Artificial Intelligence for Emotion-Semantic Trending and People Emotion Detection During COVID-19 Social Isolation
Hamed Jelodar, Rita Orji, Stan Matwin, Swarna Weerasinghe, Oladapo Oyebode, Yongli Wang
NNoname manuscript No. (will be inserted by the editor)
Artificial Intelligence for Emotion-SemanticTrending and People Emotion Detection DuringCOVID-19 Social Isolation
Hamed Jelodar · Rita Orji · StanMatwin · Swarna Weerasinghe · Oladapo Oyebode · Yongli Wang Received: date / Accepted: date
Abstract
Taking advantage of social media platforms, such as Twitter, thispaper provides an effective framework for emotion detection among those whoare quarantined. Early detection of emotional feelings and their trends helpimplement timely intervention strategies. Given the limitations of medicaldiagnosis of early emotional change signs during the quarantine period, ar-tificial intelligence models provide effective mechanisms in uncovering earlysigns, symptoms and escalating trends. Novelty of the approach presented
H. [email protected]. [email protected]. [email protected]. [email protected]. [email protected]. [email protected] School of Computer Science and Technology, Nanjing University of Science andTechnology, Nanjing 210094, China Faculty of Computer Science, Dalhousie University, Halifax, NS, Canada Faculty of Medicine, Dalhousie University, Halifax, NS, Canada Institute of Computer Science Polish Academy of Sciences, Warsaw, Poland a r X i v : . [ c s . A I] J a n Hamed Jelodar1 2
H. [email protected]. [email protected]. [email protected]. [email protected]. [email protected]. [email protected] School of Computer Science and Technology, Nanjing University of Science andTechnology, Nanjing 210094, China Faculty of Computer Science, Dalhousie University, Halifax, NS, Canada Faculty of Medicine, Dalhousie University, Halifax, NS, Canada Institute of Computer Science Polish Academy of Sciences, Warsaw, Poland a r X i v : . [ c s . A I] J a n Hamed Jelodar1 2 et al. herein is a multitask methodological framework of text data processing, imple-mented as a pipeline for meaningful emotion detection and analysis, based onthe Plutchik/Ekman approach to emotion detection and trend detection. Wepresent an evaluation of the framework and a pilot system. Results of confirmthe effectiveness of the proposed framework for topic trends and emotion de-tection of COVID-19 tweets. Our findings revealed Stay-At-Home restrictionsresult in people expressing on twitter both negative and positive emotional se-mantics (feelings), where negatives are “Anger” (8.5% of tweets), followed by“Fear” (5.2%), “Anticipation” (53.6%) and positive emotional semantics are“Joy” (14.7%) and “Trust” (11.7%). Semantic trends of safety issues relatedto staying at home rapidly decreased within the 28 days and also negativefeelings related to friends dying and quarantined life increased in some days.These findings have potential to impact public health policy decisions throughmonitoring trends of emotional feelings of those who are quarantined. Theframework presented here has potential to assist in such monitoring by usingas an online emotion detection tool kit.
