Taking advantage of social media platforms, such as Twitter, this paper
provides an effective framework for emotion detection among those who are
quarantined. Early detection of emotional feelings and their trends help
implement timely intervention strategies. Given the limitations of medical
diagnosis of early emotional change signs during the quarantine period,
artificial intelligence models provide effective mechanisms in uncovering early
signs, symptoms and escalating trends. Novelty of the approach presented herein
is a multitask methodological framework of text data processing, implemented as
a pipeline for meaningful emotion detection and analysis, based on the
Plutchik/Ekman approach to emotion detection and trend detection. We present an
evaluation of the framework and a pilot system. Results of confirm the
effectiveness of the proposed framework for topic trends and emotion detection
of COVID-19 tweets. Our findings revealed Stay-At-Home restrictions result in
people expressing on twitter both negative and positive emotional semantics.
Semantic trends of safety issues related to staying at home rapidly decreased
within the 28 days and also negative feelings related to friends dying and
quarantined life increased in some days. These findings have potential to
impact public health policy decisions through monitoring trends of emotional
feelings of those who are quarantined. The framework presented here has
potential to assist in such monitoring by using as an online emotion detection
Early detection of emotional feelings and their trends help implement timely intervention strategies.
Given the limitations of medical diagnosis of early emotional change signs during the quarantine period, artiﬁcial intelligence models provide eﬀective mechanisms in uncovering early signs, symptoms and escalating trends.
1School of Computer Science and Technology, Nanjing University of Science and Technology, Nanjing 210094, China
中国南京科学技術大学 コンピュータ科学研究科 南京210094
2Faculty of Computer Science, Dalhousie University, Halifax, NS, Canada
2Faculty of Computer Science, Dalhousie University, Halifax, NS, Canada
3 Faculty of Medicine, Dalhousie University, Halifax, NS, Canada
4 Institute of Computer Science Polish Academy of Sciences, Warsaw, Poland
ワルシャワ・ポーランド科学アカデミー(ポーランド語: institute of computer science polish academy of sciences)は、ポーランドのコンピュータ科学研究所。
1 2 0 2 n a J
1 2 0 2 n a J
6 1 ] I A .
6 1 【私】 A!
s c [ 1 v 4 8 4 6 0
sc [ 1 v 4 8 4 6 0
. 1 0 1 2 : v i X r a
. 1 0 1 2 : v i X r a
2 Hamed Jelodar 1 2 et al
2 Hamed Jelodar 1 2 et al
herein is a multitask methodological framework of text data processing, implemented as a pipeline for meaningful emotion detection and analysis, based on the Plutchik/Ekman approach to emotion detection and trend detection.
In this is a multitask methodological framework of text data processing, implemented as a pipeline for meaningful emotion detection and analysis, based on the Plutchik/Ekman approach to emotion detection and trend detection。
We present an evaluation of the framework and a pilot system.
Results of conﬁrm the eﬀectiveness of the proposed framework for topic trends and emotion detection of COVID-19 tweets.
Our ﬁndings revealed Stay-At-Home restrictions result in people expressing on twitter both negative and positive emotional semantics (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%).
Keywords Twitter, NLP, Deep Learning, COVID-19, Emontion
キーワード: twitter, nlp, deep learning, covid-19, emontion
1 Introduction Over 73 million people have been aﬀected by COVID-19 across the globe .
This more than a yearlong outbreak is likely to have a signiﬁcant impact on mental health of many individuals who lost loved ones, who lost personal contacts with others due to strictly enforced public health guidelines of mandatory social segregation.
However, social media providing a platform of risk communication and exchange of feelings and emotions to curb social isolation, this text data provides a wealth of information on the natural ﬂow of people’s emotional feelings and expressions .
This rich source of data can be utilized to curb the data collection barriers during the pandemic.
The goal of this research was to use AI to uncover the hidden, implicit signal related to emotional health of people subject to mandatory quarantine, embedded in a latent manner in their twitter messages.
The purpose of the NLP system used herein is to show the meaning and emotions of users’ expressions related to a particular topic, which can be used to understand their psychological health and emotional wellbeing.
The two-fold objectives of this paper are: (a) to develop and AI framework based on machine learning models with for emotion detection and (b) to pilot this model on unstructured tweets that followed quarantine regulation using stay at home messaging during the ﬁrst wave of COVID-19.
The AI framework is described and its implemented pipeline pilot herein can be used in the emotion detection of social media information exchange during the second wave of COVID-19 and beyond to investigate the impact on any future public health guideline.
