Penerapan Fitur Artificial Intelligence Chatbot Berbasis Retrieval-Augmented Generation Pada Studi Kasus Tanya Jawab Standar Operasional Prosedur Di Industri Hulu Gas Dan Minyak Bumi

Fadhilah, Muhammad Rifqi (2025) Penerapan Fitur Artificial Intelligence Chatbot Berbasis Retrieval-Augmented Generation Pada Studi Kasus Tanya Jawab Standar Operasional Prosedur Di Industri Hulu Gas Dan Minyak Bumi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211228-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Standar operasional prosedur (SOP) merupakan aturan penting dalam hal pelaksanaan kerja dan keselamatan operasional di industri hulu gas dan minyak bumi. Namun, tantangan dalam aksesibilitas dokumen SOP masih menjadi hambatan di lapangan. Penelitian ini mengusulkan penerapan fitur Artificial Intelligence chatbot berbasis Retrieval-Augmented Generation (RAG) sebagai solusi untuk mendukung proses tanya jawab terkait SOP secara presisi dan memiliki nilai similarity yang tinggi. Sistem ini dirancang untuk berjalan secara lokal (on-premise), sehingga tetap menjaga privasi dan kerahasiaan data perusahaan. Dalam pengembangan sistem, digunakan model Large Language Models (LLM) Qwen2- 1.5B sebagai komponen generatif dan BAAI/bge-base-en-v1.5 sebagai model embedding. Embedding disimpan dalam Qdrant sebagai basis vektor untuk pencarian yang kontekstual. Chatbot ini diintegrasikan ke dalam portal web yang dikembangkan dengan Next.js. Skenario uji coba dilakukan dengan lima gaya prompting instruktif, kasual, Bahasa Indonesia, analytical, dan evaluator untuk menilai kemampuan model dalam merespons pertanyaan dalam berbagai konteks komunikasi dan dilakukan perbandingan dengan empat model Large Language Models (LLM) Qwen2-1.5B, Qwen2-1.5B-Instruct, DeepSeek-R1- Distill-Qwen-1.5B, dan TinyLlama-1.1B-Chat-v1.0 untuk melihat model mana yang memiliki nilai evaluasi paling tinggi. Hasil menunjukkan bahwa kombinasi model qwen/Qwen2-1.5B dengan gaya prompting instruktif menghasilkan tingkat presisi dan similarity tertinggi dengan skor BLEU 0.4305, ROUGE-1 0.6629, ROUGE-2 0.5720, ROUGE-L 0.6179, dan ROUGE- Lsum 0.6174. Temuan ini menunjukkan bahwa pendekatan Retreival-Augmented Generation RAG dengan model dan konfigurasi tersebut mampu meningkatkan presisi respons chatbot terhadap pertanyaan berbasis SOP.
=============================================================================================================================================
Standard operating procedures (SOPs) are important rules in terms of work implementation and operational safety in the upstream oil and gas industry. However, challenges in the accessibility of SOP documents are still an obstacle in the field. This study proposes the implementation of an Artificial Intelligence chatbot feature based on Retrieval-Augmented Generation (RAG) as a solution to support the question and answer process related to SOPs with precision and high similarity values. This system is designed to run locally (on-premise), thus maintaining the privacy and confidentiality of company data. In developing the system, the Large Language Models (LLM) Qwen2-1.5B model is used as a generative component and BAAI/bge-base-en-v1.5 as an embedding model. The embedding is stored in Qdrant as a base vector for contextual search. This chatbot is integrated into a web portal developed with Next.js. The test scenario was conducted with five prompting styles—instructive, casual, Indonesian, analytical, and evaluator—to assess the model's ability to respond to questions in various communication contexts, and a comparison was made with four large language models (LLMs)—Qwen2-1.5B, Qwen2-1.5B-Instruct, DeepSeek-R1-Distill-Qwen-1.5B, and TinyLlama-1.1B-Chat-v1.0—to see which model has the highest evaluation value. The results show that the combination of the Qwen/Qwen2-1.5B model with the instructive prompting style produces the highest level of precision and similarity with a BLEU score of 0.4305, ROUGE- 1 0.6629, ROUGE-2 0.5720, ROUGE-L 0.6179, and ROUGE-Lsum 0.6174. These findings indicate that the Retrieval-Augmented Generation (RAG) approach with these models and configurations is capable of improving the precision of chatbot responses to SOP-based questions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Chatbot, Retrieval-Augmented Generation (RAG), Question-Answered
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Rifqi Fadhilah
Date Deposited:	31 Jul 2025 07:53
Last Modified:	31 Jul 2025 07:53
URI:	http://repository.its.ac.id/id/eprint/125086

Actions (login required)

View Item