Keywords
Twitter, NLP, Deep Learning, COVID-19, Emontion
Over 73 million people have been affected by COVID-19 across the globe [3].This more than a yearlong outbreak is likely to have a significant impact onmental health of many individuals who lost loved ones, who lost personal con-tacts with others due to strictly enforced public health guidelines of mandatorysocial segregation. Complex psychological reactions to COVID-19 regulatorymechanisms of mandatory quarantine and related emotional reactions has beenrecognized as hard to disentangle [1] – [4]. A study conducted in Belgium foundsocial media being positively associated with constructive coping for adoles-cents with anxious feelings during the quarantine period of COVID-19 [4]. An-other study conducted among social media users during COVID-19 pandemicin Spain was able to capture added stress placed on people’s emotional healthduring the pandemic period [5]. However, social media providing a platformof risk communication and exchange of feelings and emotions to curb socialisolation, this text data provides a wealth of information on the natural flowof people’s emotional feelings and expressions [6]. This rich source of data canbe utilized to curb the data collection barriers during the pandemic. The goalof this research was to use AI to uncover the hidden, implicit signal related toemotional health of people subject to mandatory quarantine, embedded in alatent manner in their twitter messages.Within the context of this paper, an NLP-based emotion detection systemaims to provide useful information by examining unstructured text data usedin social media. The purpose of the NLP system used herein is to show themeaning and emotions of users’ expressions related to a particular topic, whichcan be used to understand their psychological health and emotional wellbeing. itle Suppressed Due to Excessive Length 3
In this regard, use of NLP-based approach for emotion detection from complextextual structures such as social media (e.g., Twitter) remains a challenge inbiomedical applications of AI.The goal of this paper is to contribute an AI based methodological frame-work that can uncover emotion semantic trends that can better understand theimpact and the design of quarantine regulations. The two-fold objectives of thispaper are: (a) to develop and AI framework based on machine learning modelswith for emotion detection and (b) to pilot this model on unstructured tweetsthat followed quarantine regulation using stay at home messaging during thefirst wave of COVID-19. We investigate emotions and semantic structures todiscover knowledge from general public tweeter exchanges. We analyze thestructure of vocabulary patterns used on Twitter with a specific focus on theimpact of the Stay-At-Home public health order during the first wave of theCOVID-19. The AI framework is described and its implemented pipeline pi-lot herein can be used in the emotion detection of social media informationexchange during the second wave of COVID-19 and beyond to investigate theimpact on any future public health guideline.We aim to demonstrate the effectiveness of deep learning models for de-tecting emotions from COVID-19 tweets. In addition, our findings will providedirections public health decision making on emotion trend detection over a fourweeks period in relation to “Stay at Home” order. The contributions of thispaper can be summarized as follows: – A cleaned and standardized tweet dataset of COVID-19 issues is built inthis research, and a new database of emotion-annotated COVID-19 tweetsis presented, and this could be used for future comparisons and implemen-tations of detection-systems based on machine learning models. – We design a triple-task framework to investigate the emotions in eightstandard positions (explained in section II B) via Plutchik’s model usingthe COVID-19 tweets in which all three different tasks are complementaryto each other towards a common goal. – We discover semantic-word trends via various models such as latent Dirich-let allocation (LDA) and probabilistic latent semantic analysis (PLSA). Weaim to have a semantic knowledge discovery based on topic trends and se-mantic structures during the first wave of the pandemic, which providesan effective mechanism for managing future waves. – A deep learning model based convolutional neural network (CNN)) is pre-sented for emotion detection from the COVID-19 tweets. To the best ofour knowledge, this is the first attempt that detects emotion automaticallyfor people’s reaction to stay at home during the pandemic based on theonline comments, especially for
Hamed Jelodar1 2
Hamed Jelodar1 2 et al. emotion detection (section 4), (c)describe data collection of twitter and re-search experiment (section 5), (d) discuss the effectiveness of the presented AIframework and future research directions (section 6) with final section on theconclusions on findings of emotion detection during stay at home(section 7).
Although machine learning based emotion detection approaches have beenproposed within social media text analysis with the context of COVID-19,there are still many challenges remained to be addressed. In this regard, mostof the existing studies related to COVID-19, on Twitter, and other social me-dia platforms were performed on a general public opinion, no research havespecifically investigated emotions related to quarantine “stay at home” order,public health policy of social segregation, that is widely used across the globe.Novelty of the methods used in this paper consists of a multi-task frameworkthat can be directly applied to COVID-19 related mood discovery, using eighttypes of emotional reaction and designing a deep learning model to uncoveremotions based on the first wave of the pandemic public health restrictionof mandatory social segregation. We argue that the framework can discoversemantic trends of COVID-19 tweets during the first wave of the pandemicto predict new concerns that may be associated with furthering into the newwaves of COVID-19 quarantine orders and other related public health regula-tions. Our novel approach presented herein can help future public health crisismanagement in the new waves of the Coronavirus pandemic.Moreover, public health decision makers need to understand the temporalpatterns of emotional reactions on the population when these public healthregulatory measures are continued. To fulfill this need, we investigate the se-mantic topic and emotion trends to better understand the people’s reactionsfrom the initial wave of the pandemic.
