Studying the Representation of the LGBTQ+ Community in RuPaul’s Drag Race with LLM-Based Topic Modeling

Mika HämäläinenNAACL 2025

TL;DR

This paper introduces a novel application of Large Language Models (LLMs) for topic modeling to analyze the representation of the LGBTQ+ community in RuPaul's Drag Race. By processing subtitle data from 16 seasons, the study reveals a nuanced portrayal of both empowering themes and challenges faced by the community, offering a more comprehensive understanding than traditional methods.

What Problem Does This Paper Address?

Understanding how marginalized communities are represented in mass media is crucial for shaping public perception and societal attitudes. RuPaul's Drag Race (RPDR) is a prominent cultural artifact that foregrounds LGBTQ+ themes, making it a significant subject for media analysis. Traditional topic modeling techniques, often relying on word-based analysis, can be limited in capturing the complex nuances and implicit narratives present in media texts. This paper addresses the need for a more sophisticated method to analyze the rich textual data from RPDR, aiming to uncover a spectrum of LGBTQ+ representations that go beyond surface-level analysis.

Methods

The core innovation of this paper lies in its use of LLM-based topic modeling. Specifically, the authors employed Gemini 1.5 Flash to process subtitle data from seasons 1-16 of RuPaul's Drag Race. The LLM was prompted to extract LGBTQ+-related topics and their descriptions for each episode. These extracted topic strings were then embedded using Google's text-embedding-004 model. Finally, HDBSCAN clustering was applied to these embeddings to group similar topics, followed by manual merging to refine the clusters. This approach moves beyond traditional word-frequency methods, allowing for a more semantic and nuanced understanding of the textual content.

Datasets

The primary dataset used in this study consists of subtitle files for seasons 1 through 16 of RuPaul's Drag Race, sourced from OpenSubtitles. The authors note that season 17 was excluded as it was not fully released at the time of the study. The subtitle files were cleaned of timestamps and HTML tags.

Key Insights and Ideas

  • LLMs offer a more nuanced understanding: The study demonstrates that LLM-based topic modeling can uncover a broader spectrum of themes and subtleties in media representation compared to traditional word-based methods.
  • RPDR as a "beacon of hope": While acknowledging critical perspectives on the show, the LLM analysis highlights a significant presence of empowering themes, suggesting RPDR serves as a source of hope and community for LGBTQ+ individuals.
  • Dual narrative of empowerment and challenge: The analysis reveals a consistent portrayal of both positive aspects like self-expression, community support, and self-acceptance, alongside challenges such as homophobia, internalized issues, and negative body image.
  • Beyond surface-level analysis: The LLM approach allowed for the identification of topics that might be missed by simpler methods, such as the complexities of family acceptance and the performative nature of drag.

Key Results

The LLM analysis identified 43 distinct topic clusters, which were subsequently refined through manual merging. The most commonly discussed themes included:

  • LGBTQ+ community and representation (219 occurrences)
  • Drag as a form of self-expression and artistry (187 occurrences)
  • Coming out and self-acceptance (159 occurrences)
  • Positive body image and self-love (129 occurrences)
  • The importance of community and family support (100 occurrences)

Less frequent but still significant negative themes included:

  • Mental health and resilience (54 occurrences)
  • Internalized homophobia and self-acceptance (77 occurrences)
  • Homophobia and discrimination (45 occurrences)

Limitations and Future Work

The authors acknowledge several limitations:

  • LLM biases: The "black box" nature of LLMs means that the topics identified may be influenced by unknown biases in how the model picks them.
  • Model capabilities: The use of the free version of Gemini API (Gemini 1.5 Flash) means that more advanced models like Gemini 2.0 or GPT-40 might have extracted even more topics and produced more accurate clustering.

Future work suggested includes:

  • Evolution of topics over time: Analyzing how topics have changed across different seasons of RPDR.
  • International adaptations: Studying the representation of LGBTQ+ themes in international adaptations of RuPaul's Drag Race.
  • Comparative analysis: Investigating how topics in these adaptations compare to the main series.

Summary updated

BibTeX
@inproceedings{hamalainen2025studying,
  author = {Mika Hämäläinen},
  title = {{Studying the Representation of the LGBTQ+ Community in RuPaul’s Drag Race with LLM-Based Topic Modeling}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2025},
}
Figure