Deep Learning-Based Popularity Prediction for Short Videos in IoT Environments

Abstract

This study focuses on predicting the popularity of short videos by combining Internet of Things (IoT) data with deep learning techniques. The proposed CPRP-CNN model uses simple yet effective inputs, such as the creator’s personal attributes and textual content, to estimate how a video will perform immediately after it is published. This approach is especially useful in cross-cultural communication, where content appeal may vary across audiences. Experimental results show that using the ReLU activation function significantly improves model performance compared to the sigmoid function, leading to higher accuracy and lower loss values. The model achieves an accuracy of 74.7% along with reduced error metrics like MSE and MAE, indicating reliable predictions. Compared to other methods, CPRP-CNN delivers better results and demonstrates the importance of integrating IoT features with deep learning. Overall, this work helps content creators and platforms make informed decisions and improve personalized content recommendations for users.