The minute-scale dynamics of online emotions reveal the effects of affect labeling

Criticism, Compassion, and Conspiracy Theories: A Thematic Analysis of What Twitter Users Are Saying About COVID-19 in Correctional Settings

We examined Twitter data using thematic analysis to understand public perceptions of the impact of COVID-19 on incarcerated people and reactions to including incarcerated populations in the early phases of the vaccine rollout. Our findings from n = 513 Tweets yielded six themes: Twitter as usual, Advocacy, Deserve to suffer, Vaccine priority debate, Inadequate response, and Misinformation. Stigma-laden statements cut across themes, highlighting the role pathologizing beliefs play in forming opinions about incarcerated people in public health crises. Trust of government response and buy-in to public health communication are positively associated with adherence to guidelines. Although public health decisions are derived from logic and research, our findings indicate that public perception may be driven by personal morals and stigma associated with justice-involved individuals. We recommend that attention be turned toward effective policy messaging, and use of social media, to increase trust and decrease stigma that tends to dominate societal perception.

Negative affect variability differs between anxiety and depression on social media

OBJECTIVE

Negative affect variability is associated with increased symptoms of internalizing psychopathology (i.e., depression, anxiety). The Contrast Avoidance Model (CAM) suggests that individuals with anxiety avoid negative emotional shifts by maintaining pathological worry. Recent evidence also suggests that the CAM can be applied to major depression and social phobia, both characterized by negative affect changes. Here, we compare negative affect variability between individuals with a variety of anxiety and depression diagnoses by measuring the levels and degree of change in the sentiment of their online communications.

METHOD

Participants were 1,853 individuals on Twitter who reported that they had been clinically diagnosed with an anxiety disorder (A cohort, n = 896) or a depressive disorder (D cohort, n = 957). Mean negative affect (NA) and negative affect variability were calculated using the Valence Aware Dictionary for Sentiment Reasoning (VADER), an accurate sentiment analysis tool that scores text in terms of its negative affect content.

RESULTS

Findings showed differences in negative affect variability between the D and A cohort, with higher levels of NA variability in the D cohort than the A cohort, U = 367210, p < .001, r = 0.14, d = 0.25. Furthermore, we found that A and D cohorts had different average NA, with the D cohort showing higher NA overall, U = 377368, p < .001, r = 0.12, d = 0.21.

LIMITATIONS

Our sample is limited to individuals who disclosed their diagnoses online, which may involve bias due to self-selection and stigma. Our sentiment analysis of online text may not completely capture all nuances of individual affect.

CONCLUSIONS

Individuals with depression diagnoses showed a higher degree of negative affect variability compared to individuals with anxiety disorders. Our findings support the idea that negative affect variability can be measured using computational approaches on large-scale social media data and that social media data can be used to study naturally occurring mental health effects at scale.

LEIA: Linguistic Embeddings for the Identification of Affect

The wealth of text data generated by social media has enabled new kinds of analysis of emotions with language models. These models are often trained on small and costly datasets of text annotations produced by readers who guess the emotions expressed by others in social media posts. This affects the quality of emotion identification methods due to training data size limitations and noise in the production of labels used in model development. We present LEIA, a model for emotion identification in text that has been trained on a dataset of more than 6 million posts with self-annotated emotion labels for happiness, affection, sadness, anger, and fear. LEIA is based on a word masking method that enhances the learning of emotion words during model pre-training. LEIA achieves macro-F1 values of approximately 73 on three in-domain test datasets, outperforming other supervised and unsupervised methods in a strong benchmark that shows that LEIA generalizes across posts, users, and time periods. We further perform an out-of-domain evaluation on five different datasets of social media and other sources, showing LEIA's robust performance across media, data collection methods, and annotation schemes. Our results show that LEIA generalizes its classification of anger, happiness, and sadness beyond the domain it was trained on. LEIA can be applied in future research to provide better identification of emotions in text from the perspective of the writer.

