CoolFace
Modelpublic

MinThu11/burmese-disaster-classifier

sourceHugging Facemitupdated 3d agoView on Hugging Face
0likes17downloads
Model Card

Burmese Disaster Social Media Classifier

Fine-tuned `xlm-roberta-base` for classifying Burmese (Myanmar) disaster-related social media posts into four actionable categories. Useful for disaster-response triage and monitoring.

Labels

LabelMeaning
Immediate_Rescue_NeededPosts requesting urgent rescue / help
Donation_CampaignPosts offering or requesting donations & aid
General_NewsNews, warnings, and situational updates
Well_Wishing_PrayerPrayers and well-wishing messages

Usage

python
from transformers import pipeline

clf = pipeline("text-classification", model="MinThu11/burmese-disaster-classifier")
print(clf("ကလေးတွေရော အဘိုးကြီးရော ရေခေါင်မိုးထိတက်လာလို့ ပိတ်မိနေပါတယ် အမြန်လာကယ်ပေးကြပါ"))

Or load directly:

python
import torch, torch.nn.functional as F
from transformers import AutoTokenizer, AutoModelForSequenceClassification

model_id = "MinThu11/burmese-disaster-classifier"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

inputs = tokenizer("...", return_tensors="pt", truncation=True, max_length=128)
with torch.no_grad():
    probs = F.softmax(model(**inputs).logits, dim=-1)[0]
print(model.config.id2label[int(probs.argmax())])

Training details

  • —Base model: xlm-roberta-base (multilingual encoder)
  • —Dataset: Myanmar disaster social media dataset (~1,000 posts, 80/20 train/test split)
  • —Epochs: 4
  • —Learning rate: 2e-5
  • —Batch size: 8
  • —Weight decay: 0.01
  • —Max sequence length: 128

Limitations

  • —Trained on a relatively small dataset (~1,000 examples); may not generalize to all disaster types or writing styles.
  • —Burmese-only; performance on other languages is not guaranteed.