key: cord-0426082-lmrkzpmj authors: Valinejad, J.; Guo, Z.; Cho, J.-H.; Chen, I.-R. title: Measuring Community Resilience During the COVID-19 based on Community Wellbeing and Resource Distribution date: 2022-05-25 journal: nan DOI: 10.1101/2022.05.23.22275454 sha: 7a0b594955fa80ebdc5dbf318fabf056b1023c10 doc_id: 426082 cord_uid: lmrkzpmj The COVID-19 pandemic has severely harmed every aspect of our daily lives, resulting in a slew of social problems. Therefore, it is critical to accurately assess the current state of community functionality and resilience under this pandemic for successful recovery. To this end, various types of social sensing tools, such as tweeting and publicly released news, have been employed to understand individuals and communities thoughts, behaviors, and attitudes during the COVID-19 pandemic. However, some portions of the released news are fake and can easily mislead the community to respond improperly to disasters like COVID-19. This paper aims to assess the correlation between various news and tweets collected during the COVID-19 pandemic on community functionality and resilience. We use fact-checking organizations to classify news as real, mixed, or fake, and machine learning algorithms to classify tweets as real or fake to measure and compare community resilience (CR). Based on the news articles and tweets collected, we quantify CR based on two key factors, community wellbeing and resource distribution, where resource distribution is assessed by the level of The recent outbreak of COVID-19 has disrupted every aspect of our daily lives. To 3 absorb and adapt against COVID-19 in an agile manner and quickly recover from it, 4 maintaining a healthy, socially connected, and prepared community is critical [1] . attention previously. Seven out of ten Americans use social media to exchange personal 13 information, interact with content, and connect with others [5] . According to a recent 14 research [6], the psychological states of a whole population can be revealed through 15 social media. Social media provides a platform for billions of users to communicate, 16 express sentiments, and provide real-time updates about human interaction on a large 17 scale [7] . Twitter is one of the major community social media platforms. In this regard, 18 numerous studies have employed tweeter to evaluate population behavior 19 [6, 7, 8, 9, 10, 11] . Unfortunately, fake news may negatively impact maintaining 20 community wellbeing and equitable resource distribution during COVID-19. The 21 Internet, social media, and mass media platforms have generated a large volume of 22 information flow during the COVID-19. Part of the information volume spreads false 23 information (e.g., misinformation or disinformation), rumors, fake news, or hoaxes [12] . 24 Fake news is usually observed as more novel than real news; in addition, it flows on 25 social/mass media noticeably faster, farther, and more broadly than real news [13] . 26 Fake news has been commonly used to manipulate and propagate false information by 27 appealing to users' ideological perspectives, emotions, and desires to spread their views 28 to other people [14] . Thus, the dissemination of fake news via social/mass media may 29 have an effect on people's social behavior. Social behavior changes can affect people's 30 well-being and resource distribution, resulting in changes in community resilience. 31 However, prior studies have rarely assessed community resilience via social media and 32 have rarely investigated the correlation between various types of news and tweets from 33 the community resilience's point of view. 34 Research Goal, Contributions, and Questions 35 In this work, we aim to quantify community resilience (CR) in terms of community 36 wellbeing (CW) and resource distribution (RD). These two factors are quantified by 37 natural language processing (NLP) tools on news articles that include real, mixed (i.e., 38 half fake and half real), and fake news as well as tweets including real and fake tweets. 39 We also examine the correlation between the measured CR from news articles and the 40 actual state of CR captured from tweets on Twitter. 41 In Fig. 1 , we illustrate our proposed framework for measuring community resilience 42 of various types of news/tweets using machine learning, natural language processing, 43 and dimension reduction techniques. 44 The key contributions of this work are as follows: 45 1. We develop novel community resilience metrics inspired by the system resilience 46 May 18, 2022 2/26 4. We analyze the correlation between measurements of CR attributes by each type of 66 news (i.e., real, mixed, or fake) and tweets (i.e., real or fake). From this analysis, 67 fake news is shown to influence people's behaviors towards undesirable states, 68 undermining CR in reality. Moreover, the CR measured based on real or mixed news 69 articles can reflect actual states of the CR measured from tweets. 70 5. We conduct a resilience analysis of various types of news (i.e., real, mixed, or fake) 71 and tweets (i.e., real and fake) via an output-oriented analysis to show the values of 72 each CR attribute over time, as well as a capacity-based analysis to demonstrate the 73 time-averaged CR measurements. We also conduct statistical analyses to examine 74 the correlation of CR attributes measured from news and tweets. 