Google,FAST,Teoma,Profusion,Copernic) (top 50)
   
3. Browse and categorize: deep, revealing, other, off-topic
4. Summarize: amount of deep and revealing
5. Identify and explain suppressing and revealing factors
6. Assess validity and reliability of methodology

Results of Experiments

Revealing/Suppressing Factors

Toward a More Objective Web:
From Power Laws to Objectively Distributed Links

• 30% of pages were objective and analytic
• web authors linked 40% to objective, 30% to opposing, 30% to agreeing pages
  (fair science citation policy)
• search engines balanced
  
  • 50% analytic (journals, criticism, studies)
    
  • 50% organizational (institutions, associations)

Conclusions from Experiments

1. Experimental methodology is inconclusive, needs refining, but raises good questions
2. Controversies are on the web, but hard to find, miss interesting and important content
   
3. Search technology is biased  to present the "sunny side" of topics.
   You have to search harder for the "dark side".
   
4. Why? Because engines reflect authors' links and searchers' choices
   
5. Many suppressing/revealing influences are controllable by authors, engines, searchers
   
• searchers: beware the sunny side,  actively seek controversy when it matters
• page authors: link more objectively, bestow links carefully
• engines: balance organizational and analytic web content