A large-scale analysis of COVID-19 tweets in the Arab region

Abstract

The COVID-19 virus has spread rapidly to the Arab World, affecting the public health and economy. As a result, people started communicating about the pandemic through social media such as Twitter. This paper utilizes text mining to extract useful insights into people's perceptions and reactions to the pandemic. First, we identified 11 general topics under which COVID-19 tweets emerging from the Arab region fall. Next, we generated training data consisting of English, multidialectal Arabic, and French tweets that were manually classified into one or more of the identified 11 topics via crowdsourcing. These training data were then used to train various deep learning models to automatically classify a tweet into one or more of the 11 topics. Our best performing models were then used to perform a large-scale analysis of COVID-19 tweets emerging from the Arab region and spanning a period of over one year. Our analysis indicates that the majority of the tweets analyzed emerged from Saudi Arabia, UAE, and Egypt and that the majority of the tweets were generated by males. We also observed a surge in tweeting about all the topics as the pandemic broke followed by a slow and steady decline over the following months. We finally performed sentiment analysis on the analyzed tweets, which indicated a strong negative sentiment until mid of September 2020, after which we observed a strong positive sentiment that coincided with the surge in tweeting about vaccines.

We chased COVID-19; Did we forget Measles? - A machine learning approach to public discourse and sentiment analysis on spiking Measles cases

“…The second limitation is that Reddit may not be considered a representative of the general public, as only a limited number of users may share their views on such a platform. However, there are already studies reported in the literature on using Reddit as a platform for public discourse and emotion analysis [36]. Further, our analysis shows that the hate groups shout the loudest and most frequently during such disease outbreaks or pandemics, which may bias the output or findings of this study.…”

Applications of Social Media and Digital Technologies in COVID-19 Vaccination: Scoping Review

“…These sentiment analysis studies showed that public sentiments were associated with real-time news, internet information, public health events, the number of COVID-19 cases, vaccine development, the pandemic, and announcements of political leaders or authorities [ 7 , 21 , 27 , 31 , 51 , 52 , 55 , 56 , 58 , 65 ]. Although public sentiments on COVID-19 vaccines varied significantly over time and geography [ 7 , 33 , 39 ], positive sentiments were more prevalent than negative ones regarding COVID-19 vaccines [ 7 , 19 , 20 , 22 , 25 , 29 , 33 , 41 - 44 , 46 , 47 , 51 , 57 , 60 , 62 , 65 , 70 , 72 , 76 , 78 , 84 , 85 ], with trust and anticipation being the predominant emotions [ 20 , 23 , 32 , 37 , 50 , 54 , 58 ]. However, some other studies found that negative sentiments overwhelmed positive ones, with fear being the dominant emotion [ 53 , 59 , 64 , 66 , 71 , 73 , 79 - 82 , 86 ].…”

“…Fifty-three studies applied social media for topic analysis on COVID-19 vaccination. Topic analysis methods included latent Dirichlet allocation (LDA) topic modeling (n=24) [ 20 , 23 , 25 , 31 , 34 , 35 , 40 , 47 , 50 , 51 , 54 - 56 , 58 , 60 , 61 , 64 , 87 - 93 ], manual coding (n=17) [ 57 , 80 , 94 - 108 ], and other algorithms (n=12) [ 26 , 43 , 62 , 109 - 117 ]. Table 2 summarizes the provaccine and antivaccine topics on COVID-19 vaccines present on social media.…”

“…Nineteen sentiment analysis studies were based on machine learning, and they trained machine learning classifiers by annotating tweets in the data set [59][60][61][62][63][64][65][66][67][68][69][70][71][72][73][74][75][76][77]. Machine learning approaches used supervised classification algorithms to extract information regarding sentiment polarity.…”

“…Fifty-three studies applied social media for topic analysis on COVID-19 vaccination. Topic analysis methods included latent Dirichlet allocation (LDA) topic modeling (n=24) [20,23,25,31,34,35,40,47,50,51,[54][55][56]58,60,61,64,[87][88][89][90][91][92][93], manual coding (n=17) [57,80,[94][95][96][97][98][99][100][101][102][103][104][105][106][107][108], and other algorithms (n=12) [26,43,62,[109][110][111][112][113][114]…”