README

This repository contains the code and data for Zhao et al. "Mapping the Spiral of Silence: Surveying Unspoken Opinions in Online Communities" to appear at CHI 2026.

Repository Structure

spiral-of-silence/
├── generate_viewpoints.py    # Script to generate topics and viewpoints using LLMs
├── utils.py                  # Utility functions for Reddit API and OpenAI API
├── paper_analyses.ipynb      # Python notebook for data analysis and visualizations
├── Regression.Rmd            # R markdown file for statistical modeling
├── data.csv                  # Survey data (participant responses)
├── prompts/                  # LLM prompts for topic/viewpoint generation
│   ├── topic_1.json          # Prompt for generating controversial topics
│   ├── viewpoint_1.json      # Prompt for generating viewpoints on topics
│   └── summarize_1.json     # Prompt for shortening viewpoints
└── README.md                 # This file

Setup

Prerequisites

• = Python 3.10

• R (for statistical analyses)
• Reddit API credentials
• OpenAI API key

Installation

1. Clone this repository:

git clone <repository-url>
cd spiral-of-silence

2. Install Python dependencies:

pip install praw openai python-dotenv pandas tqdm numpy matplotlib scipy jupyter

3. Create a .env file in the root directory with your API credentials:

REDDIT_CLIENT=your_reddit_client_id
REDDIT_SECRET=your_reddit_secret
OPENAI_API=your_openai_api_key

Usage

Generating Topics and Viewpoints

The generate_viewpoints.py script generates controversial topics and corresponding viewpoints for specified Reddit subreddits:

from generate_viewpoints import main

# Generate topics and viewpoints for subreddits
subreddits = ['politics', 'technology', 'science']
filepath = 'output_{}.json'  # Output file pattern
main(subreddits, filepath, model='gpt-4o')

The script performs three steps:

1. Topic Generation: Generates controversial topics that would lead to disagreement within each subreddit
2. Viewpoint Generation: Creates viewpoints representing different sides of each topic
3. Viewpoint Shortening: Condenses viewpoints to shorter statements

Data

The data.csv file contains survey responses with the following key variables:

• Participant characteristics: Age, Gender, Race, Political, Karma, Usage, Posting, WTSC
• Community characteristics: Subreddit, Diversity, Inclusion, Safety, percent (content removal rate), modratio
• Topic/Viewpoint: Topic, Viewpoint, Agreement, Sub_Agree, Incongruency
• Outcomes: Comment_Likelihood, Upvote_Likelihood

Citation

@inproceedings{zhao2026spiral,
   author = {Dora Zhao and Diyi Yang and Michael S. Bernstein},
   title = {Mapping the Spiral of Silence: Surveying Unspoken Opinions in Online Communities},
   booktitle = {Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems},
   year = {2026}
}