Prediction Models for Network Multi-source Dissemination of Information Based on Multivariate Chaotic Time Series

Accurately predicting the future tendency of network sentiment plays a critical role in supervising and guiding the diffusion of network public event. In response to the distinction of different network platform in occurring time and dissemination mechanism, according to information diffusion theory, this paper adopts data on multi-platform, which could more completely describe the propagation of hot events. Univariate time series is extended to multivariate time series in order to improve the forecasting accuracy. At the same time, the phase space of multivariate time series from diverse platform is reconstructed and the existence of chaotic characteristic of real multi-source network information diffusion is proved by small-data method. In the end, the predicting model of wavelet neural network based on chaotic time series is constructed to forecast the future public opinion trend of real information. Simulation experiments show that compared with the way of prediction by the unit chaotic time series, multivariate time series has better predictive effects, which provides a new way for forecasting online hotspots.