CommonsenseQA-AI

Answer common sense questions using ConceptNet and PyTorch Geometric GNNs.

Quick Start

# Install dependencies
pip install torch torch-geometric

# Build graph from ConceptNet profiles
python build_graph.py

# Train GNN model
python train_gnn.py

# Ask questions
python main.py

Example

Question: What is a dog?
Answer: A dog is a type of pet and mammal. It is loyal and friendly. Dogs can bark and run.

Question: Where do you find cats?
Answer: You can typically find a cat at home, house, or outside.

How It Works

1. Parse ConceptNet semantic profiles → nodes and edges
2. Build PyTorch Geometric graph (988 nodes, 1.3k edges)
3. Train 2-layer GCN on link prediction task
4. Answer questions using learned embeddings + graph structure

Files

• parse_profile.py - Parse ConceptNet text files
• build_graph.py - Build PyTorch Geometric graph
• train_gnn.py - GNN training and model management
• main.py - Question answering interface

Architecture

ConceptNet Profiles → Graph → GCN → Embeddings → Q&A System
                      (988 nodes)  (32-dim)

Requirements

• Python 3.10+
• PyTorch 2.0+
• PyTorch Geometric
• ConceptNet semantic profile text files in ./conceptnet_profiles/

License

MIT