[Paper Review] Men Are Elected, Women Are Married: Events Gender Bias on Wikipedia

Jiao Sun, Nanyun Peng|arXiv (Cornell University)|Jun 3, 2021

Wikis in Education and Collaboration4 citations

TL;DR

This paper presents the first event-centric analysis of gender bias in Wikipedia, using a curated corpus of 7,854 celebrity fragments to detect and calibrate gender-asymmetric event associations. It reveals that Wikipedia disproportionately intermingles personal life events (e.g., marriage) with professional achievements for women, while men’s career events remain separate, highlighting systemic implicit bias in editorial practices.

ABSTRACT

Human activities can be seen as sequences of events, which are crucial to understanding societies. Disproportional event distribution for different demographic groups can manifest and amplify social stereotypes, and potentially jeopardize the ability of members in some groups to pursue certain goals. In this paper, we present the first event-centric study of gender biases in a Wikipedia corpus. To facilitate the study, we curate a corpus of career and personal life descriptions with demographic information consisting of 7,854 fragments from 10,412 celebrities. Then we detect events with a state-of-the-art event detection model, calibrate the results using strategically generated templates, and extract events that have asymmetric associations with genders. Our study discovers that the Wikipedia pages tend to intermingle personal life events with professional events for females but not for males, which calls for the awareness of the Wikipedia community to formalize guidelines and train the editors to mind the implicit biases that contributors carry. Our work also lays the foundation for future works on quantifying and discovering event biases at the corpus level.

Motivation & Objective

• To investigate how gender biases are encoded in Wikipedia through the lens of event-level representations.
• To identify asymmetric associations between events and gender in public knowledge sources like Wikipedia.
• To develop a method to detect and calibrate event-level gender bias, minimizing confounding from the event extraction model.
• To raise awareness among Wikipedia editors about implicit gender biases in content structuring.
• To provide a foundation for future corpus-level analysis of event-based biases in NLP and knowledge repositories.

Proposed method

• Curated a corpus of 7,854 fragments from 10,412 celebrities across 8 occupations, including demographic, career, and personal life descriptions.
• Applied a state-of-the-art event detection model (Han et al., 2019) to extract events from the text fragments.
• Proposed a calibration technique using strategically generated templates to offset potential gender bias in the event detection model.
• Identified gender-distinct events with significantly higher occurrence rates for one gender using statistical comparison.
• Ranked detected events by frequency to verify that findings are not driven by rare or infrequent events.
• Analyzed event distribution across sections (Career vs. Personal Life) to detect structural bias in content organization.

Experimental results

Research questions

• RQ1 How are professional and personal life events distributed differently across male and female celebrities on Wikipedia?
• RQ2 To what extent do event-level associations with gender reflect societal stereotypes in public knowledge sources?
• RQ3 Can event detection models introduce confounding biases that skew the perception of gender disparities in Wikipedia?
• RQ4 Are the detected gender-biased events frequent or rare in the corpus, and do they reflect meaningful patterns?
• RQ5 How can editorial practices on Wikipedia be improved to reduce implicit gender bias in content structuring?

Key findings

• Wikipedia pages intermingle personal life events such as marriage and divorce with professional achievements for women, while men’s career events remain isolated in the Career section.
• For female celebrities, personal life events are significantly more likely to appear in the Career section, suggesting structural bias in content presentation.
• Male celebrities’ Wikipedia pages include professional events like award ceremonies in the Personal Life section, indicating a gendered asymmetry in event categorization.
• The detected gender-distinct events are not rare; many rank in the top 10% of event frequencies, confirming their representativeness and significance.
• The calibration technique successfully reduced model-induced bias, allowing for more accurate detection of corpus-level gender bias.
• The study reveals that Wikipedia’s editorial practices may reinforce traditional gender roles by framing women’s lives around relationships and men’s around achievements.