75 Our study will answer the following research questions: 76 1. What are the main trends observed in community resilience and its key attributes, 77 i.e., community wellbeing and resource distribution? 78 2. What are the key differences and correlations between the community resilience 79 measured on various types of news and tweets? 80 3. What are the levels of the community resilience metrics, e.g., absorption and recovery 81 during COVID-19 on various types of news and tweets? 82 Research Assumptions and Limitations 83 We conduct our study by assuming the following intuitions. First, real tweets/news can 84 represent community resilience better than mixed/fake tweets/news. Second, knowing a 85 current situation with accurate information can lead people to make more rational 86 decisions to handle a faced disaster, which is COVID-19 in this work. Although the 87 scope of this work is limited to measuring and analyzing community resilience using 88 tweets and news, further investigation to prove the above as the hypothesis will be 89 conducted in our future work. As no research work cannot be faultless, our work also 90 has a number of limitations: Community resilience (CR) refers to the ability of a social system to absorb the impact 107 of the stress and cope with threats and adapt to post-event situations by reorganizing, 108 changing, or learning to cope with the threat from the disasters [16, 17] . This definition 109 is well aligned with the general concept of system resilience in terms of its fault 110 tolerance (i.e., functioning under threats or errors), adaptability (i.e., adapting to 111 disruptions), and recoverability (i.e., recovering quickly from the disrupted 112 situations) [15] . Community resilience has been measured based on various types of 113 metrics [18, 19, 20] . CR can be defined differently depending on different disasters faced 114 in the past [21] . However, it has been commonly considered with a measure of resilience 115 whether a society functions in terms of social, economic, institutional, infrastructure, 116 community capital, and ecological aspects [22, 23] . Work [24] proposed the wellbeing theory discussing a measure of community 118 wellbeing in terms of positive emotions, engagement, relationships, meaning, and 119 accomplishment. [1] discussed 'health' in terms of behavioral, physical, social, and 120 environmental wellbeing. Higher psychological wellbeing can introduce higher 121 sustainability, equality, resilience, and inclusion [1, 24] . The key factors impacting 122 people's resilience to disasters were also studied, such as family distress, available 123 support systems, disruption of school/job programs, or loss of loved ones/property [25] . 124 The distribution state of physical and social resources is another indicator of (LIWC) framework [33, 34] , risk perception, negative emotions (e.g., sadness, anger, 146 anxiety), and behavioral responses (e.g., panic buying) to COVID-19 from the dataset 147 of Sina Weibo, Baidu search engine, and Ali e-commerce marketplace using LIWC [35] . 148 Aggressive panic buying behaviors were more prominently observed when more 149 misinformation or rumors on the COVID-19 were disseminated [35] . Emotions (e.g., 150 surprise, disgust, fear, anger, sadness, anticipation, joy, and trust) in replies were also 151 captured from real and false tweets using the National Research Council Canada 152 (NRC) [36] and LIWC [13] . Mingxuan et al. [37] measured people's mental health based 153 on emotions extracted from social media data, which was analyzed using machine 154 learning (ML) or NLP techniques [38, 39] . However, to our knowledge, no prior work has estimated community resilience based 156 on community wellbeing and resource distribution using both social media news articles 157 (i.e., real, mixed, and fake) and tweets (i.e., real and fake) to compare their 158 measurements and investigate their correlations. Measurement of Community Resilience Using Social 160 Media Information In this section, we discuss how community resilience is measured using social media 162 information, including both news articles and tweets. Community Resilience Metrics 164 We measure the community functionality in terms of community wellbeing and resource 165 distribution. Fig. 2 represents the community functionality, CF (t), with time t. We 166 define community resilience based on the concept of system resilience [15] , consisting of 167 absorption (i.e., fault tolerance), adaptability, and recoverability. We interpret the time 168 until a community does not function as the time period for absorption, namely TFA 169 (i.e., time from t 0 to t 1 ). Absorption (ABS) refers to the community's capacity to 170 absorb the shock and adverse effects caused by COVID-19. High TFA implies that the 171 community tolerates hardships introduced by a disaster so that the community can still 172 function by providing at least critical, minimum services, such as food, employment, 173 schools, or health services. Note that a higher absorption is more desirable. Community 174 Non-Functioning (CNF) is a term that refers to situations in which the community's 175 functionality falls below a critical