Mining software insights: uncovering the frequently occurring issues in low-rating software applications

Area of Science:

• Software Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• App stores are crucial for software distribution, but research often overlooks feedback from low-rated applications.
• End-user feedback is vital for software evolution and quality improvement, yet underutilized for less popular apps.
• Existing studies predominantly focus on high-rating apps, neglecting valuable insights from low-rated ones.

Purpose of the Study:

• To analyze end-user feedback from low-rated apps to identify critical factors for software evolution.
• To develop and evaluate machine learning (ML) and deep learning (DL) models for classifying user feedback issues.
• To improve the explainability of ML/DL models using the SHAP approach for actionable insights.

Main Methods:

• Collected and analyzed end-user reviews from 64 low-rated apps across 14 categories in the Amazon App Store.
• Developed a grounded theory to categorize feedback into UI/UX, functionality, compatibility, performance, support, and security.
• Applied natural language processing (NLP) and feature engineering to train and optimize ML/DL classifiers (e.g., MLP, RF, BiGRU, CNN, LSTM).

Main Results:

• Identified key areas for software improvement including UI/UX, functionality, performance, and customer support based on user feedback.
• Achieved high classification accuracies for user feedback issues using various ML/DL models, with MLP and RF reaching 94%.
• The SHAP approach successfully identified critical features influencing the classification of specific user-reported issues.

Conclusions:

• End-user feedback from low-rated apps provides essential insights for enhancing software quality and user experience.
• ML and DL models demonstrate significant potential for automatically classifying user feedback, aiding developers in prioritizing improvements.
• This research offers a data-driven approach for developers to address shortcomings in low-rated applications and improve overall software.