| license: mit | |
| tags: | |
| - graph-neural-network | |
| - semantic-communication | |
| - webnlg | |
| - dart | |
| # GraphSemCom GNN (rel4593) | |
| This repo contains: | |
| - `encoder_best.pt`: GNN encoder weights | |
| - `edge_clf_best.pt`: edge relation classifier weights | |
| - `rel2id.json`: mapping of relation string -> class id | |
| - `config.json`: metadata | |
| ## How to use (concept) | |
| 1. Build a graph Data(x, edge_index, edge_type) | |
| 2. Run encoder -> h | |
| 3. Add AWGN to h (optional) | |
| 4. Predict relations with edge_clf(h[src], h[dst]) | |
| Trained with AWGN augmentation (random SNR 0–20 dB). | |