NLP and Machine Learning has been used within the context of identifyingthe type of emotions in twitter texts. In this section, we provide a review of lit-erature on recent emotion detection studies with focus on; Emotion detectionin online health communities, Emotion-based Lexical models, Deep learningand machine learning, and Directions for Public health decision making usingsocial media during COVID-19 related text analytics.Emotion detection analytics through information retrieval and NLP as a mech-anism have been used to explore large text corpora of online health communitycommunications in psychiatry, dentistry, cancer and health and fitness. Forexample, a communication tool was introduced for mental health care to un-derstand counseling content based on emotion detection and natural languageprocessing using chat assistants [7] - [12]. Similar to the proposed approach in itle Suppressed Due to Excessive Length 5 our work, a research analyzed messages in online health communities (OHCs)to understand the most prominent emotions in health-related posts and pro-posed a computational model that can exploit the semantic information fromthe text data [9]. They presented a dataset from a cancer forum with thesix common emotions based on the Ekman model and investigated the mostprominent emotions in OHCs. We proposed to use broader types of emotionsusing Plutchik’s model that contains eight emotions.In our previous work [13], sentiment and latent-topics techniques applica-tion to COVID-19 comments in social media shed light on the usefulness ofNLP methods in uncovering issues directly related to COVID-19 public healthdecision making. We expect to extend the methodology, in this study, withinour goal of extracting meaningful knowledge of emotional expression wordsfrom people’s reactions during mandatory quarantine using the StayAtHomehashtag on Twitter. This knowledge is essential as it can help decision makersto take necessary actions to control the adverse emotional effects of variouspublic health policies, especially during the emerging waves of the pandemic.Clearly, negative emotional effects such as anger and fear can lead to negativesocial reactions. To the best of the authors’ knowledge, little research havebeen done to understand the emotional expression during mandatory quaran-tine, partly due to difficulties in collecting such personal level data during thepandemic. The authors of a study in India analyzed real-time posts on Twit-ter during COVID-19, and they were able to identify the expression of moodof the nation through their analysis [14]. Also, they developed a platform forviewing the trends in emotional change across the country during a specificinterval. As a result, their system allowed users to view the mood of India onspecific events happening in the country during COVID-19. Development ofsuch a platform for Canada may be a far reaching goal of this study. This studypresents the first step towards development of such online tools to monitor themoods and emotions to inform public health decision makers.
This paper’s methods provide step-by-step approach to text data processing,emotion detection and intensity scoring, emotion semantic trends calculationand finally evaluation of the deep learning algorithm using training and testingdata.4.1 Multi-Task FrameworkIn this section, we present a multi-task framework based on Plutchik’s Emotionmodel [15] and deep learning techniques to address aforementioned researchobjectives of this paper. Our approach includes three main tasks. Plutchik’s isan operationalization of Ekman [16]. The first task is to create models to in-vestigate emotional reaction to mandatory restrictions of Stay-At-Home using
Hamed Jelodar1 2
Hamed Jelodar1 2 et al. tweets. The second task, is to show how to discover semantic and emotionaltrends to obtain patterns depicted in the first wave of the pandemic for the30 days period from April 28th to June 1st, 2020. Finally, the third task, amachine learning deep neural network is built as an emotion detection systemthat can be used for social media exchange data analysis during the quarantineperiod. Our framework, including these three tasks, is presented in Fig. 1.1) itle Suppressed Due to Excessive Length 7
Fig. 1: Research framework and pipeline for the COVID-19 tweet emotioncapture and analysisfor any of the eight emotions. A tweet that does not associate with any emo-tion receives a score of zero (0), as showed in Table 1. This AI based emotiondetection task uncovers emotion semantics with emotion valuation (strength)attached to each emotion lexicon in each of the tweets.Since the labelling is done automatically and no human tagging is used, wewill have consistent data annotation. This model defines eight basic emotionsand makes it possible to provide consistent classification of texts to uncoverthe trend in the data to reach the objectives of this research. Fig. 2 providesan example of selecting the score for COVID-19 tweets. For example, fromthe tweet after text processing showed “Sad man friend whos livin skin cantstand company” and will have the emotion SAD associated with FEAR andthis emotional expression provided the highest score from NRC based Lexicon
Hamed Jelodar1 2
Hamed Jelodar1 2 et al.
