The influence maximization algorithm for integrating attribute graph clustering and heterogeneous graph transformer.

Area of Science:

• Social Network Analysis
• Artificial Intelligence
• Information Science

Background:

• Traditional influence maximization algorithms neglect node attributes and network heterogeneity, leading to suboptimal propagation efficiency and accuracy.
• Effective influence maximization is crucial for understanding and leveraging information dissemination in social networks.

Purpose of the Study:

• To develop an advanced influence maximization model that incorporates node attributes and network heterogeneity.
• To enhance the accuracy and efficiency of influence propagation in social networks.

Main Methods:

• Constructed a social network influence maximization propagation model.
• Utilized auto-encoder and graph convolutional autoencoder for attribute extraction.
• Employed Transformer and heterogeneous graph neural networks for feature representation learning.
• Designed an algorithm combining attributed graph clustering and heterogeneous graph Transformer.

Main Results:

• The fusion algorithm exhibited strong fitting performance with low loss values (0.619 and 0.17).
• Achieved high clustering accuracy with recall rates of 92.5% and F1 scores of 0.90.
• The influence maximization model demonstrated significant active node coverage (67% and 48% on two datasets).

Conclusions:

• The designed model effectively spreads influence by addressing limitations of traditional methods.
• The findings contribute to a better understanding of influence dissemination mechanisms in social networks.
• This research promotes advancements in social network analysis and application.