The contributions of this paper can be summarized as follows:
– A cleaned and standardized tweet dataset of COVID-19 issues is built in this research, and a new database of emotion-annotated COVID-19 tweets is presented, and this could be used for future comparisons and implementations of detection-systems based on machine learning models.
– We design a triple-task framework to investigate the emotions in eight standard positions (explained in section II B) via Plutchik’s model using the COVID-19 tweets in which all three diﬀerent tasks are complementary to each other towards a common goal.
To the best of our knowledge, this is the ﬁrst attempt that detects emotion automatically for people’s reaction to stay at home during the pandemic based on the online comments, especially for #StayAtHome.
This paper is organized into sections: (a) a review of literature on existing models on emotion detection for social media pertaining to health online communities (section 3) (b) introduce a multi-tasks framework to COVID-19
emotion detection (section 4), (c)describe data collection of twitter and research experiment (section 5), (d) discuss the eﬀectiveness of the presented AI framework and future research directions (section 6) with ﬁnal section on the conclusions on ﬁndings of emotion detection during stay at home(section 7).
Although machine learning based emotion detection approaches have been proposed within social media text analysis with the context of COVID-19, there are still many challenges remained to be addressed.
In this regard, most of the existing studies related to COVID-19, on Twitter, and other social media platforms were performed on a general public opinion, no research have speciﬁcally 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 framework that can be directly applied to COVID-19 related mood discovery, using eight types of emotional reaction and designing a deep learning model to uncover emotions based on the ﬁrst wave of the pandemic public health restriction of mandatory social segregation.
We argue that the framework can discover semantic trends of COVID-19 tweets during the ﬁrst wave of the pandemic to predict new concerns that may be associated with furthering into the new waves of COVID-19 quarantine orders and other related public health regulations.
In this section, we provide a review of literature on recent emotion detection studies with focus on; Emotion detection in online health communities, Emotion-based Lexical models, Deep learning and machine learning, and Directions for Public health decision making using social media during COVID-19 related text analytics.
Emotion detection analytics through information retrieval and NLP as a mechanism have been used to explore large text corpora of online health community communications in psychiatry, dentistry, cancer and health and ﬁtness.
5 our work, a research analyzed messages in online health communities (OHCs) to understand the most prominent emotions in health-related posts and proposed a computational model that can exploit the semantic information from the text data .
In our previous work , sentiment and latent-topics techniques application to COVID-19 comments in social media shed light on the usefulness of NLP methods in uncovering issues directly related to COVID-19 public health decision making.
We expect to extend the methodology, in this study, within our goal of extracting meaningful knowledge of emotional expression words from people’s reactions during mandatory quarantine using the StayAtHome hashtag on Twitter.
This knowledge is essential as it can help decision makers to take necessary actions to control the adverse emotional eﬀects of various public health policies, especially during the emerging waves of the pandemic.
Clearly, negative emotional eﬀects such as anger and fear can lead to negative social reactions.
To the best of the authors’ knowledge, little research have been done to understand the emotional expression during mandatory quarantine, partly due to diﬃculties in collecting such personal level data during the pandemic.
4 METHODS This paper’s methods provide step-by-step approach to text data processing, emotion detection and intensity scoring, emotion semantic trends calculation and ﬁnally evaluation of the deep learning algorithm using training and testing data.
Application of this inclusion criteria is a critical step because the quality of the input data directly aﬀects the output.Lexical text analysis, Data-Dimension Reduction, and NLP Preprocessing of data are necessary to clean the data by removing the noisy and inconsistent tweets and also analyzing the relevant data to identify relevant and appropriate information related to the topic of interest.
For this purpose, four NLP techniques are used: sentence splitting and word tokenization, removing stop-words, HTML cleaning to remove unnecessary contents, removing of stop-words and hashtags, and stemming to remove preﬁxes and suﬃxes hence returning to the root.
Each tweet passed through a set of ﬁlters that are created based on the points described above.
As stated in the objectives of this paper we want to detect the emotions expressed in the tweets that are in English language, non-English tweets are eliminated from the dataset and the classiﬁer is then trained using English tweets only.
