Show me a "Male Nurse"! How Gender Bias is Reflected in the Query Formulation of Search Engine Users

Title of the Paper

Show me a "Male Nurse"! How Gender Bias is Reflected in the Query Formulation of Search Engine Users

Bibliographic Information

• Research Domain: Analysis of gender bias in human-computer interaction and information retrieval
• Keywords: Search queries, information retrieval, gender bias, prototype effects, user studies

Research Background and Problem Statement

• Problem Statement: Search engines may reinforce historical gender stereotypes and societal discrimination through biased data and algorithms. However, there is limited research on whether user behavior itself reflects such biases and contributes to their propagation.
• Importance of the Research: Search engines are critical tools for information access in daily life. Addressing and mitigating gender bias is essential for fostering equitable information access and social inclusion.
• Motivation and Related Work:
  • Previous studies have primarily focused on gender bias at the algorithm and data levels, with insufficient exploration of how user behavior contributes to bias propagation.
  • This study is grounded in prototype theory from psychology, investigating user search query behavior within the context of gender stereotypes.

Proposed Solution

• Methodology: The authors conducted an online user study targeting participants from the United States and the United Kingdom to examine whether gender bias is reflected in search queries and to explore whether educational texts can enhance users' awareness of bias.
• Innovations:
  • Introduction of prototype theory to analyze gender bias in user behavior through the lens of "prototype matching and mismatching."
  • A focus on user behavior rather than algorithms, linking it to the reconstruction of biased actions.
• Implementation Steps:
  1. Prepare documents containing gender-related or neutral texts as materials for user tasks.
  2. Instruct users to generate search queries that match the content of these documents.
  3. Analyze whether users' queries explicitly mention gender and whether they exhibit "prototype" or "anti-prototype" tendencies.
  4. Investigate the influence of individual characteristics (e.g., political orientation, gender identity) on gender-biased behavior.
  5. Incorporate educational texts into the main experiment to assess their impact on users' search behavior.

Research Findings

• Key Results:
  1. Users demonstrated gender bias in their search queries: in stereotype-defying contexts (e.g., male caregivers), users were more likely to explicitly mention gender.
  2. Conservative political views were associated with higher tendencies toward gender bias, while gender and education level had no significant impact.
  3. Educational texts had a weak to moderate effect in reducing bias among certain user groups, particularly men.
• Advantages Over Existing Solutions:
  • Unlike traditional studies focusing on algorithmic bias, this research provides deep insights into user behavior bias.
  • Introduced quantifiable measures for assessing gender bias (e.g., "Mention Gap Index").
• Experimental or Evaluation Results:
  • The Mention Gap was quantified at 0.43, indicating significantly higher mentions of non-prototype characteristics compared to prototype ones.
  • Educational texts significantly influenced male users, while showing a reverse effect on female users.
• Limitations and Future Directions:
  • The study is limited to a traditional binary gender framework and does not address more complex gender expressions.
  • Results are based on specific cultural contexts (UK and US); future research should expand to multicultural settings.
  • Investigate the integration of user educational information into search engine interfaces and its practical effects.