4.1.1. The Patterns of Online Self-Organized Groups
K-means clustering was employed to explore the structural characteristics and patterns of online self-organized groups. We selected six critical indicators as the clustering standards: number of followers, number of Weibo posts, in-degree, out-degree, betweenness centrality, and closeness centrality. Specifically, number of followers represented the influence of the node; number of Weibo posts represented the activity of the node; in-degree represented the number of times the node was mentioned by other nodes; out-degree represented the number of times the node referenced to or mentioned other nodes; betweenness centrality represented the number of shortest path through the node in the network, reflecting the node’s ability to control the information exchange in the network; finally, closeness centrality represented the average distance from one node to all other nodes in the network, which could range from 0 to 1, and the bigger its value, the farther its distance to other nodes. The parameter K of K-means clustering was determined by the elbow method. In the end, we divided the online self-organized groups into ten classes. Results are shown in
Table 5.
According to our clustering results, online self-organized groups contain seven patterns, with three outliers that do not belong to any pattern. Each pattern and outlier was characterized to reveal the composition of online self-organized groups. In
Figure 2, the node depth of each node is 1.
Cluster 1 refers to Shanghai netizens active on Weibo. As shown in
Table 5, this cluster has middle-range indicator values (i.e., mean of in-degree = 2.20, mean of out-degree = 1.22, mean of betweenness centrality = 29,038.48, mean of closeness centrality = 0.19). This group has a relatively high number of followers (1,622,873.46) and posts (43,643.85) among these seven categories. These Shanghai netizens have been active on Weibo for a long time. Their online behaviors manifest primarily in consulting relief measures, raising epidemic prevention problems, participating in epidemic discussions, and providing encouragement and advice. They attend small-scale online gatherings and are permanently active in various local topics on Weibo (see
Figure 2A).
Cluster 2 contains Shanghai netizens who participate in discussions temporarily. As seen in
Table 5, of the seven groups, the Cluster 2 group has relatively few followers (9811.44) and posts (1782.62). Furthermore, it has the highest mean of closeness centrality (0.81), the lowest mean of in-degree (0.81), and relatively low means of out-degree (1.35) and betweenness centrality (1673.88). Before the pandemic began, these netizens would use Weibo infrequently, but due to the health crisis, they had become temporarily involved in Weibo topics about asking for help. Their online behaviors focus on pointing out problems, asking for information, seeking help, and providing support. This group is not keen on online discussion, and instead are more interested in following the issues they care about. They rarely attend online gatherings, and go offline once their goals are reached (see
Figure 2B).
Cluster 3 comprises Shanghai netizens who are eager to participate in online relief. This group is more active on Weibo, though their number of followers (90,439.02) and posts (1782.62) are roughly equivalent to the average across all seven groups, their values for means of in-degree (62.10) and out-degree (15.58) are high. The Cluster 3 group has relatively high betweenness centrality (2,093,072.22) and closeness centrality (0.19), as seen in
Table 5. Their online behavior primarily involves posting help-seeking information on Weibo and obtaining help and advice from other enthusiastic netizens. At the same time, they would also share their own experiences online to provide advice and support to netizens in similar situations. On Weibo, they form tight-knit, interactive groups around help-seeking topics (see
Figure 2C).
Cluster 4 contains the volunteer organization accounts which emerged during the pandemic. As shown in
Table 5, this group has high means of in-degree (25.33) and out-degree (89.24), as well as a high mean of betweenness centrality (8,393,435.28), although its number of followers (301.86) and posts (3435.19) are low in comparison to those of the other groups. Their online behavior is mainly reflected in actively searching for help-seekers on Weibo and using their own resources to provide them with help and support. For example, the Marine Compass Volunteer Team, one Cluster 4 account, had assigned staff in the Weibo super-topic square who were tasked to retrieve help-seeking information, match needs, connect channels, and recruit offline helpers. This type of node is displayed in the network graph in a radical shape (see
Figure 2D).