In this research, three sub processing steps are carried out on every annotated tweet: (a) identifying type of emotion using Plutchik-theory and hence reproducibility is warranted, (b) assigning the emotion score obtained from the National Research Council Canada (NRC)/ NRC Emotion Lexicon  and (c) identifying the emotion and the maximum association score based on the scores computed according to the following rule.
a)plutchik-theoryを用いた感情のタイプを特定し、したがって再現性が保証される、(b)national research council canada (nrc)/nrc emotion lexicon から得られた感情スコアを割り当て、(c) 次の規則に従って算出されたスコアに基づいて感情と最大連想スコアを識別する。 訳抜け防止モード: 本研究では,アノテートされたツイート毎に3つのサブ処理ステップを実行する。 (a) Plutchik理論を用いた感情のタイプを特定し,再現性を保証する。 (b)カナダ国立研究会議(NRC)/NRC感情レキシコンから得られる感情スコアを割り当てる c) 以下のルールに従って算出したスコアに基づいて感情と最大関連スコアを同定した。
In part (b) for assigning the score, we calculate the total emotion association score as the sum of scores of the terms depicting higher scores for higher intensity of the emotion of the tweets (See Table I, example).
Fig 2 provides an example of selecting the score for COVID-19 tweets.
For example, from the tweet after text processing showed “Sad man friend whos livin skin cant stand company” and will have the emotion SAD associated with FEAR and this emotional expression provided the highest score from NRC based Lexicon
例えば、テキスト処理後のツイートから、"Sad man friend whos livin skin cant stand company" が示され、FEARと関連付けられた感情SADを持ち、この感情表現はNRCベースのLexiconから最高スコアを得た。
8 Hamed Jelodar 1 2 et al
8 Hamed Jelodar 1 2 et al
Fig. 2: Example of the process of determining the score for a pure tweet that is related to #StayAtHome
Table 1: EXAMPLE OF THE TWEETS WITH VARIOUS EMOTIONS
Tweet ID. A B
ツイートid。 A B
C Tweet Today has been a challenging day, here’s to tomorrow Score Tweet A day is a long time in the coronavirus pandemic.
C ツイート・トゥデイ(Tweet Today)は、今日から明日のScore Tweet A Dayは、新型コロナウイルスのパンデミックの中で長い時間だ。
Score Tweet Score
Tweet without stop-words Anger=0∼Anticipation=1∼Disgust=0∼Fear=0 ∼Joy=0 ∼Sadness=0∼Surprise=0∼Trust=0 Anger=0∼Anticipation=2∼Disgust=0∼Fear=0 ∼Joy=0 ∼Sadness=0∼Surprise=0∼Trust=0 Looking forward to those summer days when I can enjoy the beach and the ocean breeze again????.
Another advantage of this method is the identiﬁcation of semantic-trends over time, which we consider in this research as means to discover unusual semantic trends based on the ﬁrst wave of the pandemic of #StayAtHome tweets.
Then by investigating the distributions of these semantic-topics in various days, we obtain semantic trends.
As a second part of this task, we compute the types of each emotion to identify the trends among diﬀerent emotions based on task I.
By considering the length and strength of the staying at home public health order in the ﬁrst wave, we believe that it is necessary to examine the changes in people’s emotions by monitoring the time trends and ﬂuctuations of directions using twitter data.
In the following, we discuss the details of the designed deep-learning model, the number of convolutional layers, and dense layers to build our COVID-19 emotion detection framework; since the input layer is the sentence representation, a convolutional layer is then deployed to obtain the sequence level feature from the sentence sequences.
5.1 Informative Trends of the ﬁrst wave: Emotion and Semantic
Trending topics, to a certain extent, describe the opinion of a community and provide the means to analyze it, knowing where public attention is at a certain time point and this has become a matter of interest for researchers and health professionals.
5.2 Relationship between semantic trends and StayAtHome Tweets
5.2 セマンティックトレンドとStayAtHome Tweets
It is diﬃcult to identify the key concepts discussed by users from a million tweets in traditional ways, so we examine NLP methods (LDA and PLSA) to extract topics based on semantic aspects to better understand behaviors and People’s reactions, during stay at home.
Therefore, we consider an LDA model for performing Task II.
To implement our analytic framework’s detection of semantic trends shown in the topics during the initial wave of the pandemic, we investigate ﬁve top topics (i.e., S1, S2, S3, S4, and S5, Fig 4) to better understand the online community reactions change over time.
It is important to notice that negative feelings grow over time, eg S4 (words Friends, Die, Virus) increases at a rate of 0.14 (p=0.0001) and S5 (Home, remote, quarantine, health life) increases at a rate of 0.06 (p=0.0005) over the course of the 28 days of tweet text data collection period.