Fig. 2: Example of the process of determining the score for a pure tweet thatis related to
Tweet ID. Tweet without stop-words LabelA Tweet Today has been a challenging day, here’s to tomorrow AnticipationScore Anger=0 ∼ Anticipation=1 ∼ Disgust=0 ∼ Fear=0 ∼ Joy=0 ∼ Sadness=0 ∼ Surprise=0 ∼ Trust=0B Tweet A day is a long time in the coronavirus pandemic. AnticipationScore Anger=0 ∼ Anticipation=2 ∼ Disgust=0 ∼ Fear=0 ∼ Joy=0 ∼ Sadness=0 ∼ Surprise=0 ∼ Trust=0C Tweet Looking forward to those summer days when I can enjoy the beach and the ocean breeze again????. Stay positive and healthy everyone. JoyScore Anger=0 ∼ Anticipation=1 ∼ Disgust=0 ∼ Fear=0 ∼ Joy=3 ∼ Sadness=0 ∼ Surprise=0 ∼ Trust=1 when the process is detecting the predominant emotion.3) Task II: Emotion/Semantic-Trends of itle Suppressed Due to Excessive Length 9 low:- The probabilistic latent semantic analysis (PLSA) model is used as an NLPtechnique that can display topical similarities between words.- The latent Dirichlet allocation (LDA) model has proven very useful for se-mantic extraction and generating trends over time. LDA has been success-fully applied in several applications such as topic discovery, temporal seman-tic trends, document classification, and finding relations between documents.Another advantage of this method is the identification of semantic-trends overtime, which we consider in this research as means to discover unusual seman-tic trends based on the first wave of the pandemic of 1 2
Tweet ID. Tweet without stop-words LabelA Tweet Today has been a challenging day, here’s to tomorrow AnticipationScore Anger=0 ∼ Anticipation=1 ∼ Disgust=0 ∼ Fear=0 ∼ Joy=0 ∼ Sadness=0 ∼ Surprise=0 ∼ Trust=0B Tweet A day is a long time in the coronavirus pandemic. AnticipationScore Anger=0 ∼ Anticipation=2 ∼ Disgust=0 ∼ Fear=0 ∼ Joy=0 ∼ Sadness=0 ∼ Surprise=0 ∼ Trust=0C Tweet Looking forward to those summer days when I can enjoy the beach and the ocean breeze again????. Stay positive and healthy everyone. JoyScore Anger=0 ∼ Anticipation=1 ∼ Disgust=0 ∼ Fear=0 ∼ Joy=3 ∼ Sadness=0 ∼ Surprise=0 ∼ Trust=1 when the process is detecting the predominant emotion.3) Task II: Emotion/Semantic-Trends of itle Suppressed Due to Excessive Length 9 low:- The probabilistic latent semantic analysis (PLSA) model is used as an NLPtechnique that can display topical similarities between words.- The latent Dirichlet allocation (LDA) model has proven very useful for se-mantic extraction and generating trends over time. LDA has been success-fully applied in several applications such as topic discovery, temporal seman-tic trends, document classification, and finding relations between documents.Another advantage of this method is the identification of semantic-trends overtime, which we consider in this research as means to discover unusual seman-tic trends based on the first wave of the pandemic of 1 2 et al.
