Research On The Evolution Analysis And Prediction Model Of Public Opinion On Social Networks

The characteristics of online social networks determine that all kinds of information published on them can attract the attention of a large number of users in a very short period of time.Therefore,many public opinion event information is more easily spread on social networks.have a huge impact.This article uses the public opinion data generated on Sina Weibo,a large social network platform,to model the evolution of public opinion through empirical analysis of the data.The specific work includes the following three aspects.First,by crawling the public opinion data in the public domain on Sina Weibo,this article studies the impact of the cascading accumulation of forwarding behaviors of users participating in public opinion events on the evolution of public opinion,and standardizes the evolution of public opinion as a ”cascading process”.This article empirically analyzes the topological structure characteristics of the public opinion dissemination network and the timing characteristics of the cascading process based on forwarding behavior.The research results show that the public opinion dissemination network has a tree topology,scale-free distribution characteristics and disparate structure,and the time series characteristics of the public opinion cascade process are non-Markovian(or memory).These empirical results will be applied to the follow-up Modeling research on the evolution of public opinion.Second,based on the empirical analysis results of the topological structure characteristics of the public opinion communication network and the timing characteristics of the cascading process,this article proposes an activity-driven independent cascading model.In terms of network modeling,the model integrates scale-free distribution and user activity heterogeneity,considers user cascading behavior memory in terms of timing,and integrates the influence of user influence and ”public opinion” popularity.In this article,a single cascade process is simulated based on the proposed independent cascade model through a data-driven approach to verify its effectiveness.Furthermore,this article considers the synergy of multiple cascading processes in the same public opinion event,and improves the independent cascading model,so that the improved model can capture the multi-cascading cooperative communication relationship and simulate the interaction between multi-cascading processes.Additive effect.Third,this article studies the long-term cascading recurrence process of public opinion events,and finds that cascading recurrence is ubiquitous,and there is a long-term correlation between recurrence cascades.On this basis,this article proposes a model for predicting the recurrent process of cascades,the Multi-Cascade Evolutionary Prediction Model(Recur MLP-SIS),which combines the Cascade Scale Prediction Algorithm and the Cascade Superposition Algorithm.Among them,the cascade scale prediction algorithm uses multi-layer perceptron to learn the cooperative propagation characteristics between recurrent cascades,and predicts the recurrent cascade scale through the historical cascade process;the cascade superposition algorithm uses the output of the cascade scale prediction algorithm to The recurrence cascade process is predicted.Compared with the SOTA model used for cascade prediction,the model has significantly improved in multiple indicators,especially the MAPE indicator has improved by more than 8%.Therefore,the results of comparative experiments show that the Recur MLP-SIS model proposed in this article has better advancedness in predicting the cascade recurrence process,and the cascade simulation model has better interpretability.