In recent years, online education has morphed from an experimental novelty to a mainstream educational tool, witnessing explosive growth. Since 2000, the online learning market has surged by an astonishing 900%. By 2024, it is expected to generate revenues of $185.20 billion, and by 2028, this number is projected to skyrocket to $257.70 billion, with an annual growth rate of 8.61%. More impressively, by the end of 2028, online learning platforms are expected to engage 1 billion users worldwide.

The transition to online education offers numerous advantages over traditional methods. Flexibility, convenience, self-paced learning, and accessibility are just a few examples. This mode of learning enhances control over the educational environment. However, like every coin has two sides, online education also faces challenges. Particularly daunting is the task of monitoring progress and quality, given the vast number of users and the physical separation between teachers and students.

Ensuring high-quality education remains a paramount priority, aligning with the United Nations Sustainable Development Goals. Therefore, educational organizations must maintain rigorous oversight of their educational processes.

Traditionally, quality assessment in online education has relied on direct comparisons of student responses with desired answers (tests, assignments) and evaluating educational metrics (return rates, Completion Rates, Success Rates, attendance statistics). Although effective, these methods fall short in systematically informing educators and course developers about student progress.

This gap has driven a surge in the adoption of automated feedback systems and quality analysis based on machine learning techniques. At ProgKids, we're at the forefront of this innovative wave.

The ProgKids Approach

At ProgKids, our programming school leverages an advanced system to analyze the engagement of both teachers and students, an essential indicator of educational quality. This system employs a series of Docker-based services structured as follows:

1. User API—developed using Flask.
2. Audio Engagement Analysis Module:
   • File Processor—validates uploaded audio files and converts them into the required format using ffmpeg.
   • Speech Recognition Module—utilizes a modified version of SOVA ASR.
   • Emotion Analysis Module—uses the SpeechBrain model to detect emotions in audio.
   • Transcript Analysis Module—analyzes the recognized text.
3. Video Engagement Analysis Module:
   • Face Detection Module—identifies faces in video.
   • Gaze Detection Module—determines the direction of the participant's gaze.

Our services adhere to a microservice architectural style, ensuring a service-oriented structure with loosely coupled, easily modifiable modules that interact via API.

Deliverables of the System

Upon completion of the analysis, our system provides:

1. Speech Analysis:
   • Presence of pauses, their count and duration, list of filler words with start times, and lists of polite and impolite words.
2. Video Analysis:
   • Absence of the participant's face and duration of absence.
   • Instances of looking away and the duration of such distractions.

The Impact

Implemented at ProgKids since December 2022, this system has significantly boosted our student completion rates. By identifying decreased engagement, we can promptly inform educators, prompting them to enhance student involvement, or if necessary, replace the instructor.

Future Potential

Our work in this arena demonstrates significant potential for providing invaluable insights. The data collected underscores the effectiveness of our proposed quality assessment strategy. This approach not only aids in maintaining high educational standards but also helps predict student dropout rates, steering the future of online education toward greater success.