Wijaya, Ardhika Krishna (2026) Pengembangan Aplikasi Mobile Automatic Question Generation Pada Pembelajaran Bahasa Inggris Tingkat Sekolah Menengah Atas. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5025221006-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5025221006-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy |
Abstract
Penguasaan bahasa Inggris merupakan kompetensi krusial dalam Kurikulum Merdeka tingkat SMA, namun penyediaan soal latihan yang variatif masih terkendala keterbatasan waktu dan sumber daya pendidik. Penelitian sebelumnya telah mengembangkan sistem Automatic Question Generation (AQG) berbasis web untuk siswa sekolah dasar, namun belum tersedia solusi berbasis mobile dengan target SMA. Tugas akhir ini mengembangkan aplikasi mobile MerdekAI yang menghasilkan soal Cloze Test dan MCQ Test secara otomatis dari teks bacaan bahasa Inggris Fase E (kelas X) dan F (kelas XI dan XII) Kurikulum Merdeka yang memanfaatkan model Masked Language Model (MLM) RoBERTa dan Causal Language Model (CLM) Mistral-7B Instruct v0.3.
Dataset dikumpulkan dari 942 entri tiga buku teks SMA Kurikulum Merdeka. Cloze Test dihasilkan melalui pipeline RoBERTa dan spaCy, sementara MCQ Test dihasilkan menggunakan Mistral-7B Instruct v0.3 yang di-fine-tuning menggunakan LoRA melalui empat studi Hyperparameter Optimization (HPO) berbasis Optuna. Aplikasi dikembangkan menggunakan Flutter dengan arsitektur client-server yang mengintegrasikan FastAPI sebagai gateway inferensi dan Firebase sebagai layanan autentikasi dan basis data.
Evaluasi metrik otomatis menghasilkan BLEU-1 sebesar 0,3893, BLEU-2 sebesar 0,3082, BLEU-3 sebesar 0,2618, BLEU-4 sebesar 0,2287, METEOR 0,4128, dan ROUGE-L 0,3980. Evaluasi oleh validator ahli menghasilkan rata-rata 4,85 untuk Cloze Test dan 4,92 untuk MCQ Test dari skala 5. Seluruh 16 test case Black-Box Testing dan 5 test case nonfungsional berstatus Pass. Pengujian pengguna terhadap 11 responden menghasilkan rata-rata seluruh kategori di atas 4,40, dengan kualitas konten soal memperoleh nilai tertinggi sebesar 4,71.
=======================================================================================================================================
English proficiency is a crucial competency in the Senior High School (SMA) Merdeka Curriculum. However, the provision of varied practice questions remains constrained by educators limited time and resources. Prior research has developed web-based Automatic Question Generation (AQG) systems targeting primary school students, yet no mobile based solution targeting senior high school has been available. This study develops MerdekAI, a mobile application that automatically generates Cloze Test and MCQ Test questions from English reading texts aligned with Phase E (grade X) and Phase F (grade XI and XII) of the Merdeka Curriculum, leveraging the Masked Language Model (MLM) RoBERTa and the Causal Language Model (CLM) Mistral-7B Instruct v0.3.
The dataset was collected from 942 entries across three SMA Merdeka Curriculum textbooks. Cloze Test generation applies a RoBERTa and spaCy pipeline, while the MCQ Test employs Mistral-7B Instruct v0.3 fine-tuned with LoRA through four Hyperparameter Optimization (HPO) studies using Optuna. The application was developed using Flutter with a client-server architecture integrating FastAPI as the inference gateway and Firebase as the authentication and database service.
Automatic metric evaluation yielded a BLEU-1 score of 0.3893, BLEU-2 score of 0.3082, BLEU-3 score of 0.2618, BLEU-4 0.2287, METEOR of 0.4128, and ROUGE-L of 0.3980. Expert validation produced average scores of 4.85 for Cloze Test and 4.92 for MCQ Test on a 5-point scale. All 16 Black Box Testing and 5 nonfungsional test cases passed. User testing involving 11 respondents yielded average scores above 4.40 across all categories, with question content quality achieving the highest score of 4.71.
Actions (login required)
 |
View Item |