This is a fine-tuned version of the XLM-RoBERTa model for sentiment analysis to classify khmer texts into 2 categories; Postive and Negative.
It can process texts up to 512 tokens and performs well on khmer text inputs.
Task: Sentiment analysis (binary classification).
Languages Supported: Khmer.
Intended Use Cases:
- Analyzing customer reviews.
- Social media sentiment detection.
Limitations: - Performance may degrade on languages or domains not present in the training data. - Does not handle sarcasm or highly ambiguous inputs well.
The model was evaluated on a test set of 400 samples, achieving the following performance:
Test Accuracy: 83.25%
Precision: 83.55%
Recall: 83.25%
F1 Score: 83.25%
Confusion Matrix:
Predicted\Actual | Negative | Positive |
---|---|---|
Negative | 166 | 42 |
Positive | 25 | 167 |
The model supports a maximum sequence length of 512 tokens. |
How to Use
from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("tykea/khmer-text-sentiment-analysis-roberta")
model = AutoModelForSequenceClassification.from_pretrained("tykea/khmer-text-sentiment-analysis-roberta")
text = "អគុណCADT"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
outputs = model(**inputs)
predictions = outputs.logits.argmax(dim=1)
labels_mapping = {0: 'negative', 1: 'positive'}
print("Predicted Class:", labels_mapping[predictions.item()])
- Downloads last month
- 122
Inference Providers
NEW
This model is not currently available via any of the supported third-party Inference Providers, and
the model is not deployed on the HF Inference API.
Model tree for tykea/khmer-text-sentiment-analysis-roberta
Base model
FacebookAI/xlm-roberta-base