**4. Discussion**

The overall shape of the network resembled a community as there were a range of users conversing amongs<sup>t</sup> each other in different clusters. When examining the most frequently used hashtags, it appeared that the most popular hashtags encouraged mask wearing among the public. It was found that a range of accounts were influential and/or mentioned, ranging from ordinary citizens, politicians, and popular culture figures. The discussion had been politicized by some users on Twitter, which led to politicians appearing as influential users within the network. Japanese hashtags also appeared within the network, which highlights the global nature of the discussions around this issue. It is important to note that some of the accounts such as the official Twitter account of YouTube might not have tweeted using the word 'mask', however, users may have shared content which contained the word 'mask' and also used the mention "@YouTube". The most common theme to emerge was the encouragemen<sup>t</sup> for the public to wear masks and discussions around this. Other themes were related to jokes and discussions related to whether face masks were safe for those who may have breathing problems. Our study also examined Twitter data emerging solely from the United States, which demonstrated that there was overlap in content. This could have occurred because the Twitter has the most active users in the United States.

It must be noted that Twitter discussions are constantly evolving and potentially alter on a weekly basis. A limitation of our study is that it focused specifically on the Twitter network from 27 June to the 4 July, hence our findings may not be applicable to other time periods. Future research could seek to expand time periods and examine Twitter discussions based on other locations. A further limitation is that with the 1% random sample we extracted, generic keyword and 7-day approach, we have captured some tweets coming from "temporary" discussions (for instance, beer and mask related humor in Japan). If the study had examined a longer time period, i.e., 6 months, and taken a 0.005% sample, the study would have been able to perform a much wider analysis (the masks and beer humor would not have been captured and/or captured on a lesser scale). At the same time, however, the approach adopted in this study had a better chance of capturing other issues such as conspiracy theories and/or short-lived time-based discussions. Future research could combine both approaches in the analysis. Further research could also seek to conduct a sentiment analysis of the data. A wider limitation related to research on Twitter is the potential of 'o ff-topic' discussions surrounding a particular keyword or hashtag to take place. Future research could seek to eliminate irrelevant content prior to analyzing data. Twitter data can be used to study a wide range of public health topics and was recently used to study views into personal health records [20], which utilises a similar methodology to this present study. Other research has also examined disclosure of patient information [21] and COVID-19 conspiracies [22]. Future research could also seek to examine the role of influential accounts during this time.
