Rahma, Khansa Adia (2026) Sistem Automated Essay Scoring Adaptif Berbasis IndoBERT Dan Meta-Learning: Pemanfaatan Konten Esai Ideal Guru Untuk Optimalisasi Penilaian Topik Baru. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5027221071-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5027221071-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy |
Abstract
Sistem Penilaian Esai Otomatis digunakan untuk menilai esai dengan cepat dan efisien. Namun, masalah yang sering muncul yaitu sistem ini kesulitan dalam memberi skor esai dari topik yang belum pernah dilatih sebelumnya (cross-prompt generalization), terutama untuk bahasa Indonesia. Penelitian ini mengusulkan pengembangan sistem AES bahasa Indonesia yang dirancang khusus untuk mengatasi masalah generalisasi lintas topik tersebut dengan menggabungkan model IndoBERT dengan kerangka Meta-Learning. IndoBERT digunakan sebagai feature extractor untuk menangkap representasi semantik mendalam dari esai, sementara Meta-Learning melatih model agar mampu beradaptasi dengan cepat pada topik baru. Penelitian ini memanfaatkan esai ideal guru sebagai konteks dalam proses Meta-Learning untuk membimbing model mempelajari karakteristik penting dari suatu topik. Evaluasi dilakukan menggunakan strategi Leave-One-Out Cross Validation pada 21 prompt dengan total 6.275 sampel esai. Hasil evaluasi menunjukkan model mencapai Quadratic Weighted Kappa (QWK) tertinggi sebesar 0,7879 dengan Root Mean Squared Error (RMSE) 0,8621 pada prompt spesifik, Analisis korelasi mengidentifikasi dua faktor utama yang memengaruhi performa, yaitu jarak distribusi MMD memiliki korelasi negatif kuat (r = -0,70) dan panjang kunci jawaban memiliki korelasi positif sedang (r = +0,56) dengan QWK. Model Meta-Learning unggul signifikan pada topik tertentu dengan peningkatan hingga 0,3947 poin dibandingkan model baseline, menunjukkan kemampuan adaptasi yang baik pada topik baru yang sesuai karakteristiknya.
================================================================================================================================
Automated Essay Scoring systems are used to evaluate essays quickly and efficiently. However, a common problem is that these systems struggle to score essays on topics that have not been previously trained (cross-prompt generalization), especially for Indonesian language. This research proposes the development of an Indonesian language AES system specifically designed to address this cross-topic generalization problem by combining the IndoBERT model with the Meta-Learning framework. IndoBERT is used as a feature extractor to capture deep semantic representations of essays, while Meta-Learning trains the model to adapt quickly to new topics. This research utilizes teacher's ideal essays as context in the Meta-Learning process to guide the model in learning important characteristics of a topic. Evaluation was conducted using the Leave-One-Out Cross Validation strategy on 21 prompts with a total of 6,275 essay samples. The evaluation results show that the model achieves the highest Quadratic Weighted Kappa (QWK) of 0.7879 with Root Mean Squared Error (RMSE) of 0.8621 on specific prompts. Correlation analysis identified two main factors affecting performance: MMD distribution distance has a strong negative correlation (r = -0.70) and answer key length has a moderate positive correlation (r = +0.56) with QWK. The Meta-Learning model demonstrates significant superiority on certain topics with improvements up to 0.3947 points compared to the baseline model, showing good adaptation capability on new topics with appropriate characteristics.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Automated Essay Scoring, IndoBERT, Meta-Learning, Cross-prompt Generalization, Esai Ideal Guru, bahasa Indonesia |
| Subjects: | Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis |
| Depositing User: | Khansa Adia Rahma |
| Date Deposited: | 22 Jan 2026 03:21 |
| Last Modified: | 22 Jan 2026 03:21 |
| URI: | http://repository.its.ac.id/id/eprint/130056 |
Actions (login required)
 |
View Item |