Table 2: TABLE II. DETAILS OF THE
This section describes the dataset, generates emotion/semantic trends and theinformative results with various experiments to evaluate the performance ofthe research model. In fact, to show the value of the framework, we conductexperiments on collected COVID-19 tweets and generate informative emo-tion/semantic trens. We provide a comparison with a standard base-line todemonstrate the superiority of our CNN model for emotion detection using F1score and accuracy as standard metrics. In this research, we use 90% of thedata for training and 10% for testing for all experiments. We focused on thetweets from the first waves of the COVID-19 pandemic based on itle Suppressed Due to Excessive Length 11
Fig. 3: The precision metric of word clustering for semantic-topic trendingbased on LDA and PLSAthe community. First, we investigate PLSA and LDA models to analyze andvalidate the relationship between semantic-topics extracted from COVID-19tweets and related issues of the pandemic. For this purpose, we use a Malletpackage. Then, we generate 100 topics and only focused on 5 top topics ofall COVID-19 tweets as a result of topic modeling for discovering semanticrelated-words. Fig 3, compares the performance of the two topic modelingmethods we have considered to identify emotional trends in our twitter data.And showed, LDA can capture semantic topics better than the PLSA modelfor extracting semantic-topics of 1 2
Fig. 3: The precision metric of word clustering for semantic-topic trendingbased on LDA and PLSAthe community. First, we investigate PLSA and LDA models to analyze andvalidate the relationship between semantic-topics extracted from COVID-19tweets and related issues of the pandemic. For this purpose, we use a Malletpackage. Then, we generate 100 topics and only focused on 5 top topics ofall COVID-19 tweets as a result of topic modeling for discovering semanticrelated-words. Fig 3, compares the performance of the two topic modelingmethods we have considered to identify emotional trends in our twitter data.And showed, LDA can capture semantic topics better than the PLSA modelfor extracting semantic-topics of 1 2 et al.
Fig. 4: Semantic trends of the initial waves of COVID-19 pandemic by itle Suppressed Due to Excessive Length 13
Fig. 5: Distribution of emotion trends in 1 2
Fig. 5: Distribution of emotion trends in 1 2 et al.
Fig. 6: Emotion Trends of the COVID-19 tweets in line graphs.gust” (mean=2.3, CI:2.2-2.4) and “Sadness” (mean=2.2, CI:2.1-2.3). Amongpositive feelings “joy” (mean=14.7,CI:14.4-15.4) is the second highest emo-tion.According to psychology literature [36], Anticipation and Surprise can berelated to positive or negative health emotional outcomes. Nevertheless, in thisstudy anticipation stemmed out of the hashtag “stay-at-home”, a restrictionon a socially undesirable action and therefore, one can assume anticipationis mostly directed towards a negative emotional feeling, of perceived suscep-tibility. It is important to consider that these tweets were exchanged duringthe early pandemic period of April 28th to June 1st, 2020 of the first waveand North America reported the peak in May, 2020. Anger may be directedtowards missing summer outdoor activities due to stay-at-home restrictions,whereas fear may be expressed by those living in high risk clusters of elderly itle Suppressed Due to Excessive Length 15 and those with chronic conditions. It is important to note that the negativefeeling of disgust was minimal and people may be aware of the importance ofquarantine regulations. These emotion expressions and trend detection overtime provide important messages where public health decision makers can beaware of in the future public health regulation ordering. Though people trustthe public health measures, their anticipation towards negative emotions needto be considered in the way the public health regulations are ordered and im-posed. Negative connotation of anticipation and fear can be overcome withpublic education using social media.5.4 Deep learning model configurations and Training detailsThe objective of task III of this work is to automatically detect emotionsfrom 1 2
Fig. 6: Emotion Trends of the COVID-19 tweets in line graphs.gust” (mean=2.3, CI:2.2-2.4) and “Sadness” (mean=2.2, CI:2.1-2.3). Amongpositive feelings “joy” (mean=14.7,CI:14.4-15.4) is the second highest emo-tion.According to psychology literature [36], Anticipation and Surprise can berelated to positive or negative health emotional outcomes. Nevertheless, in thisstudy anticipation stemmed out of the hashtag “stay-at-home”, a restrictionon a socially undesirable action and therefore, one can assume anticipationis mostly directed towards a negative emotional feeling, of perceived suscep-tibility. It is important to consider that these tweets were exchanged duringthe early pandemic period of April 28th to June 1st, 2020 of the first waveand North America reported the peak in May, 2020. Anger may be directedtowards missing summer outdoor activities due to stay-at-home restrictions,whereas fear may be expressed by those living in high risk clusters of elderly itle Suppressed Due to Excessive Length 15 and those with chronic conditions. It is important to note that the negativefeeling of disgust was minimal and people may be aware of the importance ofquarantine regulations. These emotion expressions and trend detection overtime provide important messages where public health decision makers can beaware of in the future public health regulation ordering. Though people trustthe public health measures, their anticipation towards negative emotions needto be considered in the way the public health regulations are ordered and im-posed. Negative connotation of anticipation and fear can be overcome withpublic education using social media.5.4 Deep learning model configurations and Training detailsThe objective of task III of this work is to automatically detect emotionsfrom 1 2 et al.