threshold. We denote the deadlock functionality 176 threshold by b. We call the time from t 1 to t 3 the time under community 177 non-functioning (TNF). A shorter TNF is considered more desirable, representing fast 178 failure and fast recovery. By following the conventional concept of system reliability, the 179 mean time to recovery (MTTR), we defined the time to recovery (TTR) estimated from 180 the time the community reaches a critical functionality point (t 1 ) to the time it fully 181 recovers from the disaster and reaches at the initial normal state (t 4 ). Recovery (REF) 182 refers to the community's capacity to recover from COVID-19. The recoverability 183 effectiveness (RE) refers to how much the community has recovered from the minimum 184 functionality point, t 2 , to the current point at t 4 . Note that a higher level of recovery is 185 more desirable. We consider the whole period from the outbreak of a disaster (e.g., Higher absorption, recovery, and adaptability are more desirable, which means the more 190 area under the curve a community has, the more resilient it is. 191 We estimate CF (t) based on the levels of community wellbeing (CW (t)) and 192 resource distribution (RD(t)) at time t. Here, CR is measured by: scaling [40] . Function f in its simple form can be the average of CW and RD. However, 197 depending on the relative relevance of CR in a given domain, CW and RD can be 198 weighted differently. In order to improve the operationality of this work, we will explore 199 the appropriate f function. We will demonstrate that the incremental PCA function is 200 the best f function. To determine the average CF during the period of the COVID-19, we measure ABS, 202 CN F , and REF as follows: • ABS is the average CF during the time period for absorption, which is given by: • CNF is the average CF over the time under the critical area of CF , which is 205 measured by: We assume that a community is entirely dysfunctional when its CR is below the 207 threshold b. May 18, 2022 6/26 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) buying [41] . Wellbeing is measured by the extent of people's moods, such as anxiety, 241 depression, and anger, which have long been recognized as typical symptoms of 242 wellbeing illness [42, 43, 44] . Therefore, we obtain the extent of community wellbeing 243 from the features of anxiety, sadness, and anger, extracted from linguistic inquiry and 244 word count (LIWC) categories. Resource distribution (RD) also measures part of CR [3, 4, 26] where the high 247 functioning in RD refers to the high ability that a community can provide services to its 248 inhabitants related to economic, infrastructure, institutional, and community capital 249 resources. We assume that sufficient and well-distributed resources can contribute to 250 the community that can better resist, recover, and/or overcome a disaster. We measure 251 RD in terms of how well each service is provided. RD is measured by: May 18, 2022 7/26 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 25, 2022. ; Table 1 . The variance information ratio (Var), the reconstruction error (error), and the time-related correlations (Time corr) of Resource distribution and community resilience by using the polynomial (Poly) Kernel PCA, the Gaussian radial basis function (RBF) Kernel PCA, the sigmoid Kernel PCA, the cosine Kernel PCA, the incremental PCA, the linear PCA, the SVD, the isomap, and the Locally Linear Embedding. where EF (t) and CCF (t) refer to the level of states related to economic, and 253 community capital functioning, respectively, with an equal weight considered. Again, 254 depending on the domain requirement, its weight can be differently considered. As 255 discussed before, function f can be as simple as the average of EF (t), and CCF (t). However, in order to improve the operationalization of this work, We will demonstrate 257 that the incremental PCA function is the best f function. increased use of work-related (e.g., 'job,' 'majors,' 'xerox'), money-related (e.g., 'Audit,' 'cash,' or 'owe') terms in the LIWC categories. • Community capital indicates a community's ability to provide social activity services 267 to its inhabitants and build trust among them. We assess community capital in terms 268 of the language patterns representing community cooperation using the LIWC 269 categories as follows: consensus and cooperation [47] . Hence, we measure the frequency of words using 278 the 'first-person plural' pronounces and 'assent' in the LIWC categories. -Social Process-Related Communications: We measure increased social engagement 280 and cooperation [48, 49] We describe the process of finalizing information associated with news in Fig. 3 categorize news articles into real, mixed, or fake, as described in Table 2 . Using these 318 classifications, we collect all news articles from the archived news regarding COVID-19 319 from these organizations for Jan. 2020 -Jun. 2021. Processing of News Articles for Analysis 321 We extract 3,437 news articles tagged with COVID-19 and coronavirus. After 322 processing the initial cleaning, such as checking news with a correct tag, we come up 323 with 3,235 news, consisting of 360 real news, 207 mixed news, and 2,668 fake news. After eliminating repetitive or irrelevant