Laporan Kerja Praktik 29 Agustus 2025 - 30 November 2025 di Laboratorium Grafika, Interaksi, Gim dan Analitik (GIGA) Institut Teknologi Sepuluh Nopember

Hermawan, Daniel and Negoro, Kresna Winata Perwiro (2026) Laporan Kerja Praktik 29 Agustus 2025 - 30 November 2025 di Laboratorium Grafika, Interaksi, Gim dan Analitik (GIGA) Institut Teknologi Sepuluh Nopember. Project Report. [s.n.], [s.l.]. (Unpublished)

Text 5025201087_5025231308-Project_Report.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Kesehatan mental merupakan aspek krusial dalam kesejahteraan masyarakat, di mana teknologi kecerdasan buatan (artificial intelligence) kini berperan penting dalam menyediakan dukungan psikologis. Aplikasi penelitian ini dibangun untuk mengklasifikasikan respons ke dalam tiga kategori empati, yaitu emotional reactions, interpretations, dan explorations. Proyek ini melakukan adaptasi ke Bahasa Indonesia serta mengembangkan arsitektur Bi-Encoder RoBERTa dengan mekanisme multi-task learning. Hasil pengujian adaptasi bahasa menunjukkan bahwa model Cahya/BERT-Indonesian mencapai performa terbaik dengan F1-Score 0,7307 pada kategori emotional reactions. Selanjutnya, pada tahap pengembangan model, arsitektur Bi-Encoder (unfrozen) terbukti sangat efektif, mencapai rationale macro F1 sebesar 0,9435 dan juga menunjukkan kemampuan generalisasi yang kuat dengan akurasi 0,912 pada validasi dataset eksternal.
==================================================================================================================================
Mental health is a crucial aspect of societal well-being, where artificial intelligence (AI) technology now plays a crucial role in providing psychological support. This research application was developed to classify responses into three empathy categories: emotional reactions, interpretations, and explorations. This project adapted the language to Indonesian and developed the RoBERTa Bi-Encoder architecture with a multi-task learning mechanism. The language adaptation test results showed that the Cahya/BERT-Indonesian model achieved the best performance with an F1-Score of 0.7307 in the emotional reactions category. Furthermore, during the model development stage, the Bi-Encoder architecture (unfrozen) proved highly effective, achieving a macro F1 rationale of 0.9435 and also demonstrating strong generalization ability with an accuracy of 0.912 on the external validation dataset.

Item Type:	Monograph (Project Report)
Uncontrolled Keywords:	Natural Language Processing (NLP), Deep Learning, Empati, RoBERTa, BERT, Klasifikasi Teks
Subjects:	B Philosophy. Psychology. Religion > BF Psychology > BF318 Learning, Psychology of (Deep learning) Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > T Technology (General) > T174 Technological forecasting T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Daniel Hermawan
Date Deposited:	21 Jan 2026 00:50
Last Modified:	21 Jan 2026 00:50
URI:	http://repository.its.ac.id/id/eprint/129880

Actions (login required)

View Item