Unveiling the dynamics of emotions in society through an analysis of online social network conversations

Social networks can provide insights into the emotions expressed by a society. However, the dynamic nature of emotions presents a significant challenge for policymakers, politicians, and communication professionals who seek to understand and respond to changes in emotions over time. To address this challenge, this paper investigates the frequency, duration, and transition of 24 distinct emotions over a 2-year period, analyzing more than 5 million tweets. The study shows that emotions with lower valence but higher dominance and/or arousal are more prevalent in online social networks. Emotions with higher valence and arousal tend to last longer, while dominant emotions tend to have shorter durations. Emotions occupying the conversations predominantly inhibit others with similar valence and dominance, and higher arousal. Over a month, emotions with similar valences tend to prevail in online social network conversations.

Text as signal. A tutorial with case studies focusing on social media (Twitter)

Sentiment analysis is the automated coding of emotions expressed in text. Sentiment analysis and other types of analyses focusing on the automatic coding of textual documents are increasingly popular in psychology and computer science. However, the potential of treating automatically coded text collected with regular sampling intervals as a signal is currently overlooked. We use the phrase "text as signal" to refer to the application of signal processing techniques to coded textual documents sampled with regularity. In order to illustrate the potential of treating text as signal, we introduce the reader to a variety of such techniques in a tutorial with two case studies in the realm of social media analysis. First, we apply finite response impulse filtering to emotion-coded tweets posted during the US Election Week of 2020 and discuss the visualization of the resulting variation in the filtered signal. We use changepoint detection to highlight the important changes in the emotional signals. Then we examine data interpolation, analysis of periodicity via the fast Fourier transform (FFT), and FFT filtering to personal value-coded tweets from November 2019 to October 2020 and link the variation in the filtered signal to some of the epoch-defining events occurring during this period. Finally, we use block bootstrapping to estimate the variability/uncertainty in the resulting filtered signals. After working through the tutorial, the readers will understand the basics of signal processing to analyze regularly sampled coded text.

Qualitative similarities and distinctions between participants' experiences with a yoga intervention and an attention control

PURPOSE

This manuscript aims to compare and contrast acceptability and perceived benefits of yoga-skills training (YST) and an empathic listening attention control (AC) in the Pro-You study, a randomized pilot trial of YST vs. AC for adults receiving chemotherapy infusions for gastrointestinal cancer.

METHODS

Participants were invited for a one-on-one interview at week 14 follow-up, after completing all intervention procedures and quantitative assessments. Staff used a semi-structured guide to elicit participants' views on study processes, the intervention they received, and its effects. Qualitative data analysis followed an inductive/deductive approach, inductively identifying themes and deductively guided by social cognitive theory.

RESULTS

Some barriers (e.g., competing demands, symptoms), facilitators (e.g., interventionist support, the convenience of clinic-based delivery), and benefits (e.g., decreased distress and rumination) were common across groups. YST participants uniquely described the importance of privacy, social support, and self-efficacy for increasing engagement in yoga. Benefits specific to YST included positive emotions and greater improvement in fatigue and other physical symptoms. Both groups described some self-regulatory processes, but through different mechanisms: self-monitoring in AC and the mind-body connection in YST.

CONCLUSIONS

This qualitative analysis demonstrates that participant experiences in a yoga-based intervention or an AC condition reflect social cognitive and mind-body frameworks of self-regulation. Findings can be used to develop yoga interventions that maximize acceptability and effectiveness and to design future research that elucidates the mechanisms by which yoga is efficacious.

