Basic statistical analysis
Similar to word frequency, which usually conforms to Zipf’s law or Pareto’s law [53, 54], there is also a large heterogeneity in the popularity of different emojis; i.e., the top 100 popular emojis account for approximately 96% of all emojis used daily, and the top 10 popular emojis account for approximately 50.4% (Fig. S3). Among the 1,840 unique emojis, we determined the top 100 most frequently used emojis anytime during the collection period as popular emojis, which yielded 154 unique emojis overall. In determining the daily top 10 popular emojis on Weibo before and after the COVID-19 outbreak (Fig. 3), we found that after the pandemic, most of the newly added popular emojis were related to negative emotions, such as sadness, anger, and disappointment. A few positive emojis, such as Cheer for Wuhan ( ) and Peace Lantern ( ), were also posted frequently, expressing support for anti-epidemic work and encouragement and cheering for medical staff. In addition, due to the shortage of living materials in quarantined communities, the number of food-related emojis increased significantly, such as Chicken Leg.
According to the emotion expressed by the emoji, we classified the 154 popular emojis into 5 categories, i.e., Happy, Encouraging, Questioned, Sad, and Angry (Table S1). By analysing the evolution in the usage of different emoji groups over time, we found that after January 18, 2020, the number of emojis in the category Happy decreased rapidly, while the number of emojis relevant to Sad increased. Furthermore, the daily proportion of Happy emojis declined significantly and remained low (approximately 0.32) for nearly 29 days after the COVID-19 outbreak, while the daily proportion of emojis in the Sad and Encouraging categories rose to 0.24 and 0.34, respectively (Fig. 4). On January 31, 2020, Li Wenliang , an ophthalmologist at Wuhan Central Hospital, posted his experience of being summoned by the police for "making false comments on the internet about unconfirmed SARS outbreaks" on Weibo. At this time, the use of anger-related emojis reached its peak. These results indicate that the COVID-19 epidemic significantly impacted individuals’ moods, and accordingly, people tended to express more sadness and encouragement after the COVID-19 outbreak.
Correlation of emoji usage with COVID-19
We found that the average number of emojis used in each post during the period was negatively correlated with the proportion of pneumonia-related posts (r = -0.698, p < 0.001); i.e., the more pneumonia-related posts were published, the fewer emojis were utilized. This suggests that Weibo users were likely to post fewer emojis when mentioning COVID-19-related topics. Moreover, the rate of emojis relevant to Happy was negatively correlated with the ratio of pneumonia-related posts (r = -0.808, p < 0.001 \(r=-0.808, p<0.001\). In contrast, the proportion of Sad emojis and the rate of posts mentioning pneumonia showed a significant positive correlation (r = 0.813, p < 0.001 \(r=0.813, p<0.001\). Surprisingly, the proportion of Encouraging emojis was also positively correlated with the proportion of pneumonia-related posts (r = 0.604, p < 0.001 \(r=0.604, p<0.001\)), indicating that although the COVID-19 outbreak negatively affected individuals’ moods, the positive emotions regarding encouragement and cheering also increased, which to some extent demonstrated the confidence and support of Wuhan residents for anti-epidemic work.
Temporal pattern of emoji usage
In the temporal analysis of emoji usage, we first divided the study time into 4 periods in accordance with the development of COVID-19. The first period (P1) covers the time from December 01, 2019, to January 01, 2020; the second period (P2) is from January 01, 2020, to January 18, 2020; the third period (P3) is from January 18, 2020, to February 18, 2020; and the fourth period (P4) is from February 18, 2020, to March 20, 2020.
