Predicting Depression From Language-Based Emotion Dynamics: Longitudinal Analysis of Facebook and Twitter Status Updates

Responsible development of clinical speech AI: Bridging the gap between clinical research and technology

This perspective article explores the challenges and potential of using speech as a biomarker in clinical settings, particularly when constrained by the small clinical datasets typically available in such contexts. We contend that by integrating insights from speech science and clinical research, we can reduce sample complexity in clinical speech AI models with the potential to decrease timelines to translation. Most existing models are based on high-dimensional feature representations trained with limited sample sizes and often do not leverage insights from speech science and clinical research. This approach can lead to overfitting, where the models perform exceptionally well on training data but fail to generalize to new, unseen data. Additionally, without incorporating theoretical knowledge, these models may lack interpretability and robustness, making them challenging to troubleshoot or improve post-deployment. We propose a framework for organizing health conditions based on their impact on speech and promote the use of speech analytics in diverse clinical contexts beyond cross-sectional classification. For high-stakes clinical use cases, we advocate for a focus on explainable and individually-validated measures and stress the importance of rigorous validation frameworks and ethical considerations for responsible deployment. Bridging the gap between AI research and clinical speech research presents new opportunities for more efficient translation of speech-based AI tools and advancement of scientific discoveries in this interdisciplinary space, particularly if limited to small or retrospective datasets.

The Role of Humanization and Robustness of Large Language Models in Conversational Artificial Intelligence for Individuals With Depression: A Critical Analysis

Unlabelled

Large language model (LLM)-powered services are gaining popularity in various applications due to their exceptional performance in many tasks, such as sentiment analysis and answering questions. Recently, research has been exploring their potential use in digital health contexts, particularly in the mental health domain. However, implementing LLM-enhanced conversational artificial intelligence (CAI) presents significant ethical, technical, and clinical challenges. In this viewpoint paper, we discuss 2 challenges that affect the use of LLM-enhanced CAI for individuals with mental health issues, focusing on the use case of patients with depression: the tendency to humanize LLM-enhanced CAI and their lack of contextualized robustness. Our approach is interdisciplinary, relying on considerations from philosophy, psychology, and computer science. We argue that the humanization of LLM-enhanced CAI hinges on the reflection of what it means to simulate "human-like" features with LLMs and what role these systems should play in interactions with humans. Further, ensuring the contextualization of the robustness of LLMs requires considering the specificities of language production in individuals with depression, as well as its evolution over time. Finally, we provide a series of recommendations to foster the responsible design and deployment of LLM-enhanced CAI for the therapeutic support of individuals with depression.

MHA: a multimodal hierarchical attention model for depression detection in social media

As a serious mental disease, depression causes great harm to the physical and mental health of individuals, and becomes an important cause of suicide. Therefore, it is necessary to accurately identify and treat depressed patients. Compared with traditional clinical diagnosis methods, a large amount of real and different types of data on social media provides new ideas for depression detection research. In this paper, we construct a depression detection data set based on Weibo, and propose a Multimodal Hierarchical Attention (MHA) model for social media depression detection. Multimodal data is fed into the model and the attention mechanism is applied within and between modalities at the same time. Experimental results show that the proposed model achieves the best classification performance. In addition, we propose a distribution normalization method, which can optimize the data distribution and improve the accuracy of depression detection.

Supervised machine learning models for depression sentiment analysis

Introduction

Globally, the prevalence of mental health problems, especially depression, is at an all-time high. The objective of this study is to utilize machine learning models and sentiment analysis techniques to predict the level of depression earlier in social media users' posts.

Methods

The datasets used in this research were obtained from Twitter posts. Four machine learning models, namely extreme gradient boost (XGB) Classifier, Random Forest, Logistic Regression, and support vector machine (SVM), were employed for the prediction task.

Results

The SVM and Logistic Regression models yielded the most accurate results when applied to the provided datasets. However, the Logistic Regression model exhibited a slightly higher level of accuracy compared to SVM. Importantly, the logistic regression model demonstrated the advantage of requiring less execution time.

Discussion

The findings of this study highlight the potential of utilizing machine learning models and sentiment analysis techniques for early detection of depression in social media users. The effectiveness of SVM and Logistic Regression models, with Logistic Regression being more efficient in terms of execution time, suggests their suitability for practical implementation in real-world scenarios.

