Evolution of Select Epidemiological Modeling and the Rise of Population Sentiment Analysis: A Literature Review and COVID-19 Sentiment Illustration

Abstract

With social networking enabling the expressions of billions of people to be posted online, sentiment analysis and massive computational power enables systematic mining of information about populations including their affective states with respect to epidemiological concerns during a pandemic. Gleaning rationale for behavioral choices, such as vaccine hesitancy, from public commentary expressed through social media channels may provide quantifiable and articulated sources of feedback that are useful for rapidly modifying or refining pandemic spread predictions, health protocols, vaccination offerings, and policy approaches. Additional potential gains of sentiment analysis may include lessening of vaccine hesitancy, reduction in civil disobedience, and most importantly, better healthcare outcomes for individuals and their communities. In this article, we highlight the evolution of select epidemiological models; conduct a critical review of models in terms of the level and depth of modeling of social media, social network factors, and sentiment analysis; and finally, partially illustrate sentiment analysis using COVID-19 Twitter data.

Modeling Behavior and Vaccine Hesitancy Using Twitter-Derived US Population Sentiment during the COVID-19 Pandemic to Predict Daily Vaccination Inoculations

“…In terms of vaccine side effects, the CDC also reports “Selected Adverse Events Reported after COVID-19 Vaccinations” but the reports are not a daily occurrence, have significant latency between events and reports, and do not claim to represent a collection of all adverse events [ 21 ]. As a supplemental measure to account for CDC virus mitigation and vaccine side effect reporting limitations and as demonstrated previously by [ 6 ].…”

“…From June 2011 to April 2019, access to commentary by individuals on social media enabled researchers to study the controversial influence on public sentiment and behavior generated through global social media channels with a particular focus on vaccine hesitancy [ 1 ]. Other researchers in Korea [ 2 ], Turkey [ 3 ], India [ 4 ], and the United States of America [ 5 , 6 ] expanded social media sentiment analysis; in particular, using Twitter tweets as a significant resource of data and analysis to rapidly track and quantified public opinions, beliefs, or behavior regarding critical events, pandemic-related events, personalities, or subjects including quickly and effectively measuring vaccine hesitancy.…”

“…The sudden increase in the number of cases, the severity of symptoms as evidenced by increasing deaths, and changing social sentiment appeared to change behavior toward vaccine acceptance and inoculation. Researchers similarly extended vaccine social media sentiment analysis [ 6 , 7 , 8 ].…”

Novel Hate Speech Detection Using Word Cloud Visualization and Ensemble Learning Coupled with Count Vectorizer

“…To augment (i.e., increase the sample size of) tweets, we perform a character level augmentation (using KeyboardAug [24], OcrAug, and RandomAug [25] methods), word level augmentation (AntonymAug [25], Contextu-alWordEmbsAug, SpellingAug SplitAug, SynonymAug, TfIdfAug, WordEmbsAug and BackTranslationAug and ReservedAug), sentence level augmentation (using Contextual-WordEmbsForSentenceAug, AbstSummAug, and LambadaAug [26]). Figure 1 shows the steps in our framework; it includes 5 major steps [27][28][29][30][31][32][33][34]. The first step is to incorporate the Twitter data upon which a comprehensive pre-processing method has been carried out, afterwards extraction of features from the resulting pre-processed tweets has been accomplished.…”

“…Figure 5 shows the most frequently used words in tweets. It can be seen that the word cloud has found the most common words for hate speech and non-hate speech that are associated with the tweets [30][31][32][33][34][35]. In Table 4, we show a sample of pre-processed and cleaned tweets.…”