Quantifying changes in societal optimism from online sentiment

Individuals can hold contrasting views about distinct times: for example, dread over tomorrow's appointment and excitement about next summer's vacation. Yet, psychological measures of optimism often assess only one time point or ask participants to generalize about their future. Here, we address these limitations by developing the optimism curve, a measure of societal optimism that compares positivity toward different future times that was inspired by the Treasury bond yield curve. By performing sentiment analysis on over 3.5 million tweets that reference 23 future time points (2 days to 30 years), we measured how positivity differs across short-, medium-, and longer-term future references. We found a consistent negative association between positivity and the distance into the future referenced: From August 2017 to February 2020, the long-term future was discussed less positively than the short-term future. During the COVID-19 pandemic, this relationship inverted, indicating declining near-future- but stable distant-future-optimism. Our results demonstrate that individuals hold differentiated attitudes toward the near and distant future that shift in aggregate over time in response to external events. The optimism curve uniquely captures these shifting attitudes and may serve as a useful tool that can expand existing psychometric measures of optimism.

Natural Language Content Mediates the Association Between Active Interactions on Social Network Services and Subjective Well-Being

Previous studies indicated that active interactions on social networking services (SNS) are positively linked to subjective well-being (SWB). However, how semantic SNS content affects the association between the degree of SNS interaction and SWB has not been investigated. We addressed this issue by conducting a mediation analysis using natural language processing. We first analyzed Twitter data and SWB scores from 217 participants and found that the degree of active interactions on Twitter (i.e., frequency of reply) was positively correlated with SWB. Next, our multivariate mediation analysis demonstrated that positive words served as SWB-promoting mechanisms for highly interactive people, whereas worrying words led to lower SWB for less interactive people, but negative words did not. This study revealed that natural language content explains why individuals who are highly interactive on SNS have higher SWB, whereas less interactive individuals show lower SWB.

Large-Scale Social Media Analysis Reveals Emotions Associated with Nonmedical Prescription Drug Use

Background

The behaviors and emotions associated with and reasons for nonmedical prescription drug use (NMPDU) are not well-captured through traditional instruments such as surveys and insurance claims. Publicly available NMPDU-related posts on social media can potentially be leveraged to study these aspects unobtrusively and at scale.

Methods

We applied a machine learning classifier to detect self-reports of NMPDU on Twitter and extracted all public posts of the associated users. We analyzed approximately 137 million posts from 87,718 Twitter users in terms of expressed emotions, sentiments, concerns, and possible reasons for NMPDU via natural language processing.

Results

Users in the NMPDU group express more negative emotions and less positive emotions, more concerns about family, the past, and body, and less concerns related to work, leisure, home, money, religion, health, and achievement compared to a control group (i.e., users who never reported NMPDU). NMPDU posts tend to be highly polarized, indicating potential emotional triggers. Gender-specific analyses show that female users in the NMPDU group express more content related to positive emotions, anticipation, sadness, joy, concerns about family, friends, home, health, and the past, and less about anger than males. The findings are consistent across distinct prescription drug categories (opioids, benzodiazepines, stimulants, and polysubstance).

Conclusion

Our analyses of large-scale data show that substantial differences exist between the texts of the posts from users who self-report NMPDU on Twitter and those who do not, and between males and females who report NMPDU. Our findings can enrich our understanding of NMPDU and the population involved.

Global evidence of expressed sentiment alterations during the COVID-19 pandemic

The COVID-19 pandemic has created unprecedented burdens on people's physical health and subjective well-being. While countries worldwide have developed platforms to track the evolution of COVID-19 infections and deaths, frequent global measurements of affective states to gauge the emotional impacts of pandemic and related policy interventions remain scarce. Using 654 million geotagged social media posts in over 100 countries, covering 74% of world population, coupled with state-of-the-art natural language processing techniques, we develop a global dataset of expressed sentiment indices to track national- and subnational-level affective states on a daily basis. We present two motivating applications using data from the first wave of COVID-19 (from 1 January to 31 May 2020). First, using regression discontinuity design, we provide consistent evidence that COVID-19 outbreaks caused steep declines in expressed sentiment globally, followed by asymmetric, slower recoveries. Second, applying synthetic control methods, we find moderate to no effects of lockdown policies on expressed sentiment, with large heterogeneity across countries. This study shows how social media data, when coupled with machine learning techniques, can provide real-time measurements of affective states.

