Implementasi Automated Essay Scoring pada Penilaian Esai Bahasa Indonesia dengan Pendekatan Transfer Learning

Aliyah, Nathania Elirica (2025) Implementasi Automated Essay Scoring pada Penilaian Esai Bahasa Indonesia dengan Pendekatan Transfer Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027211057-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (2MB) | Request a copy

Abstract

Esai adalah salah satu metode penilaian yang penting untuk mengukur tingkat kemampuan siswa dalam berpikir kritis, merumuskan argumen yang kukuh, dan mengekspresikan konsep dengan benar. Sampai saat ini, proses penilaian esai siswa di Indonesia masih dilakukan secara manual sehingga menghasilkan nilai yang kurang konsisten dan objektif. Selain itu, proses manual ini memakan waktu yang cukup lama bagi tenaga pendidik. Maka dari itu, sistem automated essay scoring (AES) hadir sebagai solusi. Sistem ini merupakan ranah penelitian yang selalu berkembang dari tahun ke tahun. Sistem AES telah diimplementasikan menggunakan berbagai macam pendekatan, seperti cosine similarity, machine learning tradisional, dan lainnya. Namun, penelitian terkait sistem AES yang khusus untuk esai berbahasa Indonesia masih terbatas dan belum sepenuhnya mencakup pemahaman model pada tingkat kalimat. Oleh karena itu, pada penelitian ini sistem AES akan diimplementasikan untuk menilai esai berbahasa Indonesia menggunakan large language model (LLM) IndoBERT dan sentence embedding IndoSBERT dengan pendekatan transfer learning. Transfer learning dilakukan dengan melakukan fine-tuning model IndoBERT serta menambahkan embedding dari IndoSBERT ke dalam arsitektur IndoBERT menggunakan metode concatenation. Model dilatih pada dataset Automated Essay Scoring 2.0 dari Kaggle yang berfokus pada penilaian tata bahasa dari esai siswa dengan skala skor 1 hingga 6 dan rubrik terperinci. Evaluasi model dilakukan dengan metrik quadratic weighted kappa (QWK) yang mengukur kesesuaian antara penilai manusia dan penilai model. Penelitian menghasilkan nilai QWK pada kombinasi IndoBERT dan IndoSBERT memiliki peningkatan terhadap IndoSBERT saja dan IndoBERT saja yaitu, 0,497 dan 0,04. Selanjutnya, model diintegrasikan dengan aplikasi website menggunakan FastAPI yang memungkinkan pengguna mengunggah file esai digital untuk mendaptkan nilai esai.
==================================================================================================================================
Essay is one of the important assessment methods to measure the level of students' ability to think critically, formulate solid arguments, and express concepts correctly. Until now, the process of assessing student essays in Indonesia is still done manually, resulting in inconsistent and objective scores. In addition, this manual process is time-consuming for educators. Therefore, automated essay scoring (AES) system comes as a solution. This system is an area of research that is always evolving from year to year. AES systems have been implemented using various approaches, such as cosine similarity, traditional machine learning, and others. However, research related to AES systems specifically for Indonesian essays is still limited and has not fully covered the understanding of models at the sentence level. Therefore, in this study, the AES system will be implemented to assess Indonesian essays using the IndoBERT large language model (LLM) and IndoSBERT sentence embedding with a transfer learning approach. Transfer learning is done by fine-tuning the IndoBERT model and adding the embedding of IndoSBERT into the IndoBERT architecture using the concatenation method. The model was trained on Kaggle's Automated Essay Scoring 2.0 dataset which focuses on grammar scoring of student essays with a score scale of 1 to 6 and a detailed rubric. Model evaluation was performed with the quadratic weighted kappa (QWK) metric which measures the agreement between human raters and model raters. The research results show that the QWK value in the combination of IndoBERT and IndoSBERT has increased compared to IndoSBERT alone and IndoBERT alone by 0.497 and 0.04 respectively. Furthermore, the model was integrated with a website application using FastAPI that allows users to upload digital essay files to get essay grades.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Automated essay scoring, Transfer learning, IndoBERT, IndoSBERT, QWK, Automated essay scoring, Transfer learning, IndoBERT, IndoSBERT, QWK
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Nathania Elirica Aliyah
Date Deposited:	22 Jan 2025 07:37
Last Modified:	22 Jan 2025 07:37
URI:	http://repository.its.ac.id/id/eprint/116598

Actions (login required)

View Item