Cluster 5 consists of Internet influencers and social media accounts of some renown. They possess a significant numbers of followers (9,250,003.31) and the highest number of posts (137,703.01) among the seven groups (see
Table 5). However, their mean values of in-degree (7.52), out-degree (0.73), and closeness centrality (0.09) are low because they seldom respond to others’ mentions and comments. Ordinary users tend to ask them for help because of their online influence (see
Figure 2E).
Cluster 6 constitutes famous media figures and influencers who are well-known at home and abroad. As shown in
Table 5, this group has the highest number of followers (88,607,499.24) and a significant number of posts (61,781.43) of all seven groups. It has a relatively high mean of in-degree (62.86) but low means of out-degree (0.05), betweenness (0.00), and closeness centrality (0.00). Such accounts are used primarily to release information to the public, and rarely respond to users’ comments or mentions during the pandemic. These accounts have strong influences both online and offline, including organizations such as CCTV News and People’s Daily. During the pandemic, netizens tended to share information with such accounts and turned to their channels to highlight topics and try to draw the attention of the relevant aid departments (see
Figure 2F).
Cluster 7 is composed of ordinary Internet users participating in online discussions. As shown in
Table 5, they have a small number of followers (40,150.65) and posts (1820.83), but are very active in fueling discussions around online rescue topics. The Cluster 7 group has relatively low mean values of in-degree (1.07), out-degree (1.50), betweenness centrality (21,851.43), and closeness centrality (0.13). Their behavior resembles that of Cluster 2, as these netizens all participate in discussions out of their concern regarding the situation (see
Figure 2G).
Outlier 1 is the Shanghai Post, the official Weibo account of the Information Office of Shanghai Municipality. With more than 9 million followers (mean of followers = 9,755,035.00), it enjoys national influence and is a critical window for the Shanghai Municipal Government to publicize as well as describe Shanghai’s economic and social development situations. It has the highest mean of in-degree (2043.00) and a relatively high number of posts (90,259.00), but the lowest means of out-degree (0.00), betweenness centrality (0.00), and closeness centrality (0.00). During the epidemic, many people sent information to Shanghai Post (using the @ function of Weibo) to express their demands, expecting their suggestions to be heard by the Shanghai Municipal Government (see
Figure 2H).
Outlier 2 is the Weibo super-topic community, a chat forum developed by Weibo. As shown in
Table 5, it has the highest number of followers (223,710,039.00) and the second-highest mean of betweenness centrality (21089535.53). However, it has relatively low values in the other indicators (i.e., mean of in-degree = 11.00, mean of out-degree = 423.00, and mean of closeness centrality = 0.20). During the lockdown period, to better gauge user opinions and respond to the demands of the masses, the Sina company issued a notice to users concerned about specific issues, announcing the launch of a “super-topic community” via a special account. Sina had set up an online community called “Shanghai Epidemic Help” to facilitate public discussion and, through a bot account, send targeted messages to accounts whose hashtags contained the keywords “Shanghai Epidemic Help” to invite them to participate in the discussion. The emergence of the bot account greatly improved the efficiency of information transmission (see
Figure 2I).
Outlier 3 is Gangbiyangzi, a well-known local Shanghai blogger who has been active on the Internet during the lockdown period. As seen in
Table 5, he enjoys the highest means of betweenness centrality (66,988,555.71), in-degree (315.00), and closeness centrality (0.22), and the second highest out-degree (158.00). He focuses on local affairs and holds some influence in Shanghai. During the pandemic, the account actively participated in discussions regarding epidemic prevention and mutual assistance among residents, continuously relaying information on patient assistance to the relevant authorities via his own account so that the voice of the people would be heard and responded to (see
Figure 2J).
