[ad_1]
Discussion
Our study examined the role of messaging shared by PIPE in determining general public discourse direction and the nature of sentiment shared via social media networks. Drawing on messaging shared by our three subgroups of influential users, findings suggest the presence of consistent patterns of emotional content shared by PIPE for the first 2 years of the COVID-19 pandemic influenced public opinion and largely stimulated online public discourse.
When comparing the overall mean score between negative and positive comments for individuals in the athletes/entertainers’ subgroup, it can be concluded that related sentiment holds a more negative tone than positive. For this subgroup, the highest number of negative comments were associated with AR and NM, with a combined mean of 0.95. Regarding EC, it is interesting that he had very few positive tweets, the least in the group and the highest score difference. This fact implies that EC has been largely criticised by the public and was the least favourite figure of the subgroup. As evidence of this argument, the tweet with the highest like count mentioning EC states, ‘Strongly disagree with [EC]…take on Covid and the vaccine and disgusted by his previous white supremacist comments. But if you reference the death of his son to criticize him, you are an ignorant scumbag’.
The theme of overarching negative sentiment continues when examining findings within the politicians’ subgroup. DT and TeC were found to have the most substantial impact within the subgroup, with a large number of combined likes totaling more than 122 000, with a monthly-based mean of 0.33. A substantial portion of tweets related to this subgroup was aimed at questioning whether politicians hold sufficient expert public health knowledge to advise constituents on medical decision making. This is reflected in the most liked tweet mentioning TeC, a user sharing, ‘I called Ted Cruz’s office asking to make an appointment to talk with the Senator about my blood pressure. They told me that the Senator was not qualified to give medical advice and that I should call my doctor. So I asked them to stop advising about vaccines’. It became increasingly clear that this subgroup was the most tumultuous. The spread, reaction and engagement by the public to posts made by politicians online was indicative of a strong level of influence, suggesting politicians play key roles in ensuring population health and should be committed to promoting health-protective behaviours rather than sensational falsehoods.
As mentioned reviously, though sentiment related to the subgroup of news media personalities was shown to be oscillatory throughout the pandemic, emotion shared was overall more negative than positive. Moreover, tweets referencing this group were typically more related to antivaccine controversy or death (ie, death of PV) rather than news about vaccine development. Though the sentiment for the subgroup was overwhelmingly negative on average, a more interesting story begins to appear with the inspection of tweet content. For example, the most liked tweet associated with JR was ‘I love how the same people who don’t want us to listen to Joe Rogan, Aaron Rodgers about the covid vaccine, want us to listen to Big Bird & Elmo’, clearly a vaccine-hesitant or antivaccine statement. Notably, it is interesting to compare the number of tweets with the total number of maximum likes. This combined set of news media personalities had a total of 14 017 with 93 974 associated highest like count. The high number of likes displayed within these tweets shows that a much higher number of users are involved in reading tweets and are therefore potentially influenced by the content.
Public health implications
We argue the application of our findings could have meaningful impacts on the public health sector, bolstering currently available surveillance tools for precision health promotion,24 management of the ongoing COVID-19 pandemic and preparing for the next crisis. As we have demonstrated, messaging shared by influential members of society can have considerable effects on the directionality of public emotion and shared health decision making. Both negative and positive online social endorsement of prevention strategies such as vaccination are key in determining population-wide compliance and uptake success. However, threats of the spread of misinformation and disinformation by those with influence stand to undermine programmes supporting protective measures such as vaccination. As misinformation poses a range of psychological and psychosocial risks (anxiety, fear, etc), public health institutions hold some responsibility for the continued development and sustainability of low human effort surveillance systems optimised for generating responses to waves of falsities shared via social media platforms. Discernment of the dynamic levels of population sentiment shared via social media would allow public health officials to design catered mitigation and communication strategies. Social campaigns aimed at directing users with COVID-19 related inquiries to high-quality sources such as the CDC and other trusted public health institutions for evidence-based recommendations and instruction. Furthermore, public health and research institutions could be more proactive in creating collaborations with PIPE to share more positive messaging regarding vaccination. Moreover, there is room for intelligent algorithmic systems that identify patterns and anomalies in shared messaging, tasked with boosting messaging from social influencers who stand in affirmation of COVID-19 mitigation strategies; however, there is a need for a broader understanding of the potential negative implications, including ethical and legal issues.23
Limitations
It is important to note some study limitations should be considered in unison with our findings. Sentiment analysis of social media shared messaging has long been challenging due to the ambiguity of natural language. Though language models such as BERT help to mitigate many of these challenges, the difficulty lies in the true detection of sarcasm, humour and complex inferences. As such, models currently available are unable to distinguish sentiment expressed towards different targets. As such, sentiment cannot be individually mined for two separate topics if shared within a short tweet. For example, if a user were to positively mention the name of a politician in support of COVID-19 vaccination while simultaneously sharing negative emotions toward communities in opposition, current language models may label the overall sentiment as negative. Though the negative emotion was not targeted to the mentioned politician, the model was limited in differentiating sentiment within the messaging. Moreover, though messaging shared by suspected bots, highly repetitive news media, highly repetitive high frequency users or duplicates were removed, it is possible a small amount could have still slipped through the data cleaning process.
[ad_2]
Source link