Fig. 7: The F1-score for COVID-19 emotion detection by comparing the CNNmodel and the LSTM model5.6 Evaluation of the effectiveness of the frameworkWe evaluated the performance of the research model with emotions classes.Fig. 7 provides a clear view of variation with different parameters using wordembedding trained. In particular, our preliminary results indicate that themulti-channel CNN out performs the LSTM-Softmax in terms of various epochsbased on a multi-class F-score as a standard metric. The advantage of the CNNmodel comes for detection of the type of emotions in COVID-19 tweets, whichenables to avoid overfitting and still be able to find complex patterns to emo-tion detection in the introduced data.
Our results, in general, suggested that the machine learning methods we useare appropriate for the emotion detection of COVID-19 tweets. The study re-sults clearly demonstrated anticipation as a prominent emotional semantic.Among various definitions for this emotion semantic analysis, anticipation isconsidered as one of “the mature ways of dealing with real stress”[30]. Re-garding this definition, people can lower their stress during the COVID-19pandemic by anticipating and preparing how they are going to deal with it.Anticipation can be interpreted as either future positive and negative events, itle Suppressed Due to Excessive Length 17 according to [31], and are aligned with hope and fear which are the typical an-ticipatory feelings that arise in response to possibilities of future such events.A study that included multiple unigrams and bi-grams related to COVID-19 twitter feeds were analyzed using machine learning approaches and theirfindings were similar to ours in that the dominant theme identified was antic-ipation with a mixed of feelings of trust, anger and fear [32].To develop a framework that can understand the type of standard emotionscontained in COVID-19 sentences in social media is among the challengingtopics of NLP for the public health and mental healthcare delivery [33] – [35].Therefore, in this paper, a multi-task framework is presented to make a smartemotion detection system based on the Stay-At-Home aspects of COVID-19tweets. All the experiments are performed using different parameter settings.The results suggest that the CNN model with two convolutional layers withfilter sizes of 3 and 4 can achieve good performance with various metrics foremotion detection and classification.Use of online social network text data to understand user health behav-iors and emotions has become an emerging research area in NLP and healthinformatics [36] – [49]. COVID-19 introduced an unprecedented global threatthat public health planning and policy making community are still strugglingto find best practices to curb the pandemic. As the pandemic evolved publichealth guidelines became strict measures imposed on the general public. Thisone track minded approach of combating the spread or better known as flat-tening the curve, neglected emotional and mental health of the individuals whowere subject to those strict public health ordering. This study findings showeda mechanism of how the emotions and semantic trends of people’s reactions toCOVID-19 public health restrictions can be obtained for knowledge discoveryand can inform related decision making. The advantage of such an approachis that identifying these online trends provide easy and helpful informationabout public reactions to particular issues and thus it has recently attractedthe attention of medical and computer researchers. The framework proposedin this research covers three practical tasks that are related to each otherwith a common goal to develop a deep-learning system for emotion detectionand analysis of informative trends from COVID-19 tweets of people’s reactionduring the stay-at-home. Our final results uncovered important directions forpublic health policy makers [40] and decision makers [41] to pay attention toemotional issues that stemmed from those strict public health restrictions.This research has some limitations, e.g the size of dataset, data inclusionlimited to emotions based on texts of COVID-19 issues. Currently, our dataconsists of 1,047,968 tweets based on 1 2
Our results, in general, suggested that the machine learning methods we useare appropriate for the emotion detection of COVID-19 tweets. The study re-sults clearly demonstrated anticipation as a prominent emotional semantic.Among various definitions for this emotion semantic analysis, anticipation isconsidered as one of “the mature ways of dealing with real stress”[30]. Re-garding this definition, people can lower their stress during the COVID-19pandemic by anticipating and preparing how they are going to deal with it.Anticipation can be interpreted as either future positive and negative events, itle Suppressed Due to Excessive Length 17 according to [31], and are aligned with hope and fear which are the typical an-ticipatory feelings that arise in response to possibilities of future such events.A study that included multiple unigrams and