| --- |
| language: multilingual |
| license: mit |
| tags: |
| - text-classification |
| - political-science |
| - euroscepticism |
| - parliamentary-speech |
| base_model: jhu-clsp/mmBERT-base |
| --- |
| |
| # EU_Ukraine |
| |
| Binary classifier: is a parliamentary sentence about the European Union (EU=1) |
| or not (EU=0)? Fine-tuned from `jhu-clsp/mmBERT-base` on hand-annotated parliamentary |
| speeches, including Ukrainian Rada data alongside other European parliaments. |
| |
| ## Labels |
| - `0` — Non-EU |
| - `1` — EU |
| |
| ## Training |
| - Base model: `jhu-clsp/mmBERT-base` |
| - Max sequence length: 320 |
| - Train/val/test split: leakage-safe (StratifiedGroupKFold on country × speech_ID) |
| - Loss: cross-entropy with balanced class weights |
|
|
| ## Usage |
| ```python |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification |
| import torch |
| |
| tok = AutoTokenizer.from_pretrained("LBenoit/EU_Ukraine") |
| mdl = AutoModelForSequenceClassification.from_pretrained("LBenoit/EU_Ukraine") |
| |
| text = "The European Commission proposed new climate targets." |
| enc = tok(text, truncation=True, max_length=320, return_tensors="pt") |
| with torch.no_grad(): |
| logits = mdl(**enc).logits |
| probs = torch.softmax(logits, dim=-1)[0] |
| pred = int(probs.argmax().item()) |
| prob = probs[1].item() |
| print(pred, prob) |
| ``` |
|
|
| ## Intended use |
| Research on parliamentary discourse about the EU. Outputs reflect the |
| training corpus and annotation scheme; downstream prevalence estimates |
| should ideally be calibrated against a base-rate-representative sample. |
|
|