How Does the World View China’s Carbon Policy? A Sentiment Analysis on Twitter Data

China has recently put forth an ambitious plan to achieve carbon peak around 2030 and carbon neutrality around 2060. However, there are quite a few differences regarding the public views about China’s carbon policy between the Chinese people and the people from other countries, especially concerning the doubt of foreign people about the fidelity of China’s carbon policy goals. Based on Twitter data related to China’s carbon policy topics from 2008 to 2020, this study shows the inter- and intra-annual trends in the count of tweets about China’s carbon policy, conducts sentiment analysis, extracts top frequency words from different attitudes, and analyzes the impact of China’s official Twitter accounts on the global view of China’s carbon policy. Our results show: (1) the global attention to China’s carbon policy gradually rises and occasionally rises suddenly due to important carbon events; (2) the proportion of Twitter users with negative sentiment about China’s carbon policy has increased rapidly and has exceeded the proportion of Twitter users with positive sentiment since 2019; (3) people in developing countries hold more positive or neutral attitudes towards China’s carbon policy, while developed countries hold more negative attitudes; (4) China’s official Twitter accounts serve to improve the global views on China’s carbon policy.

What drives people to repost social media messages during the COVID-19 pandemic? Evidence from the Weibo news microblog

COVID-19 poses an unprecedented challenge to human society. To cope with the pandemic, people seek information from various communication channels. Microblog websites are highly influential information channels for the public to get timely information during the pandemic. Building on the heuristic-systematic processing model, this study identifies the multiple characteristics (content, author, and social features) that may play a role in triggering long cascades of reposts of COVID-19-related news microblogs. With a large-scale news microblog database collected from Weibo and an innovative information gain method, we find that heuristic thinking plays a dominant role in COVID-19 pandemic-related news microblog reposting decisions and further discloses the specific influencing factors of such behavior.

The dynamics of affect across the wake-sleep cycle: From waking mind-wandering to night-time dreaming

Affective experiences occur across the wake-sleep cycle-from active wakefulness to resting wakefulness (i.e., mind-wandering) to sleep (i.e., dreaming). Yet, we know little about the dynamics of affect across these states. We compared the affective ratings of waking, mind-wandering, and dream episodes. Results showed that mind-wandering was more positively valenced than dreaming, and that both mind-wandering and dreaming were more negatively valenced than active wakefulness. We also compared participants' self-ratings of affect with external ratings of affect (i.e., analysis of affect in verbal reports). With self-ratings all episodes were predominated by positive affect. However, the affective valence of reports changed from positively valenced waking reports to affectively balanced mind-wandering reports to negatively valenced dream reports. These findings show that (1) the positivity bias characteristic to waking experiences decreases across the wake-sleep continuum, and (2) conclusions regarding affective experiences depend on whether self-ratings or verbal reports describing these experiences are analysed.

Exploring the sentiment of entrepreneurs on Twitter

Sentiment analysis is an evolving field of study that employs artificial intelligence techniques to identify the emotions and opinions expressed in a given text. Applying sentiment analysis to study the billions of messages that circulate in popular online social media platforms has raised numerous opportunities for exploring the emotional expressions of their users. In this paper we combine sentiment analysis with natural language processing and topic analysis techniques and conduct two different studies to examine whether engagement in entrepreneurship is associated with more positive emotions expressed on Twitter. In study 1, we investigate three samples with 6.717.308, 13.253.244, and 62.067.509 tweets respectively. We find that entrepreneurs express more positive emotions than non-entrepreneurs for most topics. We also find that social entrepreneurs express more positive emotions, and that serial entrepreneurs express less positive emotions than other entrepreneurs. In study 2, we use 21.491.962 tweets to explore 37.225 job-status changes by individuals who entered or quit entrepreneurship. We find that a job change to entrepreneurship is associated with a shift in the expression of emotions to more positive ones.

