An empirical study of automated privacy requirements classification in issue reports

Abstract The recent advent of data protection laws and regulations has emerged to protect privacy personal information individuals. As the cases breaches vulnerabilities are rapidly increasing, people aware more concerned about their privacy. These bring a significant attention software development teams address concerns in developing applications. today’s adopts an agile, issue-driven approach, issues issue tracking system become centralised pool that gathers new requirements, requests for modification all tasks project. Hence, establishing alignment between those requirements is important step privacy-aware systems. This also facilitates compliance checking which may be required as underlying part organisations. However, manually alignments labour intensive time consuming. In this paper, we explore wide range machine learning natural language processing techniques can automatically classify reports. We employ six popular namely Bag-of-Words (BoW), N-gram Inverse Document Frequency (N-gram IDF), Term Frequency-Inverse (TF-IDF), Word2Vec, Convolutional Neural Network (CNN) Bidirectional Encoder Representations from Transformers (BERT) perform classification on privacy-related reports Google Chrome Moodle projects. evaluation showed BoW, IDF, TF-IDF Word2Vec suitable classifying addition, IDF best performer both