Tackling Explicit Material from Online Video Conferencing Software for Education Using Deep Attention Neural...

Area of Science:

• Computer Science
• Artificial Intelligence
• Education Technology

Background:

• The COVID-19 pandemic necessitated a global shift to distance learning, impacting approximately 1.5 billion students.
• Online learning environments have inadvertently exposed minors to inappropriate, particularly sexual, content, posing risks to their emotional and mental well-being.
• Existing policies to protect children online have proven insufficient in addressing the specific challenges of distance education platforms.

Purpose of the Study:

• To present an advanced attention neural architecture designed to detect explicit material within online education video conference applications.
• To introduce a novel, computationally efficient intelligent mechanism for identifying inappropriate content during live educational sessions.
• To enhance the safety and mental health of students participating in remote learning environments.

Main Methods:

• Implementation of a Generative Adversarial Network (GAN) model.
• Integration of a local, sparse attention mechanism to overcome the computational limitations of traditional attention models.
• Development of a system capable of accurately detecting obscene and sexual content in real-time video streams from educational platforms.

Main Results:

• The proposed architecture effectively detects explicit and sexual content in online video conferencing for education.
• The novel attention mechanism achieves high accuracy without the quadratic time and memory complexity associated with standard attention methods.
• This represents a significant advancement in content moderation for educational technology.

Conclusions:

• The developed neural network offers a robust solution for identifying and mitigating exposure to harmful content in distance learning.
• This technology is crucial for ensuring a safer online educational experience for minors.
• The study provides a scalable and efficient method for content moderation in educational video conferencing.