Model Bidirectional Encoder Representations from Transformers untuk klasifikasi laporan Barang Milik Negara

Firdaus, Akhdan Musyaffa (2023) Model Bidirectional Encoder Representations from Transformers untuk klasifikasi laporan Barang Milik Negara. Sarjana thesis, UIN Sunan Gunung Djati Bandung.

Preview	Text (Cover) 1_cover.pdf Download (102kB) | Preview
Preview	Text (ABSTRAK) 2_abstrak.pdf Download (57kB) | Preview
Preview	Text (DAFTAR ISI) 3_daftarisi.pdf Download (174kB) | Preview
Preview	Text (BAB I) 4_bab1.pdf Download (132kB) | Preview
	Text (BAB II) 5_bab2.pdf Restricted to Registered users only Download (411kB) | Request a copy
	Text (BAB III) 6_bab3.pdf Restricted to Registered users only Download (614kB) | Request a copy
	Text (BAB IV) 7_bab4.pdf Restricted to Registered users only Download (1MB) | Request a copy
	Text (BAB V) 8_bab5.pdf Restricted to Registered users only Download (73kB) | Request a copy
	Text (DAFTAR PUSTAKA) 9_daftarpustaka.pdf Restricted to Registered users only Download (132kB) | Request a copy

Abstract

INDONESIA : Penelitian ini bertujuan untuk membangun asisten pelaporan dan monitoring kondisi Barang Milik Negara (BMN) pada Direktorat Jenderal Pendidikan Islam Kementerian Agama Republik Indonesia (Ditjen Pendis Kemenag RI), di mana terdapat indikasi terjadinya overcost maintenance akibat keterlambatan dalam mendeteksi permasalahan kendaraan. Permasalahan kendaraan dapat di telusuri berdasarkan laporan kondisi kendaraan yang dilakukan oleh pengemudi kepada petugas BMN. Laporan kendaraan umumnya berbentuk teks bebas dan tidak berstruktur sehingga sering menyebabkan ambiguitas dan kesalahan identifikasi masalah kendaraan. Dalam upaya mengatasi permasalahan tersebut, dibutuhkan sebuah teknologi yang dapat mempermudah pendeteksian kondisi kendaraan melalui klasifikasi kondisi tanpa mengubah alur proses bisnis konvensional. Teknologi chatbot yang diintegrasikan dengan model Bidirectional Encoder Represenation from Transformers (BERT) untuk mempermudah proses pelaporan dan monitoring kondisi BMN. Model BERT digunakan karena kemampuannya dalam memahami konteks suatu kalimat meski memiliki struktur yang bervariasi. Dilakukan proses fine-tuning dalam pembangunan model klasifikasi teks untuk menghasilkan output berbentuk multi-label classification dan pengingat melalui chatbot. Model dibatasi dengan bahasa Indonesia dan dikembangkan menggunakan bahasa pemrograman Javascript dan Python dalam pembangunan dan pengintegrasiannya. Metodologi penelitian mengadopsi pendekatan Standar Kompetensi Kerja Nasional Indonesia (SKKNI) bidang Data Science dari Peraturan Menteri Ketenagakerjaan No. 299 tahun 2020 yang melibatkan langkah-langkah mulai dari pemahaman bisnis hingga deployment. Dengan capaian akurasi sebesar 98.83% menggunakan alat ukur confusion matrix, diketahui bahwa model klasifikasi teks bekerja dengan baik dan memiliki potensi besar untuk digunakan dalam berbagai kasus dan situasi yang serupa. ENGLISH : This study aims to build an assistant for reporting and monitoring the condition of State Property (BMN) at the Directorate General of Islamic Education of the Ministry of Religious Affairs of the Republic of Indonesia (Ditjen Pendis Kemenag RI), where there are indications of overcost maintenance due to delays in detecting problems vehicle. Vehicle problems can be traced based on vehicle condition reports made by the driver to BMN officers. Vehicle reports are generally free-text and unstructured, often leading to ambiguity and misidentification of vehicle problems. In an effort to overcome these problems, a technology is needed that can facilitate the detection of vehicle conditions through condition classification without changing the flow of conventional business processes. Chatbot technology is integrated with the Bidirectional Encoder Represenation from Transformers (BERT) model to facilitate the process of reporting and monitoring BMN conditions. The BERT model is used because of its ability to understand the context of a sentence even though it has a varied structure. A fine-tuning process is carried out in the construction of a text classification model to produce output in the form of multi-label classification and reminders via chatbot. The model is limited to Bahasa Indonesia and developed using Javascript and Python programming languages in its development and integration. The research methodology adopts the approach of the Indonesian National Work Competency Standard (SKKNI) in the field of Data Science from the Minister of Manpower Regulation No. 299 of 2020 which involves steps from business understanding to deployment. With an accuracy achievement of 98.83% using confusion matrix measuring instruments, it is known that the text classification model works well and has great potential to be used in various cases and similar situations.

Actions (login required)

View Item