Emotion Naming Impedes Both Cognitive Reappraisal and Mindful Acceptance Strategies of Emotion Regulation

Friends and therapists often encourage people in distress to say how they feel (i.e., name their emotions) with the hope that identifying their emotions will help them cope. Although lay and some psychological theories posit that emotion naming should facilitate subsequent emotion regulation, there is little research directly testing this question. Here, we report on two experimental studies that test how naming the emotions evoked by aversive images impacts subsequent regulation of those emotions. In study 1 (N = 80), participants were randomly assigned into one of four between-subjects conditions in which they either (i) passively observed aversive images, (ii) named the emotions that these images made them feel, (iii) regulated their emotions by reappraising the meaning of images, or (iv) both named and regulated their emotions. Analyses of self-reported negative affect revealed that emotion naming impeded emotion regulation via reappraisal. Participants who named their emotions before reappraising reported feeling worse than those who regulated without naming. Study 2 (N = 60) replicated these findings in a within-participants design, demonstrated that emotion naming also impeded regulation via mindful acceptance, and showed that observed effects were unrelated to a measure of social desirability, thereby mitigating the concern of experimenter demand. Together, these studies show that the impact of emotion naming on emotion regulation opposes common intuitions: instead of facilitating emotion regulation via reappraisal or acceptance, constructing an instance of a specific emotion category by giving it a name may "crystalize" one's affective experience and make it more resistant to modification.

Supplementary Information

The online version contains supplementary material available at 10.1007/s42761-021-00036-y.

Twitter, time and emotions

The study of temporal trajectories of emotions shared in tweets has shown that both positive and negative emotions follow nonlinear circadian (24 h) and circaseptan (7-day) patterns. But to this point, such findings could be instrument-dependent as they rely exclusively on coding using the Linguistic Inquiry Word Count. Further, research has shown that self-referential content has higher relevance and meaning for individuals, compared with other types of content. Investigating the specificity of self-referential material in temporal patterns of emotional expression in tweets is of interest, but current research is based upon generic textual productions. The temporal variations of emotions shared in tweets through emojis have not been compared to textual analyses to date. This study hence focuses on several comparisons: (i) between Self-referencing tweets versus Other topic tweets, (ii) between coding of textual productions versus coding of emojis, and finally (iii) between coding of textual productions using different sentiment analysis tools (the Linguistic Inquiry and Word Count-LIWC; the Valence Aware Dictionary and sEntiment Reasoner-VADER and the Hu Liu sentiment lexicon-Hu Liu). In a collection of more than 7 million Self-referencing and close to 18 million Other topic content-coded tweets, we identified that (i) similarities and differences in terms of shape and amplitude can be observed in temporal trajectories of expressed emotions between Self-referring and Other topic tweets, (ii) that all tools feature significant circadian and circaseptan patterns in both datasets but not always, and there is often a correspondence in the shape of circadian and circaseptan patterns, and finally (iii) that circadian and circaseptan patterns obtained from the coding of emotional expression in emojis sometimes depart from those of the textual analysis, indicating some complementarity in the use of both modes of expression. We discuss the implications of our findings from the perspective of the literature on emotions and well-being.

Individuals with depression express more distorted thinking on social media

Depression is a leading cause of disability worldwide, but is often underdiagnosed and undertreated. Cognitive behavioural therapy holds that individuals with depression exhibit distorted modes of thinking, that is, cognitive distortions, that can negatively affect their emotions and motivation. Here, we show that the language of individuals with a self-reported diagnosis of depression on social media is characterized by higher levels of distorted thinking compared with a random sample. This effect is specific to the distorted nature of the expression and cannot be explained by the presence of specific topics, sentiment or first-person pronouns. This study identifies online language patterns that are indicative of depression-related distorted thinking. We caution that any future applications of this research should carefully consider ethical and data privacy issues.

