README.md

---
language:
- my
license: mit
tags:
- aspect-based-sentiment-analysis
- absa
- burmese
- myanmar
- multi-label-classification
- xlm-roberta
pipeline_tag: text-classification
metrics:
- f1
- precision
- recall
- accuracy
base_model:
- FacebookAI/xlm-roberta-large
library_name: transformers
---

# Myanmar ABSA - Stage 1 (Multi-Label Aspect Detection)

This model is a fine-tuned version of `xlm-roberta-large` on a balanced dataset of 7,500+ Burmese customer reviews. It predicts the presence of 6 business aspects.

## Aspect Targets:
1. `product_or_service_quality`
2. `fulfillment_and_speed`
3. `price_and_value`
4. `digital_experience`
5. `customer_support`
6. `variety_and_availability`

## Usage Example

```python
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

MODEL_NAME = "Fixaro/myanmar-absa-stage1-aspect-detection"

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)

text = "ဒီဆိုင်က ဝန်ဆောင်မှု အရမ်းကောင်းတယ်၊ ဒါပေမဲ့ ပို့ဆောင်ခ အရမ်းကြီးတယ်။"
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    logits = model(**inputs).logits
    probs = torch.sigmoid(logits).squeeze().tolist()

ASPECTS = [
    'product_or_service_quality', 'fulfillment_and_speed', 
    'price_and_value', 'digital_experience', 
    'customer_support', 'variety_and_availability'
]

for aspect, prob in zip(ASPECTS, probs):
    if prob > 0.5:
        print(f"Detected: {aspect} ({prob*100:.1f}%)")