Pengembangan Model Chatbot Berbasis Large Language Model untuk Pasien Continous Ambulatory Peritonial Dialysis

Amin, Muhammad (2024) Pengembangan Model Chatbot Berbasis Large Language Model untuk Pasien Continous Ambulatory Peritonial Dialysis. Other thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 5025201251-Undergraduate_Thesis.pdf] Text
5025201251-Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2026.

Download (2MB) | Request a copy

Abstract

Continuous Ambulatory Peritoneal Dialysis (CAPD) merupakan opsi terapi pengobatan untuk pasien dengan Penyakit Ginjal Tahap Akhir (PGTA) di Indonesia. CAPD menawarkan dialisis mandiri tanpa perlu ke rumah sakit, dengan kualitas hidup lebih tinggi dibanding metode lainnya. Namun, risiko kematian pada pasien CAPD tinggi akibat kerumitan yang disebabkan oleh kelalaian dan kesalahan teknis. Oleh karena itu, diusulkan pengembangan model chatbot menggunakan dataset FAQ para pasien CAPD. Dengan dataset berupa FAQ dari berbagai macam sumber yang telah divalidasi oleh dokter dari Rumah Sakit Universitas Airlangga. Penelitian ini menjelaskan mengenai cara pembuatan chatbot pada aplikasi sahabat capd untuk membantu pasien capd. Dataset FAQ telah dilakukan parafrase dan preproses agar data dapat dijadikan dataset untuk training chatbot. Pada tahap selanjutnya, data telah dipisahkan menjadi train dan validation dengan persentase 80% untuk train. Beberapa arsitektur embedding model seperti BERT akan dicoba dengan ditambahkan layer ekstra untuk menyesuaikan luaran yang diinginkan untuk mendapatkan chatbot terefektif. Model Chatbot juga diimplementasi menggunakan RAG dengan LLM generasi teks seperti mistral, phi, aya. Pada model deep learning, didapatkan rerata akurasi test menggunakan pertanyaan yang ditulis manual sebesar 67%. Transfer Learning berhasil meningkatkan performa hingga 75%. Pada RAG, rerata nilai ROUGE yang dihasilkan melalui LLM model generasi teks sebesar 0.33 dan mampu menjawab beragam pertanyaan dalam satu kueri. Akan tetapi, waktu yang dibutuhkan untuk memberikan prediksi jawaban mulai dari 1 detik hingga 40 detik.
==============================================================================================================================
Continous Ambulatory Peritoneal Dialysis (CAPD) is a treatment therapy option for patients with End Stage Renal Disease (PGTA) in Indonesia. CAPD offers independent dialysis without the need to go to hospital, with a higher quality of life than other known methods. However, the risk of death in CAPD patients is high due to complications caused by negligence and technical errors. Therefore, a chatbot is proposed to be developed using the FAQ dataset of CAPD patients. With a dataset in the form of FAQs from various sources that have been validated by doctors from Rumah Sakit Universitas Airlangga. This research explained how to create a chatbot for Sahabat CAPD application to help CAPD patients. The FAQ dataset is preprocessed so that the data can be used as dataset for chatbot training. In the next stage, the data is splat to train and validation in a 4:1 ratio. Several model embedding architectures such as BERT is tried with extra layers added to adjust the desired output to get the most effective chatbot. The Chatbot model is also implemented using RAG with LLM generation text such as mistral, phi, aya. With Deep Learning, the average test accuracy using manually written questions is 67%. The use of model embedding succeeded in increasing performance by up to 75%. In RAG, the average ROUGE value produced through the LLM text generation model is 0.33 and is able to answer various questions in one query. However, the time required to provided a predicted answer ranges from 1 to 40 seconds.

Item Type: Thesis (Other)
Uncontrolled Keywords: Continuous Ambulatory Peritoneal Dialysis (CAPD), Chatbot, Model Embedding, Large Language Model (LLM)
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User: Muhammad Amin
Date Deposited: 22 Aug 2024 04:30
Last Modified: 22 Aug 2024 04:30
URI: http://repository.its.ac.id/id/eprint/112317

Actions (login required)

View Item View Item