Core Concepts
The author explores the automated detection of self-admitted technical debt using natural language processing and machine learning algorithms to assist developers efficiently.
Abstract
The systematic literature review examines the use of NLP and ML/DL algorithms in detecting technical debt. Various feature extraction techniques are compared for performance across different software development activities. The study highlights the importance of addressing technical debt early to prevent future issues.
The content discusses the prevalence of technical debt in software development, emphasizing the trade-offs made during development that can impact maintainability. It delves into self-admitted technical debt (SATD) and its acknowledgment within source code comments by developers. Automated approaches using NLP and ML/DL algorithms are explored to enhance efficiency in identifying and managing technical debt.
Key points include the taxonomy of feature extraction techniques, comparison of ML/DL algorithms, mapping TD types to software development activities, and implications for researchers and practitioners. The study provides insights into improving performance in detecting technical debt through automated approaches.
Stats
SATD instances ranged from 2.4% to 31% in open-source projects.
23 types of technical debt were identified across various software development activities.
Pre-trained embeddings like BERT were used for efficient detection of SATD instances.