Time Series Analysis On Social Network

In the past decades,Internet has experienced a period of rapid development,and become a new media beyond three traditional media of newspaper,radio and television.Hot events are the representative of the network information.They not only affect people's attitudes and opinions in realistic society,but also affect the direction of public opinion.The information propagation process is no longer restricted by time and space by means of social network,which significantly accelerates the speed of spreading.Therefore,analyzing the propagation process of hot events,and controlling the negative rhetoric is very concerned.This thesis focus on the perspective of time series to analysis propagation rules of hot events.Even though the characteristics and attributes of each individual in social network is different,the collective behavior of all individuals would offset of the differences,resulting in the overall trend,which turns to the waveform changes in time series.Firstly 300 hot events are crawled from Sina-Weibo,then they are clustered to three centroids by applying the K-Spectral Centroid(K-SC)clustering algorithm,and find that the diffusion process is divided into two step as the temporal patterns are consist of two spikes.The differences between two stages are the proportion of the two spike and the peak point in time.For the purpose of fitting the centroids,a new model named SpikeM-G is introduced base on SpikeM.In SpikeM-G model,both the first stage caused by opinion lead and the second stage on behalf of the general audience in the process of the propagation are considered,the characteristic of power law distribution is modeled as well.The simulation results demonstrate that the new model describes all the rise and fall patterns with high accuracy,while SpikeM is only capable of fitting the first spike.The fitted model parameters show that the first stage has a shorter duration,while the second stage has a longer lasting time.It illustrates that the information is spread among key nodes of the social network,and then it will cover the entire network in the second stage.