4.1.2. The Structures of Online Self-Organized Communities
Community is one of the essential components of online self-organization. Community clustering aims to explore the patterns and structural characteristics of online self-organization. We explored the seven possible communities through the statistical inference in
Section 3.1 and obtained 18 indicators for each community, which included the node number, the number of nodes from different self-organized groups, the community graph density, the community clustering coefficient, the average path length of the community, and the number of node behavior types in each community. Next, we used the 18 indicators to conduct K-means clustering for each possible community. The node number of each community represents the size of that community, and the number of nodes from different self-organized groups calculated the number of nodes coming from each of the seven groups as defined in
Section 4.1.1. The community graph density refers to the degree of connectivity between nodes within a community. The community clustering coefficient represents the degree of interconnection between neighborhood nodes, and the larger the clustering coefficient, the more obvious the small group phenomenon is in the community. The average path length represents the shortest distance between each node and its farthest node in the community. The number of node behavior types in each community was calculated using the number of node behaviors related to offering help or advice, encouragement and support, other, or no response. The K parameter of community clustering was determined using the elbow method. In the end, the rescue self-organized community was divided into six patterns. The results are shown in
Table 6.
Based on the clustering results with consideration to the actual current situation, the online rescue self-organized communities were divided into five patterns and one outlier. This section will analyze the characteristics and functions of each of these patterns (see
Figure 3).
Cluster 1 refers to medium-sized communities with close contacts. As shown in
Table 6, all mean values of members, number of nodes from different groups, and number of different behaviors within this pattern are in the middle range among those of all clusters. The mean values of the community clustering coefficients are the largest among all the clusters, indicating that the communities have a noticeable effect in terms of small groups and frequent interaction between nodes. In examining the information posted by these community members, we found that this community tends to discuss particular epidemic-related issues, such as the situation of the mobile cabin hospital, supply distribution channels, and the quarantine site environment. Most of these community members are stakeholders in the target topic, and are keen to discuss it to exchange information, to better understand the situation, and to find emotional release. Most of these participants are ordinary netizens who are very much affected by the pandemic, who engage in the discussions temporarily. The community topics develop continuously, and constantly expanding thanks to new information shared by a few community activists, and new rounds of discussion are continuously raised. The characteristics graph of this pattern is shown in
Figure 3A.
Cluster 2 represents the sparse or small communities with loose connections. As seen in
Table 6, their membership is small and consists mainly of ordinary Internet users. The community in this pattern have infrequent internal discussions and limited influence online. Information disseminated by community members mainly involves online help, situation reports, and advertising. Community members are not motivated to exchange views, will often disappear from the Internet after posting information or achieving their desired purpose, and rarely participate in follow-up questions or discussions. Due to the lack of participation or engagement, this community’s size is the smallest among the clusters. The characteristics graph of this community is shown in
Figure 3B.
Cluster 3 consists of large-scale and active communities, with the highest number of members across the five patterns. As shown in
Table 6, this pattern includes a number of activists, such as members from Cluster 1 and Cluster 3 as defined in
Section 4.1.1, who effectively fuel information dissemination and discussion within the community, leading to the increased community size. A review of the information released by the community revealed that the community concentrates on timely issues that need to be addressed, such as living conditions in quarantine sites, community services, procurement of supplies, and mobile cabin hospital environments. There is a phenomenon of empathy in the community, where its members tend to gripe and complain about a particular issue, which then leads to more discussion and, ultimately, an emotional consensus within the community. The characteristics of this community are shown in
Figure 3C.
Cluster 4 consists of the communities in urgent need of assistance. Communities in this pattern organize themselves primarily around emergency services, with topics of posts including emergency surgery or positive nucleic acid tests. Members of this community flooded to the accounts of Cluster 1, Cluster 5, and Cluster 6 as defined in
Section 4.1.1 with messages sent via the Weibo @ function to draw the attention of celebrities and key media so that problems might be fixed quickly (see
Table 6). These community members disseminate information with a strong purpose, wanting help-seekers to know they are cared for and to help them pull through their difficulties quickly. As a result, there is less communication within the community, and the information transmission is manifested as suggestions provided directly or help to deliver information to important network nodes, which explains the higher ratio of the mean value of non-responsive nodes to the mean value of community members. The characteristics of this community network are plotted in
Figure 3D.
