Add Code and Data Availability links to report (PDF/DOCX/LaTeX); add configs to repo
3025401 verified | model: | |
| name: FastText+BiLSTM | |
| architecture: BiLSTM | |
| embedding: pretrained FastText (subword) | |
| embedding_dim: 300 | |
| hidden_dim: 256 | |
| dropout: 0.5 | |
| max_seq_len: 128 | |
| training: | |
| optimizer: Adam | |
| learning_rate: 0.001 | |
| weight_decay: 0 | |
| epochs: 10 | |
| batch_size: 32 | |
| loss: binary_crossentropy | |
| data: | |
| dataset: PRISM (combined_hate_speech_dataset, Kaggle) | |
| target: hate_label | |
| classes: 2 | |
| split: stratified 70/30 holdout, train pool 60/10 train/val | |
| best_result: | |
| regime: langwise_regular | |
| strategy: english | |
| accuracy: 0.752355 | |
| f1: 0.768844 | |
| auc_roc: 0.825146 | |