CST-Net: community-guided structural-temporal convolutional networks for popularity prediction

Area of Science:

• Computational Social Science
• Machine Learning
• Network Science

Background:

• Predicting online content popularity is vital across various domains.
• Challenges include popularity inequality and complex influencing factors.
• Existing methods (feature-driven, generative, deep learning) have limitations.

Purpose of the Study:

• To introduce CST-Net, an end-to-end deep learning framework for improved popularity prediction.
• To address the shortcomings of current popularity prediction methodologies.

Main Methods:

• Learned low-dimensional user embeddings from historical interactions.
• Clustered users into communities and represented information cascades as community interaction matrices.
• Applied a convolutional architecture to extract cascade representations.
• Combined structural and temporal features for incremental popularity prediction.

Main Results:

• CST-Net demonstrated superior performance on microblogging and academic citation datasets.
• The model consistently outperformed existing competitive popularity prediction methods.
• Validation on population-scale datasets confirmed effectiveness.

Conclusions:

• CST-Net offers a robust and effective approach to predicting online content popularity.
• The framework's ability to capture complex cascade dynamics is key to its success.
• This work advances the field of computational social science and predictive modeling.