During public health emergencies, a large number of epidemic-related rumor events occur and co-propagate on social media. Personalized recommendation algorithms and many individual and group psychology factors lead to emergence of echo chamber effect in rumor spread, which can reinforce rumor belief and nullify rumor refutation, further affecting people’s correct understanding of scientific epidemic prevention and contributing to the epidemic outbreak.

In this study, the rumor events that attracted much attention on Sina Weibo during COVID-19 pandemic were selected, and all users participating in the discussion of each rumor event were regarded as being in the same virtual community. Firstly, with content analysis, sentiment computing, social network analysis, this study developed quantitative indicators of echo chamber effect for each community in terms of two dimensions: rumor standpoint-distribution homogeneity and standpoint-interaction homogeneity. The differences in the echo chamber effect within different event-communities, based on different interaction-mechanisms, were comparatively analyzed at the macro level. Fitting results of network generation statistical models suggested that users’ cross-community interactions drove the contagion of echo chamber effect between communities, especially standpoint-distribution homogeneity. Similarity analyses suggested that this contagion effect was related to topic and sentiment similarities between communities. Then, with the help of subgraph analysis, content analysis, sentiment computing, and chi-square tests, it was clarified that the significant presence of echo chamber effect within the community amplified the chain spread and radial diffusion of rumor events with the help of cascade modes and reciprocity modes, reduced the quality of interactive content, such as promoting the expression of negative sentiments, lowering the tendency of information seeking, and inducing uncivilized discourses, and weakened the ability of local groups to identify and correct rumors.

Given the significant presence and negative impact of echo chamber effect, next part of this study, based on users’ rumor engagement behavior during the crisis, constructed the polarization indicator for each user at the micro level. Following the process path of “outbreak of the public health emergency - crisis perception - generation of online uncertain information - rumor spread - emergence of rumor echo chamber effect”, this study carried out feature engineering from three aspects, namely, crisis situation factors, information environment factors, and individual attributes. The extracted features and constructed polarization indicator were input as each user-sample into machine learning regression models. Based on regression results of the optimal model, impact degree and direction of single and multiple features on user polarization were visualized. Highly polarized user portrait was outlined, by summarizing the key features which promoted micro user polarization and further drove macro emergence of echo chamber effect. After predicting or targeting highly polarized users, based on users’ “preventive immunization” and “therapeutic immunization”, this study designed management strategies, applicable to the different stages of the formation and reinforcement of the rumor echo chamber effect. Improved UAR and SEIS models were deployed in the three-layer network structure of “community network - user communication network - user contact network”, to describe the competing spread of rumor and knowledge, and the physical diffusion of the epidemic, respectively. In Agent-based simulation evolution experiments, this study observed the governance effectiveness of immunization strategies with different immunity ranges and strengths, from the aspects of depolarization of rumor echo chamber effect, suppression of rumor spread, penetration of knowledge into rumor in communication network, and containment of epidemic diffusion in physical contact network.

The research has made up for the shortcomings in research on the echo chamber effect within and across virtual communities, enriched research related to user polarization under controversial topics during crises, and expanded research on the coupling dynamics of information dissemination, user behavior and epidemic spreading. Findings can help social media platforms, public opinion managers, and public health managers achieve joint intervention and guidance in cyberspace and geo-space, quickly locating highly homogeneous chain rumor events or highly polarized users, implementing targeted governance strategies, improving the speed of response to the infodemic outbreak to achieve the ultimate goal of suppressing the epidemic outbreak.