Pretty Princess vs. Successful Leader: Gender Roles in Greeting Card Messages

Document Title

Pretty Princess vs. Successful Leader: Gender Roles in Greeting Card Messages

Document Information

• Subject Area: Human-Computer Interaction (HCI), Natural Language Processing (NLP), Social Computing
• Keywords: Gender role awareness, greeting card messages, visualization system, gender bias, human-computer interaction, natural language processing, information systems, gender stereotypes

Research Background and Issues

• What problems or challenges did the authors identify? Greeting card messages are a widely used medium for expressing emotions, and their content may contain gender stereotypes, such as focusing on appearance and family for women and career achievements for men. However, systematic research and tool development related to gender stereotypes in greeting card messages remain underexplored.

• Why is this issue important? Gender stereotypes can subtly reinforce societal biases toward gender roles, posing significant barriers to gender equality. As a direct medium of interpersonal communication, greeting card messages play a critical role in overcoming these biases. Additionally, with the widespread use of natural language generation technologies (e.g., GPT-2), such tools may unintentionally amplify these biases, making in-depth analysis essential.

• Research Motivation and Related Work The authors based their study on four research questions:

  1. Do greeting card messages exhibit gender stereotypes? Are women more associated with beauty and family, while men are linked to career achievements?
  2. Are gender-related topics in messages correlated with the recipient's age?
  3. Do people want to understand the gender roles in their greeting card messages?
  4. If people wish to understand gender roles, how can tools be designed to help raise their awareness?

Solution

• What methods or solutions did the authors propose? The authors proposed a visualization writing assistant named GreetA, designed to analyze the fine-grained topics in users' drafted greeting card messages and calculate their gender perception scores. The tool aims to enhance users' awareness of gender roles in their messages rather than forcibly altering the content.

• What are the innovative aspects of this solution?

  1. Dataset Construction: Compiled a dataset of 18,000 greeting card messages, including website templates, real-world social media messages, and content generated by language models.
  2. Gender Stereotype Analysis: Utilized topic modeling and statistical language processing tools (e.g., WEAT) to quantify gender-related topics.
  3. User Feedback and Tool Design: Conducted surveys to understand user needs and designed a tool to enable effective gender role awareness without enforcing content changes.
  4. Tool Features: Provided topic classification, gender-related topic scores, and topic exploration functionalities to help users intuitively understand gender roles.

• What are the implementation steps and key technologies used?

  1. Data Collection: Scraped content from eight popular greeting card websites, combined GPT-2-generated messages, and real social media data.
  2. Data Analysis:
     • Used the Empath tool for topic modeling, extracting 200 high-level topics.
     • Quantified gender topic differences using Odds Ratio (OR).
     • Verified the reliability of gender-topic associations using the Word Embedding Association Test (WEAT).
  3. GreetA Tool Development:
     • Visualized gender associations and topic contributions in user messages.
     • Provided topic exploration features to help users discover diverse expression methods.
  4. Evaluation: Conducted qualitative and quantitative user studies to test the tool's usability and efficiency.

Research Findings

• What specific results were achieved?

  • Greeting card messages indeed exhibit gender role biases, with women's cards often centered on appearance and family, while men's cards focus more on leadership and career achievements.
  • Gender biases in GPT-2-generated greeting card messages were further amplified.
  • People appreciate a non-intrusive tool to enhance their sensitivity to gender roles.

• What advantages does this solution have compared to existing ones?

  • Enabled large-scale analysis of greeting card messages.
  • Reduced intervention in content modification, supporting users' expression styles rather than enforcing changes.
  • System design directly addressed fine-grained exploration of gender roles and topic analysis, combining practicality and intuitiveness.

• What are the experimental or evaluation results?

  1. Qualitative Study: Most users found GreetA easy to use and helpful in raising awareness of gender roles.
  2. Quantitative Study: In a series of comparative experiments, users' accuracy in identifying gender roles in greeting card messages significantly increased from 70.1% to 87.7% after using GreetA.

• Limitations and Future Directions

  • The Empath tool struggles with topic extraction in specific contexts; future work could incorporate more refined topic modeling techniques.
  • The dataset primarily consists of template messages, which may differ from real-time, authentic user expressions; future studies could expand to broader social media data.
  • This study did not address non-binary gender groups; future research could explore methods for processing greeting card messages for diverse gender identities.

This study demonstrates how analyzing gender role awareness can help people better avoid unconscious stereotypes and foster a more equitable communication environment.