Cluster 5 comprises communities with volunteers acting as intermediaries. As seen in
Table 6, the mean values of members, number of nodes from different groups, number of different behaviors, and clustering coefficient are all under the midpoint. The main characteristic of this community is that it is involved with networking between members through volunteers. For example, the accounts of Shanghai Defender, a volunteer team, posted to seek out netizens in need through several of Weibo’s functions (i.e., post search, super-topics recommendations), as well as leaving messages in the help-seekers’ comment areas to provide them with assistance and contact information. The community in this pattern is not limited to a specific topic or post, but rather spans a wide range of information topics regarding assistance. As long as netizens need help, a community may be formed in this pattern through the center node of a volunteer account. The characteristics graph of this community is shown in
Figure 3E.
Outlier 1 represents the self-radiating community. There are several self-radiating netizens who are both help-seekers and information sharers. A community grew out of their signals for help, and continues to expand through their constant updates. In the process, many zealous users join in to offer them suggestions or encouragement, which in turn incentivizes the original poster to share more information. At the same time, their experience in seeking assistance sets an example for others in similar predicaments. Hence, the repeated interactions between help-seekers and providers creates multiple information cascades, significantly enhancing the community’s activity. The characteristics of this community network are shown in
Figure 3F.
4.1.3. The Evolution of Online Self-Organizations
Analyzing the evolution of online self-organization based on the patterns of self-organized groups and communities can amplify the microscopic process of self-organized rescue during public health emergencies. The cumulative trend graph of the daily new asymptomatic cases in Shanghai during the defined period shows that the shape of the cumulative growth curve of asymptomatic infections conforms to the logistic growth model. Therefore, according to its characteristics, the process of online self-organization evolution can be divided into four stages. We calculated the distribution of the different patterns of online self-organized groups and communities at different stages and mapped the evolutionary network. The results are shown in
Figure 4,
Figure 5 and
Figure 6.
As seen in
Figure 4 and
Figure 5, at each stage of the pandemic, the self-organized relief groups were composed primarily of ordinary netizens and ad hoc participants, and the community pattern was also based on the sparse and small communities formed by them. Specifically, the first stage (1 March 2022 to 31 March 2022) was the initial phase of lockdown when there were relatively few asymptomatic infections. The proportion of medium- and large-scale communities was relatively small, and the network of online self-organization was scattered, as shown in
Figure 6A. In the second stage (1 April 2022 to 15 April 2022), the number of asymptomatic infections exploded, and the obstacles caused by the epidemic and lockdown were fully revealed. To solve practical problems in real life, Internet users actively participating in self-organized rescue began to increase significantly, and volunteer groups emerged. Common issues such as panic buying of goods, supply distribution problems, and effects of centralized isolation began to arouse wide-spread attention and discussion, leading to an increased proportion of medium- and large-scale communities. At this stage, the densest online self-organization is reflected through active online activities, as shown in
Figure 6B. The third stage (16 April 2022 to 30 April 2022) witnessed the resolution of the problems that emerged in the previous stage, and volunteer groups became skilled in their rescue activities. When comparing the second and third stages, the proportion of self-organized groups did not alter significantly.
Figure 6C shows a decline in the proportion of large-scale communities regarding certain specific issues as well as of communities in urgent need of assistance, as well as a slow decline in the density of the overall network. In the fourth stage (1 May 2022 to 31 May 2022), with the pandemic now under control, the number of self-organized rescue groups decreased significantly, and medium- and large-scale communities established to address some of the previously common problems declined significantly. Network activity also decreased, resulting in sparser network density, as shown in
Figure 6D.