Pengembangan Chatbot Berbasis LLM Dengan Retrieval-Augmented Generation dan Rule Based Untuk Rekomendasi Kecocokan Obat Bebas Indonesia

Kurniawan, Gilang Raya (2026) Pengembangan Chatbot Berbasis LLM Dengan Retrieval-Augmented Generation dan Rule Based Untuk Rekomendasi Kecocokan Obat Bebas Indonesia. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027221045-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Tingginya angka self-medication di Indonesia, yang mencapai lebih dari 35% rumah tangga, seringkali tidak dibarengi dengan literasi kesehatan yang memadai. Hal ini menyebabkan risiko fatal, di mana sekitar 18-22% kasus kematian akibat keracunan zat kimia disebabkan oleh penggunaan obat yang tidak sesuai indikasi atau dosis. Penggunaan Large Language Model (LLM) standar untuk edukasi medis memiliki risiko "halusinasi" yang membahayakan. Oleh karena itu, penelitian ini mengusulkan pengembangan chatbot berbasis LLM untuk rekomendasi obat bebas di Indonesia dengan arsitektur hybrid. Sistem ini mengintegrasikan metode Retrieval-Augmented Generation (RAG) dan Rule-Based untuk meningkatkan akurasi serta keamanan informasi. Metode RAG menggunakan qwen-embeddings dan FAISS untuk mengambil data secara dinamis dari basis data obat yang telah terverifikasi. Sementara itu, pendekatan Rule-Based diterapkan sebagai filter keputusan klinis untuk menangani kondisi berisiko tinggi pada kelompok rentan. Penelitian ini membandingkan tiga model LLM, yaitu BioMistral, LLaMA, dan GPT-J, yang dioptimasi menggunakan teknik Parameter-Efficient Fine-Tuning berupa LoRA (Low-Rank Adaptation). Evaluasi dilakukan menggunakan rouge score dengan kinerja sistem diukur secara end-to-end menggunakan metrik ROUGE-1 Precision, ROUGE-1 Recall, ROUGE-1 F1, ROUGE-2 F1 dan ROUGE-L F1. Hasil penelitian ini diharapkan dapat memberikan kontribusi pada pengembangan asisten cerdas yang aman bagi masyarakat dalam mengelola kesehatan secara mandiri.
==================================================================================================================================
The high rate of self-medication in Indonesia affecting more than 35% of households is often not accompanied by adequate health literacy. This poses a fatal risk, as approximately 18–22% of deaths from chemical poisoning are caused by the use of medications that do not match the indicated use or dosage. The use of standard Large Language Models (LLMs) for medical education carries the risk of dangerous "hallucinations." Therefore, this study proposes the development of an LLM-based chatbot for over-the-counter medication recommendations in Indonesia using a hybrid architecture. This system integrates Retrieval-Augmented Generation (RAG) and Rule-Based methods to improve the accuracy and safety of the information provided. The RAG method uses qwen-embeddings and FAISS to dynamically retrieve data from a verified medication database. Meanwhile, the Rule-Based approach is applied as a clinical decision filter to address high-risk conditions in vulnerable groups. This study compares three LLM models—BioMistral, LLaMA, and GPT-J—which were optimized using the Parameter-Efficient Fine-Tuning technique known as LoRA (Low-Rank Adaptation). Evaluation was conducted using the ROUGE score, with system performance measured end-to-end using the ROUGE-1 Precision, ROUGE-1 Recall, ROUGE-1 F1, ROUGE-2 F1, and ROUGE-L F1 metrics. The results of this study are expected to contribute to the development of safe intelligent assistants that help people manage their health independently.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Chatbot, LLM, RAG, Rule-Based, Obat Bebas, ROUGE, LoRA, Over-The-Counter
Subjects:	T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Gilang Raya Kurniawan
Date Deposited:	29 Jul 2026 07:02
Last Modified:	30 Jul 2026 06:49
URI:	http://repository.its.ac.id/id/eprint/139041

Actions (login required)

View Item