PREDICTION OF USER BEHAVIOR IN SOCIAL HOTSPOTS USING MULTI-MESSAGE INTERACTION AND NEURAL NETWORKS

Abstract

The variety of communications under social hot themes in network public opinion analysis play a significant effect in user engagement behaviour. This paper suggests a prediction model of user participation behaviour during repeated messaging of hot social subjects while taking into account interactions between many messages and complex user behaviors. In order to better forecast user involvement behaviour, a multimessage interaction influence-driving mechanism was first presented. It takes into account how multimessage interaction affects user participation behaviour. Second, based on a multimessage interaction-driving mechanism and a BP neural network, this study proposes a user participant behaviour prediction model of social hotspots in light of the behavioral complexity of users participating in multimessage hotspots and the simple structure of BP neural networks (which can map complex nonlinear relationships). Finally, the user behaviour is iteratively guided by the multimessage interaction, which readily leads to overfitting of the BP neural network. A simulated annealing approach is used to optimize the conventional BP neural network to get around this issue and increase prediction accuracy. In evaluation studies, the model not only predicted user participation patterns in real-world scenarios including many messages, but it also quantified the relationships between various messages on trending subjects.