Matupom/wangchan-sentiment-ft

Thai sentiment classification model fine-tuned from airesearch/wangchanberta-base-att-spm-uncased on the Wisesight Sentiment dataset (3-way: pos, neu, neg).
The model is implemented with the Hugging Face Transformers library (AutoModelForSequenceClassification).

---

Model description

• Model ID: Matupom/wangchan-sentiment-ft
• Base model: airesearch/wangchanberta-base-att-spm-uncased (RoBERTa-style Thai language model)
• Architecture: CamemBERT / RoBERTa encoder with a classification head
• Task: Single-label text classification (sentiment analysis)
• Languages: Thai

Labels

This model uses the following label mapping:

• 0 → pos – positive sentiment
• 1 → neu – neutral sentiment
• 2 → neg – negative sentiment

The same mapping is stored in the config as:

model.config.id2label = {0: "pos", 1: "neu", 2: "neg"}
model.config.label2id = {"pos": 0, "neu": 1, "neg": 2}