Reducing Misinformation in Query Autocompletions
Djoerd Hiemstra <hiemstra@cs.ru.nl>

Open Search is a BIG deal!

• 
• CC BY-ND 2.0 by Cornelia Kopp

... technically not THAT big a deal.

• 
• "Google 1997" (Dean 2009)

1. To improve the service

• 
• Google search

2. Targeted advertisements

• 
• Vintage pain killer advertisement (1908)

Lawsuits over autocompletions

• Lawsuit France (2010): "cnfdi arnaque"
• Lawsuit Italy (2011): "[name] truffatore"
• Lawsuit Germany (2012): "bettina wulff prostituierte"
• Lawsuit Japan (2012): shows crimes by a man with the same name...

• Google update (2013) filters offensive completions for person names.

Autocompletion Stereotypes

• P. Baker and A. Potts (2013). Why do White people have thin lips? Google and the perpetuation of stereotypes via auto-complete search forms. Critical Discourse Studies 10(2), 187–204.

• Google can produce suggested terms which could be viewed as racist, sexist or homophobic (...)
  Gay and black identities appeared to attract higher numbers of questions that were negatively stereotyping
   
  

Harmful Autocompletions

• 
• C. Cadwalladr (2016) Google is not ‘just’ a platform. It frames, shapes and distorts how we see the world. The Guardian. 11 December 2016.

Experiment design

1. Autocompletions from web content:
   anchor texts from the ClueWeb09 web crawl
   by Callan and Hoy (2009)
2. Autocompletions from user data:
   query log by Pass et al. (2006)

Experimental results

Conclusion

• Better prediction for longer queries...
• ... but need another key-stroke for short queries
• Improvements:
  • Ask users about the usefulness of autocompletions
  • Ask users about the ethics of autocompletions
  • Linguistic knowledge to improve predictions
  • PageRank and Spam scores to improve ethics
    (links can also be manipulated)
• arXiv 2007.02620v2

Open Search Outlook

• Research challenge: Content beats behavioral data!
• Do NOT track users/queries/clicks;
• Others (Ecosia, Qwant, Startpage) will add cool features.
• Research challenge: Federated search beats monolithic engines

Acknowledgments

Thanks