TEP-BS: Public opinion evolution prediction based on stochastic competitive learning-taking the hot case of platform...

Rapid advances in social media have made the evolution of public opinion both highly dynamic and complex. An accurate prediction of this evolution is vital for government and corporate decision-making. To address the challenge of predicting the evolution of public opinion in social networks, this paper first proposes the BRT model, which incorporates the structural features and temporal weight optimization of social networks into the topic modeling process based on the BERTopic framework, thereby achieving high-precision topic identification in social network scenarios. Next, we model topic competition as a particle-dynamic game and propose TEP-BS based on BRT, a prediction model based on Stochastic Competitive Learning (SCL). TEP-BS employs particle stochastic walks, propagation-decay, and node-control mechanisms to forecast public opinion evolution. Experiments on four trending datasets from September 2024 show that the BRT and TEP-BS significantly outperform current state-of-the-art methods in network topic distribution calculation and public opinion evolution prediction, the F1 score of TEP-BS on those datasets is increased by an average of 60 %, 42 %, 36 %, and 44 % compared with TF-IDF-LDA, NMF, HDP, and GAT respectively.