Linguistic Methodologies to Surveil the Leading Causes of Mortality: Scoping Review of Twitter for Public Health Data

BACKGROUND

Twitter has become a dominant source of public health data and a widely used method to investigate and understand public health-related issues internationally. By leveraging big data methodologies to mine Twitter for health-related data at the individual and community levels, scientists can use the data as a rapid and less expensive source for both epidemiological surveillance and studies on human behavior. However, limited reviews have focused on novel applications of language analyses that examine human health and behavior and the surveillance of several emerging diseases, chronic conditions, and risky behaviors.

OBJECTIVE

The primary focus of this scoping review was to provide a comprehensive overview of relevant studies that have used Twitter as a data source in public health research to analyze users' tweets to identify and understand physical and mental health conditions and remotely monitor the leading causes of mortality related to emerging disease epidemics, chronic diseases, and risk behaviors.

METHODS

A literature search strategy following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) extended guidelines for scoping reviews was used to search specific keywords on Twitter and public health on 5 databases: Web of Science, PubMed, CINAHL, PsycINFO, and Google Scholar. We reviewed the literature comprising peer-reviewed empirical research articles that included original research published in English-language journals between 2008 and 2021. Key information on Twitter data being leveraged for analyzing user language to study physical and mental health and public health surveillance was extracted.

RESULTS

A total of 38 articles that focused primarily on Twitter as a data source met the inclusion criteria for review. In total, two themes emerged from the literature: (1) language analysis to identify health threats and physical and mental health understandings about people and societies and (2) public health surveillance related to leading causes of mortality, primarily representing 3 categories (ie, respiratory infections, cardiovascular disease, and COVID-19). The findings suggest that Twitter language data can be mined to detect mental health conditions, disease surveillance, and death rates; identify heart-related content; show how health-related information is shared and discussed; and provide access to users' opinions and feelings.

CONCLUSIONS

Twitter analysis shows promise in the field of public health communication and surveillance. It may be essential to use Twitter to supplement more conventional public health surveillance approaches. Twitter can potentially fortify researchers' ability to collect data in a timely way and improve the early identification of potential health threats. Twitter can also help identify subtle signals in language for understanding physical and mental health conditions.

Indirect Identification of Perinatal Psychosocial Risks from Natural Language

During the perinatal period, psychosocial health risks, including depression and intimate partner violence, are associated with serious adverse health outcomes for birth parents and children. To appropriately intervene, healthcare professionals must first identify those at risk, yet stigma often prevents people from directly disclosing the information needed to prompt an assessment. In this research we use short diary entries to indirectly elicit information that could indicate psychosocial risks, then examine patterns that emerge in the language of those at risk. We find that diary entries exhibit consistent themes, extracted using topic modeling, and emotional perspective, drawn from dictionary-informed sentiment features. Using these features, we use regularized regression to predict screening measures for depression and psychological aggression by an intimate partner. Journal text entries quantified through topic models and sentiment features show promise for depression prediction, corresponding with self-reported screening measures almost as well as closed-form questions. Text-based features are less useful in predicting intimate partner violence, but topic models generate themes that align with known risk correlates. The indirect features uncovered in this research could aid in the detection and analysis of stigmatized risks.

Analyzing User-Generated Web-Based Posts of Adolescents' Emotional, Behavioral, and Symptom Responses to Beliefs About Depression: Qualitative Thematic Analysis

BACKGROUND

Depression is common during adolescence. Early intervention can prevent it from developing into more progressive mental disorders. Combining information technology and clinical psychoeducation is a promising way to intervene at an earlier stage. However, data-driven research on the cognitive response to health information targeting adolescents with symptoms of depression is lacking.

OBJECTIVE

This study aimed to fill this knowledge gap through a new understanding of adolescents' cognitive response to health information about depression. This knowledge can help to develop population-specific information technology, such as chatbots, in addition to clinical therapeutic tools for use in general practice.

METHODS

The data set consists of 1870 depression-related questions posted by adolescents on a public web-based information service. Most of the posts contain descriptions of events that lead to depression. On a sample of 100 posts, we conducted a qualitative thematic analysis based on cognitive behavioral theory investigating behavioral, emotional, and symptom responses to beliefs associated with depression.