Depression alters the circadian pattern of online activity

Human sleep/wake cycles follow a stable circadian rhythm associated with hormonal, emotional, and cognitive changes. Changes of this cycle are implicated in many mental health concerns. In fact, the bidirectional relation between major depressive disorder and sleep has been well-documented. Despite a clear link between sleep disturbances and subsequent disturbances in mood, it is difficult to determine from self-reported data which specific changes of the sleep/wake cycle play the most important role in this association. Here we observe marked changes of activity cycles in millions of twitter posts of 688 subjects who explicitly stated in unequivocal terms that they had received a (clinical) diagnosis of depression as compared to the activity cycles of a large control group (n = 8791). Rather than a phase-shift, as reported in other work, we find significant changes of activity levels in the evening and before dawn. Compared to the control group, depressed subjects were significantly more active from 7 PM to midnight and less active from 3 to 6 AM. Content analysis of tweets revealed a steady rise in rumination and emotional content from midnight to dawn among depressed individuals. These results suggest that diagnosis and treatment of depression may focus on modifying the timing of activity, reducing rumination, and decreasing social media use at specific hours of the day.

Charting the development of emotion comprehension and abstraction from childhood to adulthood using observer-rated and linguistic measures

This study examined two facets of emotion development: emotion word comprehension (knowing the meaning of emotion words such as "anger" or "excitement") and emotion concept abstraction (representing emotions in terms of internal psychological states that generalize across situations). Using a novel emotion vocabulary assessment, we captured how a cross-sectional sample of participants aged 4-25 (N = 196) defined 24 emotions. Smoothing spline regression models suggested that emotion comprehension followed an emergent shape: Knowledge of emotion words increased across childhood and plateaued around age 11. Human coders rated the abstractness of participants' responses, and these ratings also followed an emergent shape but plateaued significantly later than comprehension, around age 18. An automated linguistic analysis of abstractness supported coders' perceptions of increased abstractness across age. Finally, coders assessed the definitional strategies participants used to describe emotions. Young children tended to describe emotions using concrete strategies such as providing example situations that evoked those emotions or by referring to physiological markers of emotional experiences. Whereas use of these concrete strategies decreased with age, the tendency to use more abstract strategies such as providing general definitions that delineated the causes and characteristics of emotions or by providing synonyms of emotion words increased with age. Overall, this work (a) provides a tool for assessing definitions of emotion terms, (b) demonstrates that emotion concept abstraction increases across age, and (c) suggests that adolescence is a period in which emotion words are comprehended but their level of abstraction continues to mature. (PsycInfo Database Record (c) 2020 APA, all rights reserved).

Mining Social Media Data for Biomedical Signals and Health-Related Behavior

Social media data have been increasingly used to study biomedical and health-related phenomena. From cohort-level discussions of a condition to population-level analyses of sentiment, social media have provided scientists with unprecedented amounts of data to study human behavior associated with a variety of health conditions and medical treatments. Here we review recent work in mining social media for biomedical, epidemiological, and social phenomena information relevant to the multilevel complexity of human health. We pay particular attention to topics where social media data analysis has shown the most progress, including pharmacovigilance and sentiment analysis, especially for mental health. We also discuss a variety of innovative uses of social media data for health-related applications as well as important limitations of social media data access and use.

Gradual positive and negative affect induction: The effect of verbalizing affective content

In this paper, we study the effect of verbalizing affective pictures on affective state and language production. Individuals describe (Study I: Spoken Descriptions of Pictures) or passively view (Study II: Passively Viewing Pictures) 40 pictures for the International Affective Picture System (IAPS) that gradually increase from neutral to either positive or negative content. We expected that both methods would result in successful affect induction, and that the effect would be stronger for verbally describing pictures than for passively viewing them. Results indicate that speakers indeed felt more negative after describing negative pictures, but that describing positive (compared to neutral) pictures did not result in a more positive state. Contrary to our hypothesis, no differences were found between describing and passively viewing the pictures. Furthermore, we analysed the verbal picture descriptions produced by participants on various dimensions. Results indicate that positive and negative pictures were indeed described with increasingly more affective language in the expected directions. In addition to informing our understanding of the relationship between (spoken) language production and affect, these results also potentially pave the way for a new method of affect induction that uses free expression.