Juanita, Safitri (2025) Klasifikasi Multi-Label Teks Jawaban Dokter Pada Konsultasi Kesehatan Daring Berdasarkan Fungsi Komunikasi Klinis Menggunakan Pendekatan Deep. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 7022201018-Doctoral.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
7022201018-Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy |
Abstract
Klasifikasi multi-label terhadap teks jawaban dokter dalam konsultasi kesehatan daring diperlukan untuk memahami variasi fungsi komunikasi klinis yang digunakan dalam interaksi medis berbasis teks. Penelitian ini mengembangkan alur analisis yang mencakup pemodelan topik menggunakan Latent Dirichlet Allocation (LDA) serta klasifikasi multi-label berdasarkan enam fungsi komunikasi klinis menurut King dan Hoppe. Dua pendekatan dievaluasi, yaitu Hybrid Machine Learning dan fine-tuned deep learning, dengan memanfaatkan korpus dwibahasa (Indonesia–Inggris) yang bersifat tidak seimbang. Evaluasi model dilakukan melalui tiga perspektif, yaitu berbasis contoh, berbasis label, dan berbasis pemeringkatan. Pada pendekatan Hybrid Machine Learning, model T–BR–AD menunjukkan kinerja terbaik pada korpus Indonesia untuk perspektif berbasis contoh dan berbasis label (Hamming Loss = 0,035; F1 = 0,973; F1-makro = 0,711). Pada korpus Inggris, model T–BR–RF mencapai performa tertinggi pada perspektif berbasis contoh (F1 = 0,987; Hamming Loss = 0,022) dan pemeringkatan (LRAP = 0,971). Selain itu, model T–BR–MLP unggul pada beberapa metrik spesifik di kedua korpus. Pendekatan fine-tuned deep learning menunjukkan performa yang lebih stabil. Pada korpus Indonesia, konfigurasi LV2 (20 epoch) menjadi yang terkuat pada perspektif berbasis contoh dan berbasis label (Hamming Loss = 0,035; F1 = 0,973; F1-mikro = 0,973). Pada korpus Inggris, konfigurasi BV1 (5 epoch) mencapai kinerja tertinggi pada seluruh perspektif evaluasi (Hamming Loss = 0,020; F1 = 0,978; F1-mikro = 0,980; LRAP = 0,971). Penelitian ini berkontribusi dengan menyusun kerangka klasifikasi multi-label untuk teks jawaban dokter dan menegaskan bahwa pemilihan pendekatan perlu mempertimbangkan karakteristik data serta perspektif evaluasi. Pengembangan lanjutan dapat diarahkan pada pemanfaatan model transformer dan integrasi representasi semantik berbasis ontologi.
====================================================================================================================================
Multi-label classification of doctors’ responses in online health consultations is necessary to understand the variety of clinical communication functions used in text-based medical interactions. This study develops an analytical pipeline comprising topic modeling using Latent Dirichlet Allocation (LDA) and multi-label classification based on the six clinical communication functions proposed by King and Hoppe. Two approaches were evaluated, namely Hybrid Machine Learning and fine-tuned deep learning, utilising an unbalanced bilingual corpus (Indonesian–English). Model performance is assessed across three evaluation perspectives: example-based, label-based, and ranking-based. Within the Hybrid Machine Learning approach, the T–BR–AD model achieves the best performance for the Indonesian corpus under the example-based and label-based perspectives (Hamming Loss = 0.035; F1 = 0.973; macro-F1 = 0.711). For the English corpus, the T–BR–RF model attains the highest performance under the example-based perspective (F1 = 0.987; Hamming Loss = 0.022) and the ranking-based perspective (LRAP = 0.971). Additionally, the T–BR–MLP model outperforms others on several specific metrics across both corpora. The fine-tuned deep learning approach demonstrates more stable performance. For the Indonesian corpus, the LV2 configuration (20 epochs) performs best under both the example-based and label-based perspectives (Hamming Loss = 0.035; F1 = 0.973; micro-F1 = 0.973). For the English corpus, the BV1 configuration (5 epochs) yields the highest performance across all evaluation perspectives (Hamming Loss = 0.020; F1 = 0.978; micro-F1 = 0.980; LRAP = 0.971). This study contributes a multilabel classification framework for doctors’ text-based responses and highlights that model selection should consider corpus characteristics and evaluation perspectives. Future work may explore transformer-based language models and the integration of ontology-driven semantic representations.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | klasifikasi multi-label, konsultasi kesehatan daring, komunikasi klinis, Hybrid Machine Learning, deep learning, LDA, korpus dwibahasa, multi-label classification, online health consultation, clinical communication, Hybrid Machine Learning, deep learning, LDA, bilingual corpus. |
| Subjects: | T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.62 Decision support systems |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis |
| Depositing User: | Safitri Juanita |
| Date Deposited: | 17 Dec 2025 00:42 |
| Last Modified: | 17 Dec 2025 00:42 |
| URI: | http://repository.its.ac.id/id/eprint/128999 |
Actions (login required)
 |
View Item |