news, we select 207 news at random out of each 325 pool of different types of news for fair consideration. Table 3 provides the distribution 326 of published news and tweets considered across months. As in Table 3 , we observe a 327 significant amount of news articles published in Mar./Apr. 2020 and prominently there 328 is a higher amount of fake news and tweets compared to those of real counterparts. The news sources are mainly newspaper interviews, TV interviews, viral images, We first classify tweets as real or fake. We first train eight existing ML classifiers on the 349 datasets described in [59] , which contain 23,481 fake tweets and 21,417 real news 350 articles. We then select the top three ML classifiers, i.e., Passive-Aggressive, Decision Tree, and AdaBoost based on their prediction performance, as shown in Table 4 . Finally, we predict the truthfulness of each tweet using these three ML algorithms and 353 determine the final prediction for each tweet based on the majority rule of the three ML 354 classifiers (i.e., at least two ML classifiers should give the same prediction result). . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 25, 2022. ; https://doi.org/10.1101/2022.05.23.22275454 doi: medRxiv preprint Subject (an amount of news) Real news politics (87), medical (29) , fauxtography (17) , entertainment (13), business (12) , viral (5), phenomena (5), crime (5), history (5), health (5) Mixed news coronavirus (67), politics (48) , health (32), facebook (29) , public (19) , medical (17) tweets. This is because when a community is threatened due to the impact introduced 400 by a disaster, people are more likely to cooperate for survival. The incremental PCA method calculates the resource distribution based on resource distribution, and finally, community resilience, measured using real, mixed, and 428 fake news as well as real and fake tweets. 429 We observe from Fig. 9 that fake news is in a better state of community wellbeing 430 (CW). In other words, released fake news implies that CW is adequate and likely 431 underestimates the detrimental effect of the COVID-19. Additionally, people's 432 communication via fake tweets demonstrates a significant level of isolation, whereas real 433 tweets show a higher level of community capital. Fig. 9 shows that while fake news 434 presents a high degree of economic resilience, real news shows a low degree of economic 435 resilience under the COVID-19. A possible reason is that fake news can trigger panic 436 buying, thus eroding economic resilience. Similarly, fake news has a greater level of 437 resource distribution than real news. Finally, fake news shows higher CR than real news. 438 Fake news has the potential to mislead people into taking inappropriate actions in TTR 17 17 17 9 15 17 17 17 9 17 17 17 17 9 17 17 17 17 9 17 Note that TFA, TNF, and TTR refer to the month-based average values. during which the community is non-functioning ranges from 0 to 17 months, depending 460 on the critical threshold level. For example, TNF is equal to 17 months when b = 0.5 for 461 real news, which means that the community functionality from the perspective of real 462 news is less than 0.5 for all 17 months. Understandably, as the critical threshold level 463 increases, the time duration associated with community dysfunction and recovery 464 increases, while that associated with absorption decreases. On the other hand, mixed 465 news has a higher level of absorption than fake news. Both fake news and mixed news 466 show a higher level of absorption than that of real news. This implies that the level of 467 community functionality is initially high and gradually declines, whereas real news 468 demonstrates a rapid decline in community functionality at the start. Therefore, we can 469 conclude that mixed/fake news tends to underestimate the negative impact of Table 11 shows the findings from our statistical analyses on the correlation between 1 1 17 10 17 1 1 17 10 17 1 1 17 10 17 0 1 16 10 0 TNF 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 2 TTR 17 17 1 8 1 17 17 1 8 1 17 17 1 8 1 17 17 1 8 17 Note that TFA, TNF, and TTR refer to the month-based average values. Note that TFA, TNF, and TTR refer to the month-based average values. TNF 10 12 0 6 0 17 14 0 17 2 17 15 0 17 4 17 16 1 17 6 TTR 17 16 1 17 1 17 17 1 17 17 17 17 1 17 17 17 17 1 17 17 Note that TFA, TNF, and TTR refer to the month-based average values. Note that and mean following or not following the same distribution, respectively. coefficient assumes that both x and y are normally distributed. When this assumption 499 does not hold, we rely on a non-parametric approach, such as Kendall tau correlation, 500 which does not make any assumption about distribution. According to Table 11 , fake 501 tweets and news have a positive correlation for resilience-related features with a 502 probability of 80 percent. Pearson and Kendall tau correlations (PC and KC) indicate 503 that the correlations between fake news and real tweets are negative, with a probability 504 of 80 percent. We also found that mixed news negatively correlates with real and fake 505 tweets across all types of CR attributes with a probability of 95 percent. Parametric 506 and non-parametric statistical hypothesis tests (PT and NT) demonstrate