Javascript must be enabled for the correct page display

Developing Deep Learning Approaches to Find and Classify Architectural Design Decisions in Issue Tracking Systems

Maarleveld, Jesse and Dekker, Arjan (2023) Developing Deep Learning Approaches to Find and Classify Architectural Design Decisions in Issue Tracking Systems. Master's Thesis / Essay, Computing Science.


Download (3MB) | Preview
[img] Text
Restricted to Registered users only

Download (131kB)


Architectural design decisions (ADDs) are considered to be an important part of the software architecture, but they are often not explicitly documented, making it challenging to understand the rationale behind a system’s structure. This lack of documentation complicates software maintenance and evolution. Additionally, software architects frequently rely on existing ADDs as a basis for creating new ones. Often, they make use of their own experience from past decisions instead of documented ADDs. Recent studies show that ADDs do tend to be discussed implicitly in some places, such as issue trackers in open-source systems. However, identifying these discussions is difficult. In this work, we built upon previous efforts to leverage deep learning techniques for finding ADDs in issue tracking systems by Dekker and Maarleveld (2022). In this work, we extended the dataset used there from 2179 to 6225 issues. Moreover, we performed a more fine-grained classification of issues while also improving classifier performance to 0.67 F1 score. We also investigated the abilities of classifiers to generalise to different projects and different domains. Our main finding is that deep learning models, and in particular large language models such as BERT, are a promising search tool to find ADDs in issue tracking systems since they are able to find ADDs with a precision ≥ 0.63 even when applied to different domains.

Item Type: Thesis (Master's Thesis / Essay)
Supervisor name: Soliman, M.A.M. and Avgeriou, P.
Degree programme: Computing Science
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 31 Aug 2023 14:00
Last Modified: 31 Aug 2023 14:00

Actions (login required)

View Item View Item