Implementasi Bidirectional Long Short-Term Memory Untuk Sistem Rekomendasi Jawaban Medis Otomatis Berbasis Web.

Lammora, Rosa Valentine (2022) Implementasi Bidirectional Long Short-Term Memory Untuk Sistem Rekomendasi Jawaban Medis Otomatis Berbasis Web. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 05111840000035-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (2MB)

Abstract

Penelitian di bidang tanya-jawab sebagai pekerjaan natural language processing (NLP) menghasilkan banyak produk yang membantu proses pencarian informasi lebih terarah. Salah satu implementasinya adalah dalam bentuk sistem rekomendasi jawaban. Pada bidang medis, belum terdapat sistem yang mengakomodasi kebutuhan ini dalam Bahasa Indonesia. Tugas akhir ini mengajukan sistem rekomendasi yang memanfaatkan data tanya-jawab dari situs telemedika Alodokter. Sistem terdiri dari pengklasifikasi topik multi-label dan pemilih rekomendasi jawaban. Klasifikasi yang diimplementasikan mengutamakan pendekatan deep learning dengan pertimbangan evaluasi subset accuracy, hamming loss, dan F1 score. Evaluasi untuk model ensemble CNN dan Bidirectional RNN, yaitu BiGRU dan BiLSTM, memperoleh hasil terbaik dengan subset accuracy keduanya 0.79, hamming loss keduanya 0.029, serta F1 score 0.87 dan 0.85. Tahap pemilihan rekomendasi jawaban melibatkan pencarian pertanyaan yang mirip dari data tanya-jawab Alodokter, kemudian mengembalikan teks jawaban dari pertanyaan tersebut yang sudah diringkas. Sistem disajikan dalam antarmuka web dan dilakukan uji coba untuk menilai relevansi hasil rekomendasi jawaban. Topik yang dapat ditanyakan dibatasi agar pengembangan sistem sesuai dengan sumber daya perangkat yang tersedia. Hasil uji coba menandakan sistem cukup mumpuni untuk menangani pertanyaan-pertanyaan, baik yang sudah ada di dataset maupun pertanyaan baru.
=================================================================================================================================
Research on question answering as a natural language processing (NLP) task has produced many outputs to help the process of information gathering be more directed. An examle of this is an answer recommendation system. In the medical field, there has not been developed a system to accommodate this need using the Indonesian language. This final project proposes an answer recommendation system that utilizes question-answer data from the telemedicine site Alodokter. The proposed system consists of a multi-label topic classifier and an answer recommender. The classifier is implemented primarily using a deep learning approach with evaluation measurements using subset accuracy, hamming loss, and F1 score. Evaluation results denote that the ensemble CNN-Bidirectional RNN models, which are BiGRU and BiLSTM, outperform baseline models with both subset accuracies scoring 0.79, both hamming losses scoring 0.029, and 0.87 F1 score for BiGRU and 0.85 F1 score for BiLSTM. The answer recommendation step involves looking for similar questions from the Alodokter dataset, then returning a summarized answer text as answer recommendation to the user. To enable easier interaction with users, this project also covers the implementation of a web-based interface. Testing is done to evaluate the relevancy of the answer recommendations produced by the system. There is a limitation of topics which can be handled by the system to ensure development conforms to available resources. Results of the final testing shows that the proposed system is able to accommodate both existing questions from the dataset and new ones.

Item Type:	Thesis (Other)
Additional Information:	RSIf 006.33 Lam i-1 2022
Uncontrolled Keywords:	deep learning, klasifikasi teks, klasifikasi multilabel, rekomendasi jawaban. answer recommendation, deep learning, multilabel classification, text classification.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Mr. Marsudiyana -
Date Deposited:	25 May 2026 04:20
Last Modified:	25 May 2026 04:20
URI:	http://repository.its.ac.id/id/eprint/133391

Actions (login required)

View Item