the 507 distribution's similarity across multiple scenarios. Fig. 10 illustrates the 508 Quantile-Quantile (Q-Q)-plot for community resilience in relation to various news types 509 (i.e., real, mixed, or fake) and tweet types (i.e., real or fake). We observe that fake 510 tweets and real tweets exhibit similarity in their distributions with the probability of 60 511 percent. This similarity implies that both tweets can properly reflect the actual states 512 of community resilience (CR) regardless of their truthfulness. Furthermore, analyzing 513 social media information and predicting CR can provide a useful indicator to measure 514 how our community is functioning against a disaster such as COVID-19. May 18, 2022 18/26 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) We summarize the findings obtained from the discussion above as follows: • Based on fake news, the public may believe that the community is resilient, which is 518 not the case. Additionally, the results indicate that fake news shares the same 519 viewpoint. They underestimate COVID-19's adverse effects and demonstrate a higher 520 level of resilience than that measured by real news. This perspective prolongs the 521 time required for actual complete recovery. Further, based on this finding, we observe 522 that fake news is not always pessimistic or negative. • From community resilience point of view, mixed news is more optimistic than real 524 news showing higher resilience. This may be because mixed news contains fake news, 525 which underestimates the impact of COVID-19. • Compared to propagated fake tweets, propagated fake news is more unrealistic from 527 community resilience point of view. Answer. Among the PCAs with various kernel types, the SVD, the isomap, and the 566 Locally Linear Embedding, we used the incremental PCA to integrate dimensions of 567 resource distribution and community resilience due to the higher level of variance 568 information ratio and the preservation of temporal dependency information. In 569 September 2020, CW reached its peak in fake tweets and real/fake news. The peaks of 570 CW in real tweets and mixed news, on the other hand, occur in February 2020 and June 571 2021, respectively. Additionally, we observe that CW reaches a low point by the end of 572 2020 when real tweets are used. Plus, the findings suggest that the resource distribution 573 trends observed in mixed/fake news and real/fake tweets are comparable to those 574 observed in community capital. On the other hand, the trend in real news about 575 resource distribution corresponds to the trend in economic functionality. Take note that 576 both real and fake news follow the same pattern in terms of resource distribution and 577 community wellbeing. Community resilience trends in real/fake news and real/fake 578 tweets are comparable to resource distribution trends. On the other hand, trends in real, 579 mixed, and fake news regarding resource distribution are similar to the trends in 580 community wellbeing. Fake news has a more even distribution of resources than real 581 news. Finally, fake news has a higher community resilience than real news. By creating 582 unrealistic optimism about the future, fake news has the potential to mislead people 583 into taking inappropriate actions in response to the COVID-19. RQ2. What are the key differences and correlations between the community resilience 585 measured on various types of news and tweets? Answer. According to the findings, fake tweet articles have an 80% probability of 587 correlating positively with fake news for resilience-related characteristics. Additionally, 588 Pearson and Kendall tau correlations indicate that the correlation between fake news 589 and real tweets is negative, with an 80 percent probability. Additionally, we discovered 590 that mixed news has a 95% probability of negatively correlating with real and fake 591 tweets across all types of CR attributes. Statistical hypothesis tests, both parametric 592 and non-parametric, demonstrate the distribution's similarity across multiple scenarios. 593 We observe that fake and real tweets have a 60% probability of having similar 594 distributions. This implies that fake tweets can accurately reflect the actual state of 595 community resilience (CR), regardless of their veracity. . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) We suggest the following future research directions. 612 First, in this work, we only used Twitter to gather all real and fake news to 613 investigate the behavior of the population. One should be extremely careful in 614 analyzing social media information. Surveys can provide high-quality data for analyzing 615 population behavior, albeit at a cost. Additional research can be conducted to examine 616 the correlations between fake/real news and survey responses. Nonetheless, considering 617 additional social media platforms for future research may be beneficial. Second, while we propose in this work to quantify community resilience using social 619 media data (e.g., tweets), we are still in the first phase of this journey, namely 620 enhancing community resilience. The literature does not include a thorough 621 examination of the prediction models for community resilience. As a result, more 622 sophisticated