Biased or Diverse? Measuring Personalization in Online Search Results

• In Event: Measuring Diversity in Algorithmically Personalized News Environments

Abstract

Personalization in online services is widely considered to have contributed crucially to the success of the Internet, yet questions still arise about the potential adverse effects that personalization may have on public discourse (Couldry & Turow, 2014). One issue regarded as a potential hazard and unintended consequence of algorithmic personalization is bias, i.e. being confronted with predominantly one-sided information. Search engines are one area where such biases may manifest, based on a unique combination of input data and algorithms (Kulshrestha et al, 2017, Hannack et al, 2016). Prior studies show that biased search rankings can influence voting preferences of undecided voters by 20% or more (Epstein et al, 2015). In our analysis, we focus on exploring the bias and diversity in the search engine results on news-related topics pertaining to the German 2017 national elections (N = 8.7 mio. URLs, N = 737,000 search queries) by presenting multiple metrics for studying the diversity in search results.

Authors

• Juhi Kulshrestha, Hans Bredow Institute for Media Research

• Cornelius Puschmann, Hans Bredow Institute for Media Research