Zongmin Li

The massive spread of misinformation on social media platforms not only leads to the generation of irrational behaviors of users, but also threatens the social order. Users occupy an important position in the governance process of misinformation. In this study, users' willingness to forward refuting information is taken as an important indicator to measure the effectiveness of misinformation governance. The neutralization technique and norm activation model are applied to misinformation governance. Psychological mechanism, normative ethics, demographic informatics characteristics, and misinformation refuting methods are all considered as indicators for predicting users' forwarding behaviors. Based on support vector machine, XGBoost, random forest and artificial neural network algorithms, four prediction models are constructed respectively. The results show that all four categories of predictive indicators have good predictability. Three types of indicators related to users' personal characteristics are more important than refuting methods. XGBoost and random forest algorithms are more suitable for constructing the prediction models in this paper. Finally, specific suggestions are proposed for the government department's misinformation governance.

The world has seen a substantial number of disastrous emergencies, which lead to heated discussion on the Internet and then chaos without proper interference. Motivated by others’ emotions through online discussion, netizens actively interact with each other and flock together, eventually reaching to the same emotional state. That's what emotion polarization means. To guide the netizens, influencing factors accounting for emotion polarization must be identified. In this paper, we take the China Eastern Airlines MU5735 crash as an example. All the hot topics,comments and retweets concerning the air crash on Sina Weibo are collected by a web crawler from March 21 to April 1, 2022. The reciprocal of sentiment entropy is first used to measure emotion polarization. 7 main factors affecting emotion polarization are identified based on Hovland's Persuasion Model. In order to mine key influencing factors, four machine learning models are established and tuned, among which Adaptive Boosting regression model (AdaBoost) has the best performance. Then three influencing factors are mined by feature importance ranking of AdaBoost. By SHapley Additive exPlanations (SHAP), a more concise outcome of feature importance ranking is represented. Finallly, according to the figures of feature importance ranking, decision-making suggestions on facilitating emotion polarization is proposed.