IMPLEMENTASI FINE-TUNED MODEL LLAMA3 PADA HUGGING FACE UNTUK MENDETEKSI TEKS PROMOSI JUDI ONLINE

Muhammad Rafid Miftah Fadhil, - and Indira Syawanodya, - and Yulia Retnowati, - (2025) IMPLEMENTASI FINE-TUNED MODEL LLAMA3 PADA HUGGING FACE UNTUK MENDETEKSI TEKS PROMOSI JUDI ONLINE. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Judi online merupakan permasalahan serius di Indonesia karena dampak sosial dan ekonomi yang ditimbulkan, diperparah oleh kemudahan akses serta strategi promosi yang masif. Penelitian ini bertujuan untuk mengimplementasikan model Large Language Model (LLM) Llama 3.2 berbasis library Hugging Face untuk mendeteksi promosi judi online, serta menganalisis performanya tanpa dan dengan teknik optimasi. Metode yang digunakan mengacu pada Design Research Methodology (DRM), dimulai dari pengumpulan data komentar terkait promosi judi dari platform Kaggle. Data kemudian melalui tahap pre-processing, seperti case folding, pembersihan, penyaringan, dan validasi untuk memastikan kualitasnya. Model Llama 3.2 di fine-tune menggunakan pendekatan Parameter-Efficient Fine-Tuning (PEFT) dan Low-Rank Adaptation (LoRA). Evaluasi kinerja model dilakukan melalui Confusion Matrix dan Classification Report, mencakup metrik accuracy, precision, recall, dan F1-score. Hasil menunjukkan bahwa model tanpa optimizer menghasilkan akurasi 0,89. Dampak penerapan optimizer AdamW, akurasi meningkat menjadi 0,92 dengan peningkatan keseimbangan antara precision dan recall pada kedua kelas (judi dan non-judi). Peningkatan performa ini menunjukkan efektivitas penggunaan optimizer dalam meningkatkan proses pelatihan dan generalisasi model. Secara keseluruhan, studi ini menunjukkan bahwa Llama 3.2 mampu mendeteksi promosi judi online secara efektif dan dapat menjadi fondasi dalam pengembangan sistem deteksi konten ilegal. -------- Online gambling is a serious problem in Indonesia due to its social and economic impacts, exacerbated by easy access and massive promotional strategies. This study aims to implement the Large Language Model (LLM) Llama 3.2 based on the Hugging Face library to detect online gambling promotions and analyze its performance with and without optimization techniques. The method used refers to the Design Research Methodology (DRM), starting with collecting comment data related to gambling promotions from the Kaggle platform. The data then goes through preprocessing stages, such as case folding, cleaning, filtering, and validation to ensure its quality. The Llama 3.2 model was refined using the Parameter-Efficient Fine-Tuning (PEFT) and Low-Rank Adaptation (LoRA) approaches. Model performance evaluation was conducted through a Confusion Matrix and Classification Report, which includes metrics of accuracy, precision, recall, and F1-score. The results show that the model without an optimizer produces an accuracy of 0.89. The impact of implementing the AdamW optimizer increases the accuracy to 0.92 with an improved balance between precision and recall in both classes (gambling and non-gambling). This performance improvement demonstrates the effectiveness of using the optimizer in improving the training process and model generalization. Overall, this study demonstrates that Llama 3.2 is capable of effectively detecting online gambling promotions and can serve as a foundation for developing illegal content detection systems.

	Text S_RPL_2107757_Title.pdf Download (900kB)
	Text S_RPL_2107757_Chapter1.pdf Download (228kB)
	Text S_RPL_2107757_Chapter2.pdf Restricted to Staf Perpustakaan Download (593kB)
	Text S_RPL_2107757_Chapter3.pdf Download (466kB)
	Text S_RPL_2107757_Chapter4.pdf Restricted to Staf Perpustakaan Download (987kB)
	Text S_RPL_2107757_Chapter5.pdf Download (180kB)
	Text S_RPL_2107757_Appendix.pdf Restricted to Staf Perpustakaan Download (653kB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?user=dkpahNwAAAAJ&hl=en&oi=ao ID SINTA Dosen Pembimbing: Indira Syawanodya: 6681751 Yulia Retnowati: 6852573
Uncontrolled Keywords:	Judi Online, Llama 3.2, Hugging Face, Optimizer AdamW, Deteksi Teks, PEFT, LoRA, Online Gambling, Llama 3.2, Hugging Face, AdamW Optimizer, Text Detection, PEFT, LoRA
Subjects:	L Education > L Education (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General)
Divisions:	UPI Kampus cibiru > S1 Rekayasa Perangkaat Lunak
Depositing User:	Muhammad Rafid Miftah Fadhil
Date Deposited:	03 Sep 2025 07:07
Last Modified:	03 Sep 2025 07:07
URI:	http://repository.upi.edu/id/eprint/137000

Actions (login required)

View Item