WAHDANA, RIO ICHSAN (2026) PERANCANGAN DAN IMPLEMENTASI CHATBOT KESEHATAN MATA BERBASIS BAHASA INDONESIA MENGGUNAKAN MODEL INDOBERT. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
Cover.pdf Download (512kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
Bab 1.pdf Restricted to Registered users only Download (40kB) |
|
|
Text (BAB II)
Bab 2.pdf Restricted to Registered users only Download (252kB) |
|
|
Text (BAB III)
Bab 3.pdf Restricted to Registered users only Download (178kB) |
|
|
Text (BAB IV)
Bab 4.pdf Restricted to Registered users only Download (442kB) |
|
|
Text (BAB V)
Bab 5.pdf Restricted to Registered users only Download (78kB) |
|
|
Text (DAFTAR PUSTAKA)
Daftar Pustaka.pdf Restricted to Registered users only Download (92kB) |
|
|
Text (LAMPIRAN)
Lampiran.pdf Restricted to Registered users only Download (239kB) |
Abstract
Eye diseases require prompt and accurate access to medical education. This study aims to develop a Natural Language Processing (NLP)-based chatbot using the IndoBERT model to classify eye health queries into 7 main intent categories. The novelty of this research lies in the hybrid integration of the IndoBERT (indobert base-p1) intent classification model with a Cosine Similarity-based Q&A Retrieval module, along with text augmentation strictly applied to the training set to prevent data leakage. The pristine dataset of 736 samples was first split into an 80% train validation set and a 20% unseen test set (148 samples). Text augmentation expanded the training data to 2,355 samples (totaling 2,944 samples). Model training was executed for 3 epochs using the AdamW optimizer with a learning rate of 2 × 10−5 and a batch size of 16. Evaluation on the unseen test set demonstrated optimal performance with an accuracy of 89.86%, a macro F1-score of 90.48%, a weighted F1-score of 89.81%, and ROC-AUC values ranging from 0.99 to 1.00. End-to-end system testing recorded an average response latency of 1.12 seconds. This study proves that the hybrid IndoBERT approach is effective in delivering responsive and accurate eye health guidance. Keywords: Chatbot, Natural Language processing, IndoBERT, Eye Health, Intent Classification. Penyakit mata membutuhkan akses edukasi medis yang cepat dan akurat. Penelitian ini bertujuan mengembangkan chatbot berbasis Natural Language Processing (NLP) menggunakan model IndoBERT untuk mengklasifikasikan kueri pertanyaan kesehatan mata ke dalam 7 kategori intent utama (penyakit_mata, pencegahan, keluhan_mata, penanganan_mandiri, bahaya, obat_mata, dan informasi_umum). Kebaruan (novelty) dari penelitian ini terletak pada integrasi arsitektur hibrid antara model klasifikasi intent IndoBERT (indobert-base-p1) dan modul Q&A Retrieval berbasis Cosine Similarity, serta penerapan text augmentation khusus pada data latih guna mencegah kebocoran data (data leakage). Dataset murni berjumlah 736 sampel dipisahkan terlebih dahulu menjadi 80% data latih-validasi dan 20% data uji murni (unseen test set, 148 data). Pengayaan data mengekspansi data latih menjadi 2.355 sampel (total dataset menjadi 2.944 data). Pelatihan model dieksekusi selama 3 epoch menggunakan optimizer AdamW dengan laju pembelajaran (learning rate) 2 × 10−5 dan batch size 16. Hasil evaluasi pada data uji murni menunjukkan model IndoBERT mencapai accuracy sebesar 89,86%, macro F1-score 90,48%, weighted F1-score 89,81%, dan nilai ROC-AUC pada rentang 0,99–1,00. Pengujian end-to-end sistem mencatatkan rata-rata waktu tanggap (response latency) sebesar 1,12 detik (0,8–1,4 detik). Penelitian ini membuktikan bahwa pendekatan hibrid IndoBERT efektif menyediakan sarana edukasi kesehatan mata yang responsif, akurat, dan stabil. Kata Kunci: Chatbot, Natural Language processing, IndoBERT, Kesehatan Mata, Intent Classification.
Actions (login required)
 |
View Item |
