EmojiFold: Emoji Semantic Manifold Discovery

Discover the mathematical structure underlying emoji semantics through systematic analysis of oppositional pairs.

🚀 Quick Start

# Initialize project
cd /Users/rob/repos/emojifold
uv venv
source .venv/bin/activate
uv pip install -e .

# Database location: ~/.emojifold/emojifold.db
# Run small test
emojifold test --pairs 50

# Run full overnight analysis
emojifold batch --model all --output results/

# Calculate model centroids
python calculate_centroids.py --db ~/.emojifold/emojifold.db

🎯 Project Goals

• Mass computation of semantic distances between emoji pairs
• Discovery of strongest oppositional dimensions in emoji space
• Cross-model validation of semantic manifold structure
• Efficient storage and analysis of large-scale results

🔬 Key Research Questions

1. Mathematical vs Visual: How do math-oriented models compare to general embeddings?
2. Universal Oppositions: Which emoji pairs are strong across all models?
3. Dimensional Structure: What are the fundamental axes of emoji semantics?

📊 Expected Results

With ~3,000 emoji testing ~9 million pairs overnight, we should discover:

• Top 100 strongest oppositional pairs
• Model-specific vs universal patterns
• Natural clustering of semantic dimensions
• Validation of our ⚫⚪ and ➕➖ findings

🛠️ Technical Stack

• UV for dependency management
• Ollama + HuggingFace for embeddings
• Optimized for M2 Ultra (64GB) + optional 4090
• SQLite for efficient result storage
• Async/parallel processing for maximum throughput

---

Ready to map the emoji semantic universe! 🌌