Summary:

• Leveraging advanced natural language processing techniques, this project investigates long-term emotional expression among openly queer users on Twitter (now X.com) from 2012 to 2017. The analysis spans two landmark U.S. Supreme Court rulings on same-sex marriage (2013, 2015), revealing how sentiment patterns evolved in response to major sociopolitical events.

• Key findings: LGBTQ+ users expressed more positive emotions than non-LGBTQ+ users, suggesting emotional resilience and the supportive nature of digital communities during major political events.

Team & Collaboration:

• Kyle Ratner Ph.D. UC Santa Barbara

Challenge

• Identifying relevant users: I needed a rough but reasonable way to detect Twitter users who were openly LGBTQ+ or discussing their identity (e.g., using emojis, hashtags, or self-descriptions such as 🏳️‍🌈).
• Establishing user-centered relevance: The study addresses a research gap in understanding how queer emotional expression evolves over time in response to sociopolitical events — a need often overlooked in sentiment and visibility research.
• Data scale and accessibility: Managing millions of tweets required building a custom web scraper and designing efficient data pipelines.
• Computational constraints: Running BERT-based sentiment analysis on such a large dataset demanded substantial compute resources and optimization strategies.

Solutions to the challenge

• User detection pipeline: Wrote custom Python scripts to extract user profile metadata and identify openly LGBTQ+ users based on self-descriptions, emojis, and hashtags.
• Web scraping infrastructure: Customized the open-source Twint package to build a scalable Twitter scraper, distributed across multiple lab machines for parallel data collection.
• High-performance sentiment analysis: Leveraged UCSB’s High Performance Computing (HPC) cluster to run BERT-based sentiment models efficiently, overcoming the limitations of personal hardware.

Goal

At its core, the goal was to understand the affective experiences of LGBTQ+ users through their digital expressions — seeing online sentiment as one potential signal of long-term well-being in the face of systemic change.

Why it matters

• Understand how government rulings shape public discourse and online emotions.
• Explore whether structural events like court rulings contribute to community resilience or distress among stigmatized groups.
• Assess whether sentiment on social media can serve as a proxy for public well-being.

Research Process

Stage 1: Identifying the Problem

• Reviewed prior literature showing that queer individuals report higher rates of depression, anxiety, and reduced emotional well-being, often linked to structural stigma and social exclusion.
• Noted that major legal rulings—such as those affirming same-sex marriage—can shift public norms, reduce perceived stigma, and potentially promote better mental health outcomes.
• Identified a gap in existing research: while emotional spikes around rulings are well-documented, few studies examine long-term emotional trajectories among stigmatized communities.
• Hypothesized that following landmark rulings, openly queer users would display more positive sentiment in the years after the ruling compared to the years before—reflecting increased visibility, solidarity, and resilience.
• Defined the central research question:

  How does queer emotional expression shift online in response to major government rulings??

Stage 2: Study Design

• User focus: Targeted everyday openly LGBTQ+ users with fewer than 1,000 followers to minimize bias from influencers or institutional accounts.
• User categorization: Identified openly LGBTQ+ users through profile cues such as emojis, hashtags, and bio descriptions.
• Pilot phase:
  • Tested keyword filters and categorization logic on small data subsets.
  • Refined classification to ensure clear differentiation between user types and ruling periods.

Data Collection Strategy:

1. Scraped tweets referencing key rulings using hashtags like #SCOTUS and #LoveWins.
2. Compiled a list of users who engaged with these rulings.
3. Scraped all tweets from these users between 2012–2017, forming a longitudinal dataset.
4. Processed tweets using BERT to generate a sentiment score for each tweet.
5. Calculated average sentiment scores per user per year (2012–2017).
6. Categorized users’ sexual orientation based on profile cues (e.g., 🌈 emojis, hashtags, bios) and tweet content (e.g., “coming out”).
7. Included only non-influencer accounts (<1,000 followers) to preserve everyday emotional expression.

Schematic Design

Stage 3: Data Collection

Programming Language: Python

• Libraries & Tools:
  • Twint + Twitter API for data scraping
  • Pandas, NumPy for data cleaning
  • OpenAI’s BERT for sentiment scoring
• Dataset:
  • Collected ~12 million tweets from 2950users (1520 openly LGBTQ+ users).
  • All users tweeted at least once per year from 2012 to 2017
  • Users identified through hashtags like #SCOTUS and #LoveWins.
  • Inclusion criteria: Tweeted at least once per year (2012–2017), with U.S. location specified.
• Category Identification:
  • Users self-identified as openly LGBTQ+ via emojis, profile text, or hashtags (e.g., 🏳️‍🌈).
  • Others classified as non-openly LGBTQ+ as a comparison group.
• Final Dataset for Analysis: Focused on 2013 and 2015, the two key ruling years.

Stage 4: Analysis

• Sentiment Scoring:
  • BERT model fine-tuned for sentiment classification.
  • Tokenized tweets → context-aware embeddings → sentiment scores (−1 = negative; +1 = positive).
• Statistical Modeling:
  • Linear Mixed Models in R.
  • Random intercepts for “username” to capture within-user variation.
  • Fixed effects: Month, Category (LGBTQ+ vs. Non-LGBTQ+), and their interaction.
• Analytic Focus:
  • Temporal trends in positivity.
  • Comparison across ruling months (especially June, July).
  • Evaluation of within-group sentiment variability.

Stage 5: Results

2013 Ruling

• Openly LGBTQ+ tweets were more positive overall (β = 0.1651, p < .001).
• Significant monthly variation, with peaks around June–August (the ruling period).
• However, from July–October, LGBTQ+ sentiment declined, showing less positivity than non-LGBTQ+ users.
• Interpretation: The initial uplift from the ruling might have been tempered by backlash, social debate, or continuing discrimination.

2015 Ruling

• Openly LGBTQ+ users again showed higher baseline positivity (β = 0.156, p < .001).
• Sentiment rose in June (ruling month) but dropped sharply July–December (βs between −0.02 to −0.04, p < .001).
• Possible Explanation: Post-ruling backlash or lingering interpersonal discrimination may have dampened long-term emotional benefits.

Additionally, seasonal trends were observed, including a sentiment dip in 2016 (possibly linked to Trump’s election)

Stage 6: Interpretation

• Unexpected Finding: LGBTQ+ tweets were generally more positive overall, contradicting assumptions about persistent negative affect due to marginalization.
• Possible Explanation: Reflects community resilience and positive self-expression patterns online.
• Psychological Context: Aligns with research showing minority group resilience (e.g., Crocker et al., 1994).
• Temporal Dynamics: Institutional recognition (rulings) may shift perceived norms rather than emotional tone—positivity spikes may be short-lived.
• Social Implications: Legal wins may not immediately alleviate interpersonal discrimination or structural stigma.

Reflection

• The study highlights how digital platforms serve as emotional ecosystems—spaces for both coping and celebration.
• Offers evidence that social media can amplify positive affect within stigmatized communities following structural change.
• Future steps include:
  • Comparing emotional trends across specific LGBTQ+ subgroups (e.g., gay vs. lesbian users).
  • Exploring non-verbal communication (emoji use, visual media).
  • Linking online sentiment to offline well-being indicators (e.g., community health metrics, survey data).

Programming & Tools Summary