We found that the hourly changes in the number of emojis used during the whole period were consistent with the number of posts; i.e., the number of emojis and posts both peaked at 10:00 pm and were minimal at approximately 5:00 am. However, the hourly emoji use rate (average number of emojis used in each post) after the COVID-19 outbreak (P3 and P4) was clearly different from the pattern before the outbreak (P1 and P2), as shown in Fig. 5A. The peak time of the emoji use rate before the pandemic was at midnight or in the early morning, when the rate during P3 and P4 was at the lowest. In addition, users became more negative at midnight (approximately 3:00 am); i.e., the proportion of emojis related to happiness and encouragement decreased significantly, while emojis in the Sad and Angry categories increased dramatically (Fig. 5B). This proves that COVID-19 negatively impacted individuals’ moods. Furthermore, online users were more psychologically vulnerable at midnight, often communicating the pessimistic part of their emotions.
On weekdays, the number of emojis used in each post was the highest (emoji use rate) during the evening (approximately 10:00 pm) and early morning (4:00–6:00 am). However, the peak period of the emoji use rate on weekends was not obvious, and the emoji use pattern daily showed a more even distribution (Fig. 5C). In addition, the diversity of emojis used on weekends was much higher than that on weekdays, even if there were not more emojis used. On weekends, the proportion of Encouraging and Happy emojis used was significantly higher in the morning (approximately 9:00 am) than at other times (Fig. 5D), indicating that people tended to be more positive and optimistic in the morning on weekends. However, whether on weekdays or weekends, the use of emojis related to negative emotions (Sad and Angry) reached its peak at midnight. This also proves that people were more likely to reveal their psychologically vulnerable and negative side late at night. We found that the difference in emoji usage between weekdays and weekends varied significantly with the development of COVID-19. During P1 and P2, there was a clear peak in the emoji use rate in the early morning (approximately 4:00 am) on working days, while after the COVID-19 outbreak, the daily trends on weekdays and weekends both remained relatively stable throughout the day (Fig. S6). This could be attributed to the stay-home quarantine for controlling the spread of COVID-19; afterwards, residents had a more even time distribution.
Emoji co-occurrence network
To understand the collocation pattern among emojis, we constructed a co-occurrence network with nodes standing for the unique emojis and the weight of edges being the number of times the two emojis have been used together. The network contains 1,711 nodes and 31,878 edges, with an average degree of 37.3; i.e., each emoji is used, on average, with approximately 37 other emojis together in posts. is the most connected node, with a degree of 591, while there are more than 45% nodes with a degree less than 10 and approximately 87% nodes with a degree less than 100. We found that the average weighted degree of the network is 298.4, indicating that, on average, each emoji appears with other emojis approximately 298 times. The node with the highest weighted degree (12,970) is , a consequence of the great number of Christmas greetings during the data collection period, and the emoji pair (edge) [ ] has the highest co-occurrence frequency in posts (weight of edge, 66,653); however, the weight of approximately 91% of the edges in the network is less than 100, and the weight of more than 74% edges is less than 10. In addition, we found that there was a clear community structure in the co-occurrence network.
Evolution of the emoji co-occurrence network
We extracted the emojis in all posts and constructed the emoji co-occurrence networks before and after the COVID-19 outbreak (Net1 and Net2, for short), as shown in Fig. 6A and B. We found that the number of nodes and edges in the network greatly decreased after the COVID-19 outbreak. This is consistent with the above findings, indicating that due to the epidemic, users reduced the use frequency of emojis. However, the average degree of this network increased from 34 to 39.8, which means that although the number of emojis used decreased, the diversity of emoji usage increased. After the COVID-19 outbreak, the average clustering coefficient of the emoji co-occurrence network also increased from 0.23 to 0.32, indicating that more triangular structures appeared in the emoji connections. The top 3 nodes with the largest weighted degree in Net2 are (Table 1), related to emotions about grievance, sadness and doubt, which are significantly different from the most connected nodes in Net1. The emoji pairs with the highest co-occurrence frequency after the pandemic also changed considerably, and the use frequency of emojis relevant to sadness (e.g., ) increased. By comparing the community structure of Net1 and Net2, we also described the flow of members (nodes) in the TOP-10 communities before and after the pandemic (Fig. 7).
