Research on social media user forwarding prediction algorithm based on data analysis

This paper introduces a novel predictive model tailored specifically to the characteristics of user forwarding behavior in social media, which integrates both machine learning and graph neural network (GNN) technologies, with the aim of enhancing the accuracy and resilience of forwarding behavior predictions. Firstly, we comprehensively collect and pre-process data from three dimensions: user personal characteristics, content attribute characteristics and social network structural characteristics to ensure the diversity and representativeness of input features. The core of the model lies in an integrated hybrid architecture that consists of two main parts: the graph neural network module and the machine learning module. Graph neural network module is responsible for processing topological information in social networks, using graph attention network (GAT) to capture complex interactions between nodes and generate node embedding representation. This part focuses on understanding the dynamic changes and structural characteristics within social networks to help extract deep network characteristics. The machine learning module is based on the output of the graph neural network, combined with the basic information and content characteristics of the user, through advanced supervised learning algorithms (such as gradient lifting decision tree GBDT) to train the model. The goal of this part is to synthesize many factors to form the final forwarding probability prediction. Multiple empirical studies have demonstrated that the hybrid model surpasses the single model in terms of performance on diverse real-world data sets, notably in reducing the root mean square error (RMSE) and mean absolute error (MAE). These results prove the validity and practicability of the new model.