Abstract: 【Purpose/significance】As one of the major social network platforms in China, Weibo has become an important media for
the dissemination of online public opinion, with real-time, rapid and decentralized information transmission. It is of great practical sig‐
nificance for the control of online public opinion to identify public opinion center figures and mine public emotions for Weibo.
【Method/process】Taking Xinjiang Cotton Event as an example, life cycle method was used to divide the evolution process of Weibo
public opinion. Use word2vec and k-means models to extract public opinion central figures in each stage of the event life cycle. A sen‐
timent analysis method combining dictionary and LSTM deep learning model was adopted. Conduct polarity analysis on the comments
and emotions related to people in the public opinion centers.【Result/conclusion】The proposed method can well mine public opinion
center figures, the type and intensity of the public's emotions, Get the guidance method that can make public opinion turn for the bet‐
ter.【Innovation/limitation】This paper innovatively applies the theme mining method to the extraction of microblog public opinion cen‐tral figures. In terms of emotion analysis method, the dictionary and deep learning method are combined to solve the limitation of deep learning method requiring manual annotation in emotion analysis. In addition, the effect of emoticons was not taken into account in the calculation of emoticons in this paper. Further studies will consider more fine-grained emotional classification.