models are required to forecast how distinct communities will respond to a 623 variety of events and epidemics. It specifically calls for developing a multi-agent model 624 that accounts for the spread of fake news. The approach described in this work must be 625 extended further to validate the model. The next step on this path is to predict 626 output-oriented community resilience using machine and deep learning techniques. Ruzek Disaster response, mental health, and community resilience Available at Building resilient communities: The importance of integrating mental health and wellbeing in effective development thinking and practice What do we mean by'community resilience' ? A systematic literature review of how it is defined in the literature PLoS Currents Using Twitter for demographic and social science research: Tools for data collection and processing Sociological methods & research Identifying and understanding communities using Twitter to connect about depression: cross-sectional study JMIR mental health Using twitter for demographic and social science research: Tools for data collection WASHINGTON UNIV SEATTLE Tien and Deligiannis, Nikos Twitter data analysis for studying communities of practice in the media industry Fact-checking effect on viral hoaxes: A model of misinformation spread in The spread of true and false news online Science STRAM: Measuring the trustworthiness of computer-based systems ACM and others =COPEWELL: a conceptual framework and system dynamics model for predicting community functioning and resilience after disasters Disaster Medicine and Public Health Preparedness A place-based model for understanding community resilience to natural disasters Global Environmental Change Khankeh Community disaster resilience: a systematic review on assessment models and tools PLOS Currents Disasters Measuring capacities for community resilience Social Indicators Research A synthesis of disaster resilience measurement methods and indices Emrich Disaster resilience indicators for bench-marking baseline conditions Building community resilience through mental health infrastructure and training in post-Katrina New Orleans Ethnicity & Disease Inter-Agency Standing Committee and others IASC guidelines on mental health and psychosocial support in emergency settings Critical analysis of the relationship between local ownership and community resiliency Rural Sociology Mental health strategies to combat the psychological impact of COVID-19 beyond paranoia and panic COVID-19: Emerging compassion, courage and resilience in the face of misinformation and adversity COVID-19) infodemic and emerging issues through a data lens: The case of China International Journal of Environmental Research and Public Health and others Using social and behavioural science to support COVID-19 pandemic response and others The role of social media as psychological first aid as a support to community resilience building The Australian Journal of Emergency Management Text-mining-based Fake News Detection Using Ensemble Methods International Journal of Automation and Computing The impact of COVID-19 epidemic declaration on psychological consequences: a study on active WEIBO users The psychological meaning of words: LIWC and computerized text analysis methods Assessment of public attention, risk perception, emotional and behavioural responses to the COVID-19 outbreak: social media surveillance in China medRxiv Sentiment and emotion Lexicons National Research Council Canada (NRC) Available at Emotion: Disentangled Representation Learning for Emotion Analysis on Social Media to Improve Community Resilience in the COVID-19 Era and Beyond Web Science Quantifying mental health signals in Twitter Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality Community resilience and social media: Remote and rural First Nations communities, social isolation and cultural preservation International Rural Network Forum. Whyalla and Upper Spencer Gulf Data mining: Concepts and techniques A No health without mental health The Lancet The impact of sheltering in place during the COVID-19 pandemic on older adults' social and mental well-being Social media use and well-being: A prospective experience-sampling study Diego The impact of the COVID-19 pandemic on subjective mental well-being: The interplay of perceived threat, future anxiety and resilience Personality and Individual Differences Personality in its natural habitat: Manifestations and implicit folk theories of personality in daily life Pronouns in marital interaction: What do you and I say about marital health? Psychological Science Analyzing cockpit communications: The links between language, performance, error, and workload Human Performance in Extreme Environments Emmanuel and others The collegial phenomenon: The social mechanisms of cooperation among peers in a corporate law partnership Gender differences in language use: An analysis of 14,000 text samples Discourse Processes Liping Distance weighted cosine similarity measure for text classification International Conference on Intelligent Data Engineering and Automated Learning Internet Live Stats Twitter usage statistics Twitter Twitter usage/company facts Detecting opinion spams and fake news using text classification Security and Privacy