File size: 1,461 Bytes
090d775
840d25c
 
 
 
 
 
 
 
090d775
 
840d25c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
---
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.