5.3 Relationship Between Emotion Trends and StayAtHome Tweets
5.3 感情傾向とStayAtHome つぶやきの関係
In Tasks 1 and 2 of the framework, we take the advantage of NRC emotional lexical, which is supported by Plutchik’s theory based on 14,000 words for ﬁnding the eight primary emotions: anger, anticipation, joy, surprise, sadness, disgust, trust, and fear.
According to Fig 5 the most of the emotions depict in tweets across time are “Anticipation”.
As shown in Fig 6 the mean percentage of “anticipation” detected per day across 30 day period is 53.6 (95% CI: 52.4-55.0) with the least shown by “Surprise” (mean 1.5% , CI: 1.4-1.6%), and “Sadness” (mean 2.2,CI:2.1-2.3%).
Nevertheless, in this study anticipation stemmed out of the hashtag “stay-at-home”, a restriction on a socially undesirable action and therefore, one can assume anticipation is mostly directed towards a negative emotional feeling, of perceived susceptibility.
Negative connotation of anticipation and fear can be overcome with public education using social media.
5.4 Deep learning model conﬁgurations and Training details
The objective of task III of this work is to automatically detect emotions from #StayAtHome tweets by enabling Multi-Channel CNN methodology as a computational model for the emotion detection of the COVID-19 tweets.
First we get the input data of the Task I, we leverage our COVID-19 tweet data to train our own word embedding with Word2Vec technique  which provides a much richer text representation than the classical, word-based approaches.
The advantage of the CNN model comes for detection of the type of emotions in COVID-19 tweets, which enables to avoid overﬁtting and still be able to ﬁnd complex patterns to emotion detection in the introduced data.
A study that included multiple unigrams and bi-grams related to COVID19 twitter feeds were analyzed using machine learning approaches and their ﬁndings were similar to ours in that the dominant theme identiﬁed was anticipation with a mixed of feelings of trust, anger and fear .
To develop a framework that can understand the type of standard emotions contained in COVID-19 sentences in social media is among the challenging topics of NLP for the public health and mental healthcare delivery  – .
As the pandemic evolved public health guidelines became strict measures imposed on the general public.
This one track minded approach of combating the spread or better known as ﬂattening the curve, neglected emotional and mental health of the individuals who were subject to those strict public health ordering.
This study ﬁndings showed a mechanism of how the emotions and semantic trends of people’s reactions to COVID-19 public health restrictions can be obtained for knowledge discovery and can inform related decision making.
The advantage of such an approach is that identifying these online trends provide easy and helpful information about public reactions to particular issues and thus it has recently attracted the attention of medical and computer researchers.
The framework proposed in this research covers three practical tasks that are related to each other with a common goal to develop a deep-learning system for emotion detection and analysis of informative trends from COVID-19 tweets of people’s reaction during the stay-at-home.
Our ﬁnal results uncovered important directions for public health policy makers  and decision makers  to pay attention to emotional issues that stemmed from those strict public health restrictions.
Although more tweets can be extracted based on #StayAtHome, we believe that the number of current tweets is suﬃcient to draw reasonable conclusions to direct possibilities of uncovering importance of consequences of public health orders and restrictions.
Further advantage of a temporally larger dataset is an opportunity of a longitudinal study combining geographicallybased tweeter -detected emotions with COVID-19 incidences and expanded public health regulations to enable geographic-area targeted public health decision making.
Moreover, it presents a more insightful understanding of COVID-19 tweets by automatically identifying the type of emotions including both negative and positive reaction and the magnitude of their presentation.
We identiﬁed ways to improve the ﬁndings in future research.
We discuss potentially signiﬁcant, realistic future work, such as extending the longitudinal character of the results, inclusion of geography-based public health orders and spatiallyannotated COVID-19 case loads.
Ethical Approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
Karmegam, D., T. Ramamoorthy, and B. Mappillairajan, A systematic review of techniques employed for determining mental health using social media in psychological surveillance during disasters.
Karmegam, D., T. Ramamoorthy, B. Mappillairajan, A systematic Review of techniques used used for mental health using social media in Psycho surveillance during disasters。
Disaster medicine and public health preparedness, 2020.
14(2): p. 265-272.
 Stead, William W., and Nancy M. Lorenzi.
”Health informatics: linking investment to value.” Journal of the American Medical Informatics Association 6, no.
医療情報学協会(american medical informatics association 6, no.)は、アメリカ医療情報学協会(american medical informatics association)が発行する学術誌。 訳抜け防止モード: 健康情報学」 と、american medical informatics association 6, no. のジャーナルに書いている。