Persistent Homology Combined with Machine Learning for Social Network Activity Analysis

Currently, the rapid development of social media enables people to communicate more and more frequently in the network. Classifying user activities in social networks helps to better understand user behavior in social networks. This paper first creates an ego network for each user, encodes the higher-order topological features of the ego network as persistence diagrams using persistence homology, and computes the persistence entropy. Then, based on the persistence entropy, this paper defines the Norm Entropy-NE(X) to represent the complexity of the topological features of the ego network, a larger NE(X) indicates a higher topological complexity, i.e., the higher the activity of the nodes, thus indicating the degree of activity of the nodes. The paper uses the extracted set of feature vectors to train the machine learning model to classify the users in the social network. Numerical experiments are conducted to evaluate the performance of clustering quality metrics such as profile coefficients. The results show that the proposed algorithm can effectively classify social network users into different groups, which provides a good foundation for further research and application.