
: Predicting student satisfaction on Metaverse platforms can improve educational effectiveness by analyzing factors like engagement, interaction, and content delivery. The benefits include personalized, immersive learning experiences and real-time feedback. However, the existing schemes suffer from overfitting issues due to highly imbalanced data and fail to generalize to diverse content. To address these limitations, a robust model, Deep High Attention Long Short-Term Memory Forward Harmonic Network (DHA-LSTM-FH Net), is proposed to predict student satisfaction in Virtual Reality (VR) teaching within the Metaverse. The model utilizes data from Metaverse platforms, including virtual spaces, Augmented Reality (AR)/VR devices, learning materials, and student information. Initially, interaction logs from VR sessions and student profiles are collected as input. The data undergoes softmax normalization to ensure consistency. Feature selection is conducted using Recursive Feature Elimination (RFE) and Elastic Net to select the key features. Local Densitybased Synthetic Minority Over-Sampling Technique (LD-SMOTE) is then applied to address data imbalance. Student satisfaction prediction is done by DHA-LSTM-FH Net, which is developed by combining Deep High Attention Neural Network (DHA-Net) and Long ShortDownloaded for personal academic use. All rights reserved. https://papernode.online/ Term Memory (LSTM) using Harmonic Analysis. Experimental results show that the model achieves a precision of 93.765%, a recall of 95.755%, an F1 Score of 94.750%, and a Cohen’s Kappa of 0.838, outperforming baseline methods. However, the model is trained on a specific VR/Metaverse platform, so its performance may drop when applied to different Metaverse setups or content types.
Download PDF: https://tirna.eu.org/fQPWZ9





