The Role of Search Engines in the Amplification and Suppression of LGBTIQ+ Polarization

TL;DR

This study addresses how search engines contribute to amplification or suppression of polarization around LGBTIQ+ issues in Europe. It combines active, automated monitoring of four engines across eight countries with a large multilingual query set and NewsGuard domain metadata to quantify exposure to low-credibility, politically slanted content. The main finding is that the choice of search engine is the dominant determinant of polarizing exposure, with Mojeek and Yandex producing substantially more low-credibility, right-leaning results than Google and Bing. The work highlights important implications for information integrity and democratic discourse, showing that differences between search technologies—not user language or location—drive the degree of polarization web users encounter, and underscores the need for policymakers and designers to account for engine diversity in addressing online polarization.

Abstract

Search engines are used and trusted by hundreds of millions of people every day. However, the algorithms used by search engines to index, filter, and rank web content are inherently biased, and will necessarily prefer some views and opinions at the expense of others. In this article, we examine how these algorithmic biases amplify and suppress polarizing content. Polarization refers to a shift toward and the acceptance of ideological extremes. In Europe, polarizing content in relation to LGBTIQ+ issues has been a feature of various ideological and political conflicts. Although past research has focused on the role of social media in polarization, the role of search engines in this process is little understood. Here, we report on a large-scale study of 1.5 million search results responding to neutral and negative queries relating to LGBTIQ+ issues. Focusing on the UK, Germany, and France, our analysis shows that the choice of search engine is the key determinant of exposure to polarizing content, followed by the polarity of the query. Location and language, on the other hand, have a comparatively minor effect. Consequently, our findings provide quantitative insights into how differences between search engine technologies, rather than the opinions, language, and location of web users, have the greatest impact on the exposure of web users to polarizing Web content.

Figures (14)

• Figure 1: Search results per country returned by all four search engines combined. Panel (a) and (b) show how the political orientation of these search results are distributed per country, for queries about topics and individuals, respectively. A greater proportion of politicized content is seen in the UK in comparison to France and Germany, and there tends to be more left-leaning content for queries about individuals. Panel (c) and (d) show how the same search results are distributed according to credibility score, for queries about topics and individuals, respectively. These search results are in response to both neutral and negative queries. Political orientations and credibility scores are provided by NewsGuard (see Section \ref{['sec:newsguard']}).
• Figure 2: Focusing on search results for all three countries combined, panels (a) and (b) show how the political orientation of these search results are distributed by search engine, for queries about topics and queries about individuals, respectively. Panels (c) and (d) show how the same search results are distributed by credibility score. Mojeek and Yandex show more content from low credibility sources. These search results include responses to both neutral and negative queries.
• Figure 3: For each search engine, the proportions of search results associated with one of 3 markers of problematic content: a credibility score below 60, a record of hosting conspiracy-related content, and a record of disinformation-related content. All markers are based on NewsGuard metadata. These results combine results from Germany, France, and the UK, and include search results relating to both neutral and negative topic queries.
• Figure 4: For each search engine, the distribution of credibility scores for each political orientation label. These distributions have been normalized to account for differences in the absolute numbers for each orientation label and search engine. Appendix \ref{['app:counts_per_se_label_and_cred_range']} details these absolute numbers.
• Figure 5: For each country, and for neutral and negative queries, the mean political orientation and credibility scores for queries about topics. The political orientation score assumes that left-leaning domains have a score of -1, right-leaning domains have a score of 1, and domains without a political orientation have a score of 0. Error bars represent standard deviations.