bi-grams related to COVID-19 twitter feeds were analyzed using machine learning approaches and theirfindings were similar to ours in that the dominant theme identified was antic-ipation with a mixed of feelings of trust, anger and fear [32].To develop a framework that can understand the type of standard emotionscontained in COVID-19 sentences in social media is among the challengingtopics of NLP for the public health and mental healthcare delivery [33] – [35].Therefore, in this paper, a multi-task framework is presented to make a smartemotion detection system based on the Stay-At-Home aspects of COVID-19tweets. All the experiments are performed using different parameter settings.The results suggest that the CNN model with two convolutional layers withfilter sizes of 3 and 4 can achieve good performance with various metrics foremotion detection and classification.Use of online social network text data to understand user health behav-iors and emotions has become an emerging research area in NLP and healthinformatics [36] – [49]. COVID-19 introduced an unprecedented global threatthat public health planning and policy making community are still strugglingto find best practices to curb the pandemic. As the pandemic evolved publichealth guidelines became strict measures imposed on the general public. Thisone track minded approach of combating the spread or better known as flat-tening the curve, neglected emotional and mental health of the individuals whowere subject to those strict public health ordering. This study findings showeda mechanism of how the emotions and semantic trends of people’s reactions toCOVID-19 public health restrictions can be obtained for knowledge discoveryand can inform related decision making. The advantage of such an approachis that identifying these online trends provide easy and helpful informationabout public reactions to particular issues and thus it has recently attractedthe attention of medical and computer researchers. The framework proposedin this research covers three practical tasks that are related to each otherwith a common goal to develop a deep-learning system for emotion detectionand analysis of informative trends from COVID-19 tweets of people’s reactionduring the stay-at-home. Our final results uncovered important directions forpublic health policy makers [40] and decision makers [41] to pay attention toemotional issues that stemmed from those strict public health restrictions.This research has some limitations, e.g the size of dataset, data inclusionlimited to emotions based on texts of COVID-19 issues. Currently, our dataconsists of 1,047,968 tweets based on 1 2 et al. dataset representing different emotions is also a limitation of our work. We didnot consider slang or emoticons to compute emotions in the tweet contents,it would be useful to build a new emotional lexical to cover slang words re-lated to COVID-19 issues. A significantly longer temporal horizon longitudinaldataset would allow us using LSTM on sequences of tweets, as well as replac-ing the cross-validation. train/test approach with one based time-stamp ratherthan the cross-validation approach. Further advantage of a temporally largerdataset is an opportunity of a longitudinal study combining geographically-based tweeter -detected emotions with COVID-19 incidences and expandedpublic health regulations to enable geographic-area targeted public health de-cision making. The framework we developed showed potential to accuratelyuncover emotional responses and temporal trend detection of mood changesdue to quarantine related public health orders.
This paper presented a novel framework for emotion detection using COVID-19 tweets in relation to the “stay-at-home” public health guidelines. For thisframework, a multi-task framework of COVID-19 emotions detection via aCNN model was presented. The research further shows that the frameworkis effective in capturing the emotions and semantics trends in social mediamessages during the pandemic. Moreover, it presents a more insightful un-derstanding of COVID-19 tweets by automatically identifying the type ofemotions including both negative and positive reaction and the magnitudeof their presentation. The framework can be applied to uncover reactions tosimilar public health policies that affect people’s well being. We identifiedways to improve the findings in future research. We discuss potentially sig-nificant, realistic future work, such as extending the longitudinal character ofthe results, inclusion of geography-based public health orders and spatially-annotated COVID-19 case loads.
Ethical Approval
All procedures performed in studies involving human participants were in ac-cordance with the ethical standards of the institutional and/or national re-search committee and with the 1964 Helsinki declaration and its later amend-ments or comparable ethical standards.
Declaration of Conflict of Interest :
All authors declare no conflict ofinterest directly related to the submitted work. itle Suppressed Due to Excessive Length 19
References1 2