Opinion Mining for App Reviews: Identifying and Prioritizing Emerging Issues for Software Maintenance and Evolution

• Ano de defesa: 2023
• Autor(a) principal: Vitor Mesaque Alves de Lima
• Orientador(a): Ricardo Marcondes Marcacini
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: opinion mining, text mining, app reviews, software engineering, software evolution
• Link de acesso: https://repositorio.ufms.br/handle/123456789/6373

Opinion mining for app reviews aims to analyze user comments on app stores to support software engineering activities, primarily software maintenance and evolution. One of the main challenges in maintaining software quality is promptly identifying emerging issues, such as bugs. However, manually analyzing these comments is challenging due to the large amount of textual data. Methods based on machine learning have been employed to automate opinion mining and address this issue. While recent methods have achieved promising results in extracting and categorizing issues from users' opinions, existing studies mainly focus on assisting software engineers in exploring users' historical behavior regarding app functionalities and do not explore mechanisms for trend detection and risk classification of emerging issues. Furthermore, these studies do not cover the entire issue analysis process through an unsupervised approach. This doctoral project advances state of the art in opinion mining for app reviews by proposing an entire automated issue analysis approach to identify, prioritize, and monitor the risk of emerging issues. Our proposal introduces a two-fold approach that (i) identifies possible defective software requirements and trains predictive models for anticipating requirements with a higher probability of negative evaluation and (ii) detect issues in reviews, classifies them in a risk matrix with prioritization levels, and monitors their evolution over time. Additionally, we present a risk matrix construction approach from app reviews using the recent Large Language Models (LLMs). We introduce an analytical data exploration tool that allows engineers to browse the risk matrix, time series, heat map, issue tree, alerts, and notifications. Our goal is to minimize the time between the occurrence of an issue and its correction, enabling the quick identification of problems. We processed over 6.6 million reviews across 20 domains to evaluate our proposal, identifying and ranking the risks associated with nearly 270,000 issues. The results demonstrate the competitiveness of our unsupervised approach compared to existing supervised models. We have proven that opinions extracted from user reviews provide crucial insights into app issues and risks and can be identified early to mitigate their impact. Our opinion mining process implements an entire automated issue analysis with risk-based prioritization and temporal monitoring.