[Objective] In order to control and intervene the spread and development of public opinion to a certain extent.

[Methods] An opinion leader mining model OLMT is firstly proposed, which integrates the propagation force and topological potential, so that opinion leader mining can be carried out from more angles and more objectively. In addition, by transforming the Transformer model used for machine translation, a social network propagation behavior prediction model mf-transformer is constructed. By using its high parallelism and attention mechanism, the forwarding behavior of opinion leaders can be predicted more efficiently and accurately.

[Results] Combining with the mining results of opinion leaders and the prediction results of propagation behaviors, the critical propagation paths composed of opinion leaders are effectively predicted in the process of public opinion propagation.

[limitation] The research and experiment of this paper are mainly oriented to Sina Weibo.

[Conclusions] The experimental results of an actual case of public opinion propagation in microblog show that the opinion leader mining model and propagation behavior prediction model proposed in this article can not only mine opinion leaders more accurately, but also effectively predict the critical paths in the process of public opinion propagation.