Presentation "COVID-19 Vaccine Brand Sentiment on Twitter" - OASIS 2022 Workshop

good afternoon welcome to our presentation titled covid19 vaccine brands sentiment on twitter my name is alina campan and i am presenting this work with one of my co-authors trian mario struta we are both in the school of computing and analytics at northern kentucky university this is the outline of our talk due to the time limitation we will provide only a general overview and we recommend that you read the entire paper online social networks osns are today a primary way to spread and consume information an important aspect of osns is that they are open which means that users can post anything of course within certain boundaries established by the provider this will lead to a proliferation of information with different degrees of truthfulness with intentional or unintentional disinformation purposes twitter linkedin and facebook are some of the most used uh social networks for our work we used um data collected from twitter since twitter allows via provide apis to collect in real time tweets that are a match for a set of given keywords vaccination hesitancy is not a new phenomenon by any means it manifested even before 2020 it is contributed by active groups that spread misinformation and or disinformation for example by exaggerating the number of occurrences of side effects of the existing vaccines um of course vaccination hesitancy remained a significant issue in stopping the spread of the kovi 19 pentagon existing feed algorithms on osn's choose relevant opinions to display to the users based on popularity these popular opinions are usually extreme and by being shared they are uh further emphasized uh and this uh in the end will be leading to more division of opinions so users uh do not really see uh or get a real feel of what's going on in the osn they will have a biased view of the information our goal in this work was to analyze the spreading of information in twitter volume wise and sentiment wise positive or negative for covid 19 vaccines overall and for some specific ov 19 vaccine brands so our contributions include in this work we collected twits with vaccination and covet 19 related keywords for four periods of 10 days each from winter to fall 2021. we analyzed the popularity based on the count and public sentiment of covet 19 19 vaccination tweets overall and with focus on individual individual vaccine brands to our knowledge this is the first study that makes this comparison between existing covid19 vaccine brands we also identified the most popular individual tweets during the analyze time windows and we analyzed these tweets with respect to the overall retweets count so how much they were respreading to the network retweets count over time so a timeline timeline of the tweets and sentiment polarity this figure shows our workflow from that data collection to summarization and sentiment analysis all the work was implemented in python and we used among others the tweepy numpy pendants and re libraries i just want to add a couple notes regarding the specifics of collection and analysis with respect to twit collection um we leveraged twitter free streaming api with uh filtering and the twippy library to collect the tweets from the twitter live stream um uh for our uh fourth day 10 day time windows uh or periods ranging from february to october of 2021. regarding the twitter sentiment analysis or tsa we used vader which the documentation um of the product advertises as being it being a lexicon and rule-based sentiment analysis tool that is specifically attuned to sentiments expressed in social media hello everybody as alina introduced me i am mario strutze and i will be presenting the results part of this paper in these figures we showed the volume of tweets for all four 10-day windows we noticed that the total tweet count was higher during the first two ten day periods we investigated and then dropped by approximately 30 percent in the last two periods there is a correlation between the number of tweets we collected and the total number of copy 19 case reported worldwide in particular for the last three periods during the first 10 day period the number of tweets was relatively high compared to the number of cases however this could be explained by the novelty of the vaccines and the fact that most developed countries were still in the early stages of their vaccination campaign each period has significant peaks in terms of magnitude there are just five peaks that reach more than 7.5 000 tweets in a 30 minutes period and those are shown with green we identified possible explanations for each of these speaks for example in the first peak label cp1 this is due to the tweet of the prime minister of india narendra modi in which he announces that he took the first dose of the kobi19 vaccine the other peaks except for one were produced in proportion of 35 or more by retweets of similar messages by user with high follower count of over 1 million followers the exception is peak cp3 for which the dominant tweet originates from a user with a relatively small number of followers 20.6 thousands in addition the relevant the retweet count for the most repeated tweet is only 16 so it looks like most volume peaks are contributed by messages from users with high follower counts in these figures we showed the average sentiment for all four 10-day windows the overall sentiment is more positive on average in the first analyzed period than in the next three one possible explanation is that during that time february 27 march 8 2021 vaccines were still in short supply but becoming more and more available there were many positive messages that advertised the increased availability of vaccines and people were hopeful that vaccines would have a decisive impact on the course of the pandemic for all other periods the vaccine availability was universal in developed countries that have a large twitter presence and focus shifted to vaccine hesitancy and convincing people to vaccinate it is worth noting that all 10 days from the first period had the average daily tweet sentiment positive while this number dropped to to six for the second period five for the third and fourth for the last 10 day period all remaining days had a neutral sentiment polarity on average at a more micro level the number of 30 minutes windows with negative sentiment also increased from only 5 in the first period to over 30 in the last period we noticed that the interest for individual copy 19 vaccine brands varied significantly between the four periods the pfizer vaccine is as reflected by the tweet volume more popular than each of the other vaccine in all of surf periods except the first more significantly the percentage of tweets that match fisher keywords out of all tweets that contain any of the eight vaccine brand keywords per time period is generally growing 26 46 35 and 63. the popularity of astrazeneca vaccine follows an opposite trend to pfizers while during the first period it had more mentions than any other brand its importance decreased later and in the last three periods it was only the third most popular vaccine brand modena is the second or third most popular vaccine throughout the time with the exception of the third period when sinovac reached its peak and took this spot three vaccines cinovax putnik and kovacsin has similar popularity trends with alternating popularity count in the last period their popularity has declined significantly somewhat surprising since the johnson johnson vaccine was approved in usa this brand has a very low number of tweets throughout the entire observed time and its popularity was decreasing continuously lastly the vaccine from novos also had low popularity during all four periods the sentiment of brand specific tweets is harder to interpret there are wide variations between periods and it is likely that the overall sentiment was influenced by tweets with relatively high count in that corresponding period in these figures we showed the popularity trend of top 10 the most frequent tweets the positive tweets are illustrated with green the negative tweets are shown in red and neutral tweets are in blue we notice that the top twist popularity often peaks fast after the tweet is created and then it decreases at the slower rate until it dies down this kind of tweet timeline features an initial maximum volume peak very soon after the tweet's creation time within the first hour or so and is common in the top 10 tweets shown in the figure half of the 20 tweets have this profile this profile is also usually associated with the author of the tweet being a user with many followers by visual inspection of the tweet timelines we see that most of tweets in this group 7 out of 10 followers numbers highlighted in yellow were altered by a user with more than 1 million followers by fewer of those tweets only 3 out of 10 followers number highlighted in green were authored by a user with less than 1 million followers there are however twits author by user will not follow account that become very popular such tweets usually reach their peak a few hours after creation usually when repeated by more influential with more followers individuals in addition the popular tweets authored by less popular users seem to keep their popularity longer and in some cases more than a day here are some conclusions the volume of all convenient vaccine-related tweets decreased from the first half of 2021 into the second half of the year the overall sentiment polarity was more positive in the first time period and decreased over time the volume of brand specific tweets is also decreasing pfizer is the most talked about covenanting vaccine brand on twitter and it was increasing its percentage of overall brand matching tweets other than nikka and justin and johnson percentages decreased at the same time moderna while decreasing in the overall count kept its popularity within brand related tweets no differences in terms of trending patterns between tweets with positive and negative polarity and the trend is influenced by the popularity number of followers of the tweet author and of individuals that retweet that tweet this concludes our presentation we hope you enjoy it and please ask us any questions you might have about this work enjoy the workshop