Automatic classification of mobile apps to ensure safe usage for adolescents

Area of Science:

• Computer Science
• Adolescent Psychology
• Cybersecurity

Background:

• Mobile device integration into adolescent lives necessitates robust safety measures.
• Advancements in 6G internet technology will enable high-quality real-time streaming, increasing exposure risks.
• Accurate classification of mobile applications (M-APPs) is vital for content moderation and adolescent protection.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying M-APPs based on their potential negative impact on adolescents.
• To enhance the security of adolescent mobile usage in anticipation of 6G network capabilities.
• To predict and mitigate exposure to inappropriate content such as violence, pornography, hate speech, and cyberbullying.

Main Methods:

• Utilized Deep Learning techniques, specifically Attentional Convolutional Neural Networks (A-CNNs).
• Employed Bidirectional Encoder Representations from Transformers (BERT) embeddings within the A-CNN architecture.
• Compared the performance of A-CNNs against multiple other Machine and Deep Learning (M/DL) models.

Main Results:

• The proposed A-CNN model achieved an average accuracy of 88.74% in classifying M-APPs.
• The A-CNN model demonstrated an improvement in recall from 99.33% to 99.65% compared to other models.
• A-CNNs based on BERT embeddings outperformed other M/DL models in the classification task.

Conclusions:

• Attentional Convolutional Neural Networks offer a promising approach for securing adolescent mobile usage.
• Accurate M-APP classification is crucial for shielding adolescents from harmful online content.
• The developed A-CNN model effectively predicts the potential negative impact of M-APPs, contributing to enhanced digital safety.