Dashboard Pelaporan Insiden Berbasis Text-to-Query Menggunakan Metode Large Language Model

Furqon, Alif (2026) Dashboard Pelaporan Insiden Berbasis Text-to-Query Menggunakan Metode Large Language Model. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241100-Master_Thesis.pdf Restricted to Repository staff only Download (18MB) | Request a copy

Abstract

Proses pelaporan insiden site down pada Divisi Operation Monitoring Center (OMC) Regional Jawa Timur masih banyak dilakukan secara manual melalui pencarian data, penyaringan informasi, dan penyusunan laporan menggunakan spreadsheet. Proses tersebut membutuhkan waktu, bergan• tung pada ketelitian operator, serta berpotensi menimbulkan keterlambatan dan ketidakkonsistenan informasi. Penelitian ini bertujuan mengembangkan dashboard pelaporan insiden berbasis Large Language Model (LLM) dengan pendekatan text-to-query untuk menerjemahkan pertanyaan bahasa alami dalam Bahasa Indonesia menjadi query SQL.
Sistem dikembangkan menggunakan Laravel, MySQL, Bootstrap, serta integrasi API GPT-4 dan Gemini. Alur sistem dimulai dari input pertanyaan bahasa alami, pembentukan prompt, penerjemahan menjadi query SQL, validasi melalui SQL safety layer, eksekusi pada basis data, dan penyajian hasil pada dashboard. Pengujian dilakukan menggunakan 40 pertanyaan yang terdiri dari 20 pertanyaan formal dan 20 pertanyaan non-formal. Evaluasi difokuskan pada validitas query SQL, kemampuan eksekusi, penggunaan token, waktu respons, dan konsistensi output.
Hasil pengujian menunjukkan bahwa GPT-4 memiliki performa lebih efisien dibandingkan Gemini. GPT-4 memperoleh rata-rata waktu respons sebesar
2,28 detik dengan rata-rata penggunaan 912,98 token, sedangkan Gemini memperoleh rata-rata waktu respons sebesar 9,32 detik dengan rata-rata penggunaan 1.375,40 token. Selain itu, GPT-4 menunjukkan struktur query SQL yang lebih konsisten, sedangkan Gemini masih menunjukkan variasi pada beberapa pertanyaan non-formal dan pertanyaan berbasis wilayah. Dengan demikian, sistem berbasis LLM dapat membantu pengguna non-teknis dalam mengakses data insiden tanpa harus menulis SQL secara manual, namun validasi query tetap diperlukan sebelum eksekusi pada basis data.
==============================================================================================================================
The reporting process of site down incidents in the Operation Monitoring Center (OMC) Division of East Java Region is still largely conducted manually through data retrieval, information filtering, and report preparation using spreadsheets. This process requires time, depends on operator accuracy, and may lead to reporting delays and information inconsistency. This research aims to develop an incident reporting dashboard based on a Large Language Model (LLM) using a text-to-query approach to translate natural language questions in Indonesian into SQL queries.
The system was developed using Laravel, MySQL, Bootstrap, and the integration of GPT-4 and Gemini APis. The system workflow begins with natural language input, prompt construction, translation into SQL queries, validation through an SQL safety layer, execution on the database, and presentation of the results on the dashboard. Testing was conducted using 40 questions consisting of 20 formal questions and 20 non-formal questions. The evaluation focused on SQL query validity, execution capability, token usage, response time, and output consistency.
The testing results show that GPT-4 achieved more efficient performance than Gemini. GPT-4 obtained an average response time of 2.28 seconds with an average token usage of 912.98 tokens, while Gemini obtained an average response time of 9.32 seconds with an average token usage of 1,375.40 tokens. In addition, GPT-4 produced more consistent SQL query structures, while Gemini still showed variations in several non-formal and region-based questions. Therefore, the LLM-based system can assist non-technical users in accessing incident data without manually writing SQL queries. However, query validation remains necessary before database execution.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Kata kunci: Large Language Model, text-to-query, dashboard insiden, SQL, GPT-4, Gemini Keywords: Large Language Model, text-to-query, incident dashboard, SQL, GPT-4, Gemini
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.6 Management information systems
Divisions:	Faculty of Electrical Technology > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Alif Furqon
Date Deposited:	30 Jul 2026 04:01
Last Modified:	30 Jul 2026 04:01
URI:	http://repository.its.ac.id/id/eprint/139578

Actions (login required)

View Item