ikhlamal/sahabat-ai-indonesian-news-analytics

Sahabat-AI Indonesian News Analytics

Model fine-tuned untuk analisis teks komentar berita bahasa Indonesia dengan tiga tugas utama:

Tasks

1. Sentiment Analysis - klasifikasi sentimen: positif, negatif, netral
2. Hate Speech Detection - deteksi ujaran kebencian: hate, not_hate
3. Stance Detection - deteksi sikap: favor, against, neutral

Model Details

• Base model: GoToCompany/gemma2-9b-cpt-sahabatai-v1-instruct
• Fine-tuning method: LoRA (r=16, alpha=16, dropout=0.1)
• Quantization: 4-bit (NF4)

Datasets Used

1. NusaX — Sentiment analysis
2. IndoLEM — Sentiment analysis
3. Ialfina — Hate speech detection
4. Okky Ibrohim — Hate speech detection
5. Indonesian COVID-19 Vaccination Stance — Stance detection

Training Configuration

• Epochs: 5 per dataset
• Batch size: 2
• Gradient accumulation steps: 4
• Learning rate: 2e-4
• Optimizer: AdamW 8-bit
• Weight decay: 0.01
• LR scheduler: Linear
• Warmup steps: 5
• Seed: 3407

Intended Use

Model ini dirancang untuk:

• Analisis sentimen berita dan opini publik berbahasa Indonesia
• Moderasi konten (deteksi ujaran kebencian)
• Analisis stance dalam diskusi publik
• Penelitian NLP bahasa Indonesia

Citation

@misc{sahabat-ai-news-analytics-2024,
  author = {Ikhlamal},
  title = {Sahabat-AI Indonesian News Analytics: Multi-task Text Classification for Indonesian},
  year = {2024},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/ikhlamal/sahabat-ai-indonesian-news-analytics}}
}

Acknowledgments

• Base model: Sahabat-AI team (GoToCompany)
• Dataset contributors: NusaX, IndoLEM, Ialfina, Okky Ibrohim, dan Indonesian COVID-19 Vaccination Stance authors

License

Apache 2.0

Contact

Untuk pertanyaan atau masukan, buka issue di repositori model atau hubungi pemilik model.