Ekstraksi Kebutuhan Nonfungsional dari Ulasan Pengguna ke Atribut Kualitas Perangkat Lunak Menggunakan Pendekatan Pembelajaran Mesin

Aldriantama, Muhammad Daffa (2024) Ekstraksi Kebutuhan Nonfungsional dari Ulasan Pengguna ke Atribut Kualitas Perangkat Lunak Menggunakan Pendekatan Pembelajaran Mesin. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025201177-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (7MB) | Request a copy

Abstract

Rekayasa kebutuhan dalam rekayasa perangkat lunak melibatkan penemuan, dokumentasi, dan pemeliharaan kebutuhan sistem. Kebutuhan tersebut dibagi menjadi kebutuhan fungsional (FR) dan kebutuhan nonfungsional (NFR). Pentingnya NFR ditunjukkan oleh kegagalan proyek United States Army yang berbiaya 2,7 miliar dolar akibat masalah performa dan kegunaan. NFR dapat diidentifikasi melalui ulasan pengguna aplikasi, tetapi jumlah ulasan pengguna yang sangat banyak dapat menghambat proses identifikasi tersebut. Penelitian sebelumnya menggunakan algoritma klasifikasi SVM dengan CountVectorizer sebagai metode praproses NLP-nya untuk melakukan klasifikasi ulasan pengguna aplikasi. Hasil penelitian tersebut menunjukkan bahwa penggunaan kedua metode tersebut menghasilkan F1-score yang rendah, yaitu 0,5874. Rendahnya hasil penelitian tersebut disebabkan oleh cara kerja CountVectorizer yang tidak memperhatikan informasi kontekstual. Untuk mengatasi kelemahan tersebut, penelitian ini menggunakan pendekatan feature-based dan fine-tuning. Feature-based dilakukan dengan menggunakan praproses NLP DistilBert yang dikombinasikan dengan algoritma NB dan SVM. Fine-tuning dilakukan dengan menyetel model DistilBERT sehingga model dapat langsung melakukan klasifikasi. Hasil dari penelitian ini menunjukkan bahwa metode DistilBERT fine-tuning memberikan hasil terbaik. Metode ini menghasilkan nilai evaluasi F1-score 0,7072, precision 0,6753, recall 0,7433, dan accuracy 0,8168.
=====================================================================================================================================
Requirement Engineering in software engineering involves the discovery, documentation, and maintenance of system requirements. These requirements are divided into functional requirements (FR) and non-functional requirements (NFR). The importance of NFR is highlighted by the failure of the United States Army project, which cost 2.7 billion dollars due to performance and usability issues. NFR can be identified through application user reviews, but the vast number of user reviews can hinder the identification process.
Previous research used the SVM classification algorithm with CountVectorizer as its NLP preprocessing method to classify application user reviews. The results of this research showed that the use of these two methods resulted in a low F1-score of 0.5874. The low results were due to CountVectorizer's approach, which does not take contextual information into account. To address this weakness, this study employs a feature-based and fine-tuning approach. The feature-based approach uses DistilBERT NLP preprocessing combined with the NB and SVM algorithms. Fine-tuning is done by adjusting the DistilBERT model so that the model can directly perform classification.
The results of this study show that the DistilBERT fine-tuning method provides the best results. This method produces evaluation scores with an F1-score of 0,7072, precision of 0,6753, recall of 0,7433, dan accuracy of 0,8168.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Kebutuhan Nonfungsional, Ulasan Aplikasi, NLP, DistilBERT Fine-tuning, Naive Bayes, SVM, MLSMOTE, Nonfunctional Requirements (NFR), App Review, NLP, DistilBERT Fine-tuning, NB, SVM, MLSMOTE
Subjects:	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:	Muhammad Daffa Aldriantama
Date Deposited:	01 Aug 2024 02:00
Last Modified:	11 Sep 2024 03:45
URI:	http://repository.its.ac.id/id/eprint/111259

Actions (login required)

View Item