Educational Data Mining by Optimally Fusing Shallow and Deep Features.

Area of Science:

• Artificial Intelligence
• Educational Technology
• Computer Vision

Background:

• Traditional recommendation algorithms lack personalized educational insights.
• Effective extraction of heterogeneous educational data is challenging.
• Detecting student emotions and academic integrity requires advanced methods.

Purpose of the Study:

• To develop a deep learning framework for personalized understanding and educational feature extraction in students.
• To propose requirements for intelligent courses addressing limitations of existing recommendation systems.
• To enable accurate emotion identification and detection of academic misconduct.

Main Methods:

• Utilized a word vector permutation discourse poke model for message data.
• Developed a learning algorithm to extract and fuse heterogeneous educational data.
• Applied Singular Value Decomposition (SVD) for feature vector reduction.
• Combined hand-extracted and deep network-extracted features for emotion recognition.
• Employed VGG16 and fine-tuned AlexNet for experimental validation.

Main Results:

• Successfully extracted personalized educational features and fused multiple data types.
• Achieved effective identification of student emotions through combined feature extraction.
• Developed a system capable of detecting copying and objection.
• Demonstrated model functionality in identifying unified channels and users.

Conclusions:

• The proposed deep learning architecture enhances personalized learning and educational feature extraction.
• The system effectively identifies student emotions and academic integrity issues.
• This approach offers a robust solution for intelligent course design and student monitoring.