Cole Augusta
Common sense reasoning remains a fundamental challenge in artificial intelligence, requiring systems to understand implicit relationships and everyday knowledge that humans take for granted. This project presents a graph neural network (GNN) approach to common sense question answering using ConceptNet, a large-scale semantic knowledge base. We construct a knowledge graph from ConceptNet semantic profiles and train a Graph Convolutional Network to learn meaningful concept embeddings. Our system combines learned embeddings with explicit graph structure to answer natural language questions about everyday concepts. We demonstrate that this hybrid approach provides an effective foundation for common sense reasoning tasks.
Links:
- Final Paper: https://drive.google.com/file/d/1zMASgthocvLJy6NQWsCZBljNyhdfVNsv/
- Final presentation (slides): https://docs.google.com/presentation/d/11icYVahwNczEDoKppQptxc1MjE4QC4V9VaLiWtq9L4A/
- Final presentation (video): https://youtu.be/82CynAPSlWA
- github repository: https://github.com/ColeAugusta/CommonSenseQA-AI
