Introduction:
The unprecedented outbreak of Covid-19 in 2020 changed everyone’s life. Public facilities were shut down, such as schools, restaurants and companies. A lot of people were forced to work/study from home, some of them even lost their jobs. This all seemed chaotic and desperate at first glance. However, this is not the first pandemic in human history. By the technology we have so far, it is possible to conduct a social network analysis on people’s reactions and discussions on and during this pandemic. Fortunately, I found an article in which a study addressed this topic by applying machine learning and artificial intelligence on Twitter posts.
Data collection and analysis:
This study was conducted by Man Hung from Roseman University of Health sciences. It drew data from Twitter, which is a social network platform with 166 million daily users. The tweets were collected from the United States, from March 20 to April 19, 2020.
First, the tweets were filtered by keywords related to covid. And they were then applied Latent Dirichlet Allocation (LDA), which is a machine learning algorithm for performing topic modeling. The tweets were categorized into 5 salient themes/topics, namely Health care environment, Emotional support, Business economy, Social change and Psychological stress. Next, a sentiment score calculated by Valence Aware Dictionary and sEntiment Reasoner (VADER) was assigned to each topic. A sentiment score ranges from -1 to 1, where -1 is the most negative, 0 is the neutral and 1 is the most positive.
The above diagram shows the number of tweets related to covid in the 1-month period.
The above diagram shows the distribution of positive, neutral and negative tweets of the 5 topics. We can see that the positive tweets take the largest proportion of each topic.
The above figure indicates the top 10 most frequently used words for each topic. The words are presented as nodes. The bigger the node is, the more frequently the word is used. The edges are the relationships between words. We can also conclude that the more links there are between two nodes, the more the two words are used together under the same context. For example, for social change, the many links between “made”, “morning”, “night”, “today” and “time” suggest that the pandemic changed people’s social living for the entire day. Something worth noting is that emotional support has the greatest closeness which implies that it is the most likely discussed topic between the 5 topics.
This map displays the average sentiment score for each state in the United States. We can see that every state has an average sentiment score that is greater than 0. Therefore, the average sentiment score for the United States is slightly positive.
Conclusion:
All the figures illustrated above reveals that overall, the positive sentiment outweighed the negative sentiment, which implies the fact that people generally hope for the better in such a public health crisis, but it may also be the case that in the early stages, people are underestimating the pandemic.
Despite the rigorous data collected for the study, there are limitations. For example, not everybody uses Twitter, so the result is not fully representative to the United States population. Also, due to the rapidly changing situation, tweets can become outdated very quickly.
But still, Twitter is an inexpensive, fast and reliable data source that can be utilized to perform network data analysis to obtain realistic and down-to-earth emotions and sentiments of people towards the pandemic. And, by referring to the response of the analysis, the authorities and the public can take measures correspondingly.
References:
Hung M, Lauren E, Hon E, Birmingham W, Xu J, Su S, Hon S, Park J, Dang P, Lipsky M
Social Network Analysis of COVID-19 Sentiments: Application of Artificial Intelligence
J Med Internet Res 2020;22(8):e22590
URL: https://www.jmir.org/2020/8/e22590
DOI: 10.2196/22590