Klasifikasi Multi-label Aspek Fungsional Dan Non-fungsional Berbasis Transformer Pada Ulasan Pengguna Aplikasi Mobile

Hidayah, Wahid Nur (2025) Klasifikasi Multi-label Aspek Fungsional Dan Non-fungsional Berbasis Transformer Pada Ulasan Pengguna Aplikasi Mobile. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032231268-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Perkembangan transformasi digital yang pesat menyebabkan banyaknya aplikasi mobile yang dikembangkan. Hal tersebut meningkatkan interaksi antara produk/layanan perusahaan dengan pengguna. Memahami pengalaman pengguna menjadi salah satu aspek penting dalam meningkatkan kualitas aplikasi dan mempertahankan pengguna dari persaingan dengan kompetitor. Salah satu cara untuk memahami pengalaman pengguna adalah melalui ulasan pada platform toko aplikasi seperti Google Play Store. Ulasan pengguna dapat diklasifikasikan ke dalam aspek fungsional dan non-fungsional. Aspek fungsional adalah yang berkaitan langsung dengan permintaan fitur yang harus disediakan aplikasi atau laporan masalah yang dihadapi pengguna pada spesifik fitur tertentu. Sedangkan aspek non-fungsional menggambarkan bagaimana sistem bekerja, mencakup kualitas performa, kemudahan penggunaan, keandalan serta kualitas pelayanannya. Penelitian ini melakukan klasifikasi aspek fungsional (bug report, feature request) dan non-fungsional (dependability, performance, supportability, usability), membangun model klasifikasi dengan transformer yaitu Robustly Optimized BERT Pretraining Approach (RoBERTa) serta memberi rekomendasi untuk membantu pihak perusahaan menentukan arah pengembangan aplikasi mobile. Model menunjukkan performa terbaik dengan nilai macro precision, macro recall, macro f1-score, hamming loss dan subset accuracy berturut-turut adalah 0.7960, 0.8295, 0.8118, 0.0924 dan 0.6292. Performa didapat saat model dilatih pada data hasil random swap augmentation dengan target distribusi label dominan menggunakan hyperparameter learning rate 0.00005, batch size 16 dan weight decay 0.1. Aspek supportability menjadi aspek yang paling dominan sehingga perlu diprioritaskan dimana pengembang aplikasi dapat mempertimbangkan penggunaan chatbot, melatih customer service untuk mempercepat pelayanan serta melakukan pengawasan dan sistem reward and punishment untuk seller dan kurir ekspedisi.
=====================================================================================================================================
The growth of digital transformation has caused many mobile applications to be developed. This increases the interaction between the company's products/services and users. Understanding user experience is one of the important aspects in improving app quality and retaining users from competitors. One way to understand user experience is through reviews on app store platforms like Google Play Store. User reviews can be classified into functional and non-functional requirements. Functional requirements are those that are directly related to feature requests that the application should provide or reports of problems faced by users on specific features. Non-functional requirements describe how the system works, including performance quality, ease of use, reliability and service quality. This research aims to classify functional (bug reports, feature requests) and non-functional aspects (dependability, performance, supportability, usability), build classification models with transformers, Robustly Optimized BERT Pretraining Approach (RoBERTa) and provide recommendations to help companies determine the direction of mobile application development. The model shows the best performance with macro precision, macro recall, macro f1-score, hamming loss and subset accuracy values of 0.7960, 0.8295, 0.8118, 0.0924 and 0.6292, respectively. The performance was obtained when the model was trained using data from random swap augmentation with the target of dominant label distribution and using hyperparameter learning rate 0.00005, batch size 16 dan weight decay 0.1. It was found that the supportability aspect is the most dominant aspect that needs to be prioritized. The management of the application developer can consider using a chatbot, train customer service to speed up service and supervise and reward and punishment system for sellers and expedition couriers.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Klasifikasi Multi-label, Aspek Fungsional dan Non-fungsional, Ulasan Pengguna Aplikasi Mobile, Transformer, Multi-label Classification, Functional and Non-Functional Requirements, Mobile App User Review, Transformer
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Wahid Nur Hidayah
Date Deposited:	30 Jul 2025 06:37
Last Modified:	30 Jul 2025 06:37
URI:	http://repository.its.ac.id/id/eprint/122945

Actions (login required)

View Item