Maximum co-occurrence frequency network
To discover the typical emoji collocation on Weibo, we built the maximum co-occurrence frequency network by retaining the node pair (edge) with the largest co-occurrence probability. The weight of the edge represents the co-occurrence probability between nodes (from 0 to 1). There were mainly 4 hub nodes ( ) in the network (Fig. 8A) during P1 and P2. After the COVID-19 outbreak, Flower () replaced Longing ( ) as a new hub node, and more discrete emoji pairs, as well as small subclusters, appeared, e.g., the emoji pairs connected by Angry () and a series of emojis about gestures (Fig. 8B). Emojis and emoji pairs related to sadness, gratitude, and encouragement were used more frequently. In addition, the cluster connected by Love ( ) decreased, but the size of the community linked by Facepalm ( ) grew significantly, and there were more emojis relevant to negative emotions, such as Dizzy ( ), Scared ( ), and Wearing Mask ( ). These results indicate that users’ habits regarding emoji usage indeed changed after the COVID-19 outbreak. Due to the emergence of newly popular emojis and the evolution of the moods users wanted to express, the fixed emoji co-occurrence patterns were significantly altered.
Topic, user level and gender
Correlation between emoji usage and post topics
To examine whether the topic affected emoji usage in the post, we extracted all posts containing hashtags and analysed the correlation between different topics and emoji use patterns. All hashtags (##) on Weibo from December 1, 2019, to March 20, 2020, with over 500 users and 500 microblog posts, were manually divided into 4 groups according to the topic, i.e., stars, daily life, games or e-sports, and social events (including COVID-19). These 4 topics covered a total of 3,830,100 posts. It was observed that before the COVID-19 outbreak, most topics discussed online were involved with stars and daily life. After the COVID-19 outbreak, discussions on these topics diminished dramatically, while the number of posts related to public social events rapidly increased (Fig. S10).
We then compared the evolution of the proportion of posts containing emojis on these four topics (Fig. 9A). Users posted emojis most frequently when discussing topics related to stars and entertainment, while the emoji use in posts related to social events was the lowest, suggesting that users tend to be more cautious and objective when discussing public social events. After the COVID-19 outbreak, the proportion of posts containing emojis in topics on social events moderately increased, while the proportion regarding stars, daily life, and games dropped greatly. In addition, we found that most emojis used in these topics were related to Happy and Encouraging, while more Sad emojis were posted in posts related to social events. Using the total number of emojis used daily to standardize the number of emojis in different categories, we found that after the COVID-19 outbreak, users’ emotions showed greater fluctuations than before when discussing public social events, and they expressed more incentives, sadness, and anger (Fig. 9B).
Differences between individual users and official users
There are a total of five user levels on Weibo, i.e., Ordinary User, Popular User, Orange V, Gold V, and Blue V. The first four levels all represent user accounts, while Orange V as well as Gold V are personal authentication accounts, standing for public figures with a certain social influence (e.g., actors). Blue V stands for officially certified enterprise or government accounts. We found that the proportion of posts containing emojis published by official users dropped dramatically after the COVID-19 outbreak, while the proportion of posts published by personal users changed much more moderately (Fig. 9C). This indicates that due to the public health emergency, official users decreased their emoji use frequency, and the texts posted by the government and enterprises tended to be more formal and serious than usual.
Emoji usage by user gender
Comparing the evolution of emoji usage between male and female users, we found that the average number of emojis posted by male users decreased significantly after the COVID-19 outbreak, while that posted by female users slightly increased (Fig. S12). Comparing the proportion of posts containing emojis between these two groups, it is also obvious that after the epidemic, the emoji use frequency of male users when posting online dramatically decreased (Fig. 9D). Male users were more inclined to utilize fewer emotional symbols and more formal language, which to some extent suggests that when facing public risks, men tend to conceal or control their emotional display, while women are more emotionally expressive [56,57,58].