Research On Prediction Algorithm Of Neural Network-based Social Network Information Propagation

In the process of social network information propagation prediction, when usually constructing the message propagation prediction model, the simple factors of establishing the node state transition equations or the inappropriate uses of prediction algorithms can lead to the propagation of information forecasting accuracy low. Therefore, this thesis will reconstruct the social network information propagation model, solve the construction problems of the state transition equation of the prediction model, use the neural network algorithms to optimize the social network information propagation prediction, and improve the information propagation accuracy.This thesis focuses on the social network hybrid information propagation model and social network information propagation prediction algorithm. The main works are as follows:(1) A hybrid social network information propagation model is constructed, which considers the impact of the network topology on the information propagation, introduces the factors including the degree of users’ interest on the information, information value, and information heat in the information propagation processing, and establishes the state transition functions of user nodes.Experiments show that the proposed hybrid information propagation model can well reflect the true propagation rules of social networks.(2) Bases on particle swarm, an optimization BP neural network(BPNN) information propagation prediction algorithm is proposed. The local particle swarm optimization algorithm is used to solve the problems of the slow convergence speed and easily falling into the local minimal values of BP neural network. The error of BP neural network is used as the basis to solve the fitness function of particle swarm optimization algorithm, BPNN weight values and threshold values are modified, and the BPNN prediction model is established. Experiments show that the information propagation prediction accuracy of the proposed algorithm is better than the BP neural network which is optimized by ARIMA or the global particle swarm.(3) Base on LSTM recurrent neural network, a social network information propagation prediction algorithm is proposed, a framework of the social network information propagation model is established, and the information propagation prediction algorithm is designed. Based on a website data, the LSTM information propagation prediction model is constructed, and the blog propagation process in the website is predicted. Experiments show that the information prediction accuracy of the proposed LSTM algorithm is better than the algorithms based on recurrent neural network and convolution neural network.