IndoELECTRA-NER-Singgalang

IndoELECTRA NER - Singgalang Dataset

Model Named Entity Recognition (NER) untuk Bahasa Indonesia menggunakan IndoELECTRA yang di-fine-tune pada dataset SINGGALANG.

📋 Deskripsi Model

Model ini dapat mendeteksi 3 jenis entitas dalam teks bahasa Indonesia:

• Person: Nama orang
• Place: Nama tempat/lokasi
• Organisation: Nama organisasi/perusahaan

🎯 Label

Model menggunakan format BIO (Begin-Inside-Outside):

• O: Bukan entitas
• B-Person, I-Person: Entitas Person
• B-Place, I-Place: Entitas Place
• B-Organisation, I-Organisation: Entitas Organisation

🔧 Training Details

• Base Model: ChristopherA08/IndoELECTRA
• Dataset: SINGGALANG (oversampled)
• Training Strategy: Parameter-efficient fine-tuning
  • Classifier head + last 2 encoder layers (unfrozen)
  • Remaining layers frozen
• Class Weighting: Applied to handle class imbalance
• Max Sequence Length: 128 tokens
• Batch Size: 16 (with gradient accumulation steps=4)
• Learning Rate: 3e-5
• Epochs: 12 (with early stopping patience=3)

📊 Performance

Model mencapai performa yang baik pada validation set dengan F1-score tinggi untuk deteksi entitas Person, Place, dan Organisation.

💻 Usage

Instalasi

pip install transformers torch

Inference

from transformers import AutoTokenizer, AutoModelForTokenClassification
import torch

# Load model dan tokenizer
model_name = "ecaaa09/IndoELECTRA-NER-Singgalang"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)

# Fungsi untuk prediksi NER
def predict_ner(sentence):
    tokens = sentence.split()
    inputs = tokenizer(tokens, is_split_into_words=True, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model(**inputs)
    
    predictions = torch.argmax(outputs.logits, dim=-1).squeeze().tolist()
    
    results = []
    word_ids = inputs.word_ids()
    prev_word = None
    
    for idx, word_idx in enumerate(word_ids):
        if word_idx is None or word_idx == prev_word:
            continue
        label = model.config.id2label[predictions[idx]]
        results.append((tokens[word_idx], label))
        prev_word = word_idx
    
    return results

# Contoh penggunaan
sentence = "Joko Widodo bertemu dengan Prabowo di Jakarta"
results = predict_ner(sentence)

for token, label in results:
    print(f"{token:<20} {label}")

Output Example

Joko                 B-Person
Widodo               I-Person
bertemu              O
dengan               O
Prabowo              B-Person
di                   O
Jakarta              B-Place

👥 Team

Tugas Besar Natural Language Processing - Institut Teknologi Sumatera

📝 Citation

@misc{indoelectra-ner-singgalang,
  author = {Rayhan Fatih Gunawan et al.},
  title = {IndoELECTRA NER - Singgalang Dataset},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/ecaaa09/IndoELECTRA-NER-Singgalang}}
}

📄 License

Apache 2.0