RESULTS

Results were organized into four themes. (1) Hopelessness, appearing as a set of negative beliefs about the future, possibly results from erroneous beliefs about the causal link between risk factors and the course of depression. We found beliefs about establishing a sturdy therapy alliance as a responsibility resting on the patient. (2) Therapy hesitancy seemed to be associated with negative beliefs about therapy prognosis and doubts about confidentiality. (3) Social shame appeared as a consequence of impaired daily function when the cause is not acknowledged. (4) Failing to attain social interaction appeared to be associated with a negative symptom response. In contrast, actively obtaining social support reduces symptoms and suicidal thoughts.

CONCLUSIONS

These results could be used to meet the clinical aims stated by earlier psychoeducation development, such as instilling hope through direct reattribution of beliefs about the future; challenging causal attributions, thereby lowering therapy hesitancy; reducing shame through the mechanisms of externalization by providing a tentative diagnosis despite the risk of stigmatizing; and providing initial symptom relief by giving advice on how to open up and reveal themselves to friends and family and balance the message of self-management to fit coping capabilities. An active counseling style advises the patient to approach the social environment, demonstrating an attitude toward self-action.

Coming together in a digital age: Community twitter responses in the wake of a campus shooting

Campus mass shootings have become a pressing policy and public health matter. Twitter is a platform used for processing events among interested community members. Examining the responses of invested community members to a mass shooting on a college campus provides evidence for how this type of violence affects the immediate community and the larger public. These responses may reflect either content (e.g. context-specific) or emotions (e.g. humor). Aims Using Twitter data, we analyzed the emotional responses as well as the nature of non-affective short-term reactions, in response to the April 2019 shooting at UNC Charlotte. Methods Drawn from a pool of tweets between 4/30/19-5/7/19, we analyzed 16,749 tweets using keywords related to the mass shooting (e.g. "shooting," "gun violence," "UNC Charlotte"). A coding team manually coded the tweets using content and sentiment analyses. Results Overall, 7,148 (42.67%) tweets contained negative emotions (e.g. anger, sadness, disgust, anxiety), 5,088 (30.38%) contained positive emotions (e.g. humor, hope, appreciation), 14,892 (88.91%) were communal responses to the shooting (e.g. prayers, healing, victim remembrance), 8,329 (49.73%) were action-oriented (e.g. action taken, policy advocacy), and 15,498 (92.53%) included information (e.g. death/injury, news). All tweets except positive emotions peaked one day following the incident. Conclusions Our findings point to peaks in most emotions in the 24 hours following the event, with the exception of positive emotions which peaked one day later. Social media responses to a campus shooting suggest college preparedness for immediate deployment of supportive responses in the case of campus violence is needed.

Digitale Phänotypisierung in der Psychologie – ein Quantensprung in der psychologischen Forschung?

Zusammenfassung. Digitale Phänotypisierung stellt einen neuen, leistungsstarken Ansatz zur Realisierung psychodiagnostischer Aufgaben in vielen Bereichen der Psychologie und Medizin dar. Die Grundidee besteht aus der Nutzung digitaler Spuren aus dem Alltag, um deren Vorhersagekraft für verschiedenste Anwendungsmöglichkeiten zu überprüfen und zu nutzen. Voraussetzungen für eine erfolgreiche Umsetzung sind elaborierte Smart Sensing Ansätze sowie Big Data-basierte Extraktions- (Data Mining) und Machine Learning-basierte Analyseverfahren. Erste empirische Studien verdeutlichen das hohe Potential, aber auch die forschungsmethodischen sowie ethischen und rechtlichen Herausforderungen, um über korrelative Zufallsbefunde hinaus belastbare Befunde zu gewinnen. Hierbei müssen rechtliche und ethische Richtlinien sicherstellen, dass die Erkenntnisse in einer für Einzelne und die Gesellschaft als Ganzes wünschenswerten Weise genutzt werden. Für die Psychologie als Lehr- und Forschungsdomäne bieten sich durch Digitale Phänotypisierung vielfältige Möglichkeiten, die zum einen eine gelebte Zusammenarbeit verschiedener Fachbereiche und zum anderen auch curriculare Erweiterungen erfordern. Die vorliegende narrative Übersicht bietet eine theoretische, nicht-technische Einführung in das Forschungsfeld der Digitalen Phänotypisierung, mit ersten empirischen Befunden sowie einer Diskussion der Möglichkeiten und Grenzen sowie notwendigen Handlungsfeldern.

