Sistem Deteksi Kesalahan pada Pengajuan Sertifikasi Halal Berbasis Knowledge Graphs dan Large Language Models

Bhamakerti, Ganendra Aby (2026) Sistem Deteksi Kesalahan pada Pengajuan Sertifikasi Halal Berbasis Knowledge Graphs dan Large Language Models. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6026232025-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Dalam praktik untuk memperoleh sertifikasi halal, tidak menutup kemungkinan terjadinya kesalahan seperti overclaim di mana pelaku usaha hanya mendaftarkan sebagian produk yang dijual untuk memperoleh sertifikasi halal alih-alih mendaftarkan semua produk. Oleh karena itu, diperlukan suatu solusi yaitu sistem deteksi kesalahan pada proses pengajuan sertifikasi halal. Sistem ini berbasis Knowledge Graph (KG) dan Large Language Model (LLM), serta menggunakan sumber data yang diperoleh dari HalalWave dan GoFood. Metodologi yang dijalankan dalam mengerjakan penelitian tesis ini yaitu dengan melakukan pengumpulan terhadap data dari HalalWave dan GoFood. Data yang didapatkan dari HalalWave merupakan kumpulan informasi produk serta penyedia makanan yang telah mendapatkan sertifikasi halal dan data yang diperoleh dari GoFood merupakan kumpulan produk serta merchant di GoFood. Tahap berikutnya yang dijalankan yaitu melakukan data preprocessing pada dataset GoFood. Kemudian dilakukan konstruksi KG dengan dataset GoFood hasil dari data preprocessing. Setelah itu dilakukan data preprocessing pada dataset HalalWave. Berikutnya dilakukan Retrieval-Augmented Generation (RAG) menggunakan API Google Maps, LLM, KG, dan dataset HalalWave hasil dari data preprocessing, tahap ini membandingkan sistem yang menggunakan KG dan LLM dengan sistem yang hanya menggunakan KG. Terakhir dilakukan evaluasi menggunakan metrik berbasis confusion matrix dan kurva ROC serta AUC. Hasil yang diperoleh dalam penelitian tesis ini antara lain sistem dengan KG dan LLM menghasilkan rata-rata nilai accuracy 0.690, precision 0.652, recall 0.987, F1-score 0.786, serta AUC 0.636, dan sistem dengan KG saja menghasilkan rata-rata nilai accuracy 0.591, precision 0.586, recall 0.994, F1-score 0.737, serta AUC 0.519. Hasil tersebut memberikan indikasi bahwa sistem yang menggunakan pendekatan KG dan LLM secara keseluruhan lebih unggul dibandingkan dengan sistem yang hanya menggunakan pendekatan KG.
======================================================================================================================================
In practice, obtaining halal certification does not rule out the possibility of errors such as overclaiming, where business actors only register a portion of the products sold to obtain halal certification instead of registering all products. Therefore, a solution is needed, namely an error detection system in the halal certification submission process. This system is based on Knowledge Graphs (KG) and Large Language Models (LLM), and uses data sources obtained from HalalWave and GoFood. The methodology used in conducting this thesis research is to collect data from HalalWave and GoFood. The data obtained from HalalWave is a collection of product information and food providers that have obtained halal certification, and the data obtained from GoFood is a collection of products and merchants on GoFood. The next stage is to perform data preprocessing on the GoFood dataset. Then, KG is constructed with the GoFood dataset resulting from the preprocessed data. After that, data preprocessing is carried out on the HalalWave dataset. Next, Retrieval-Augmented Generation (RAG) is carried out using the Google Maps API, LLM, KG, and the HalalWave dataset resulting from data preprocessing, this stage compares the system using KG and LLM with the system that only uses KG. Finally, an evaluation was conducted using confusion matrix-based metrics and ROC curves and AUC. The results obtained in this thesis research are the system with KG and LLM yielded an average accuracy value of 0.690, precision 0.652, recall 0.987, F1-score 0.786, and AUC 0.636, and the system with KG alone yielded an average accuracy value of 0.591, precision 0.586, recall 0.994, F1-score 0.737, and AUC 0.519. These results provide an indication that the system using the KG and LLM approach is overall superior compared to the system using only the KG approach.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Knowledge Graph, Large Language Model, Sertifikasi Halal, Sistem Deteksi Kesalahan, Retrieval-Augmented Generation. = Knowledge Graph, Large Language Model, Halal Certification, Error Detection System, Retrieval-Augmented Generation.
Subjects:	Q Science > QA Mathematics > QA166 Graph theory Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.6 Computer programming. T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Ganendra Aby Bhamakerti
Date Deposited:	28 Jan 2026 07:43
Last Modified:	28 Jan 2026 07:43
URI:	http://repository.its.ac.id/id/eprint/130808

Actions (login required)

View Item