Predicting Mental Health Status in Remote and Rural Farming Communities: Computational Analysis of Text-Based Counseling

Background

Australians living in rural and remote areas are at elevated risk of mental health problems and must overcome barriers to help seeking, such as poor access, stigma, and entrenched stoicism. e-Mental health services circumvent such barriers using technology, and text-based services are particularly well suited to clients concerned with privacy and self-presentation. They allow the client to reflect on the therapy session after it has ended as the chat log is stored on their device. The text also offers researchers an opportunity to analyze language use patterns and explore how these relate to mental health status.

Objective

In this project, we investigated whether computational linguistic techniques can be applied to text-based communications with the goal of identifying a client’s mental health status.

Methods

Client-therapist text messages were analyzed using the Linguistic Inquiry and Word Count tool. We examined whether the resulting word counts related to the participants’ presenting problems or their self-ratings of mental health at the completion of counseling.

Results

The results confirmed that word use patterns could be used to differentiate whether a client had one of the top 3 presenting problems (depression, anxiety, or stress) and, prospectively, to predict their self-rated mental health after counseling had been completed.

Conclusions

These findings suggest that language use patterns are useful for both researchers and clinicians trying to identify individuals at risk of mental health problems, with potential applications in screening and targeted intervention.

What Are the Characteristics of User Texts and Behaviors in Chinese Depression Posts?

Social media platforms provide unique insights into mental health issues, but a large number of related studies have focused on English text information. The purpose of this paper is to identify the posting content and posting behaviors of users with depression on Chinese social media. These clues may suggest signs of depression. We created two data sets consisting of 130 users with diagnosed depression and 320 other users that were randomly selected. By comparing and analyzing the two data sets, we can observe more closely how users reveal their signs of depression on Chinese social platforms. The results show that the distribution of some Chinese speech users with depression is significantly different from that of other users. Emotional sadness, fear and disgust are more common in the depression class. For personal pronouns, negative words and interrogative words, there are also great differences between the two data sets. Using topic modeling, we found that patients mainly discussed seven topics: negative emotion fluctuation, disease treatment and somatic responses, sleep disorders, sense of worthlessness, suicidal extreme behavior, seeking emotional support and interpersonal communication. The depression class post negative polarity posts much more frequently than other users. The frequency and characteristics of posts also reveal certain characteristics, such as sleep problems and reduced self-disclosure. In this study, we used Chinese microblog data to conduct a detailed analysis of the users showing depression signs, which helps to identify more patients with depression. At the same time, the study can provide a further theoretical basis for cross-cultural research of different language groups in the field of psychology.

Analysis and classification of privacy-sensitive content in social media posts

User-generated contents often contain private information, even when they are shared publicly on social media and on the web in general. Although many filtering and natural language approaches for automatically detecting obscenities or hate speech have been proposed, determining whether a shared post contains sensitive information is still an open issue. The problem has been addressed by assuming, for instance, that sensitive contents are published anonymously, on anonymous social media platforms or with more restrictive privacy settings, but these assumptions are far from being realistic, since the authors of posts often underestimate or overlook their actual exposure to privacy risks. Hence, in this paper, we address the problem of content sensitivity analysis directly, by presenting and characterizing a new annotated corpus with around ten thousand posts, each one annotated as sensitive or non-sensitive by a pool of experts. We characterize our data with respect to the closely-related problem of self-disclosure, pointing out the main differences between the two tasks. We also present the results of several deep neural network models that outperform previous naive attempts of classifying social media posts according to their sensitivity, and show that state-of-the-art approaches based on anonymity and lexical analysis do not work in realistic application scenarios.

Digital Technologies for Emotion-Regulation Assessment and Intervention: A Conceptual Review

The ability to regulate emotions in response to stress is central to healthy development. While early research in emotion regulation predominantly employed static, self-report measurement, the past decade has seen a shift in focus toward understanding the dynamic nature of regulation processes. This is reflected in recent refinements in the definition of emotion regulation, which emphasize the importance of the ability to flexibly adapt regulation efforts across contexts. The latest proliferation of digital technologies employed in mental health research offers the opportunity to capture the state- and context-sensitive nature of emotion regulation. In this conceptual review, we examine the use of digital technologies (ecological momentary assessment; wearable and smartphone technology, physical activity, acoustic data, visual data, and geo-location; smart home technology; virtual reality; social media) in the assessment of emotion regulation and describe their application to interventions. We also discuss challenges and ethical considerations, and outline areas for future research.

Identifying Insomnia From Social Media Posts: Psycholinguistic Analyses of User Tweets

Background

Many people suffer from insomnia, a sleep disorder characterized by difficulty falling and staying asleep during the night. As social media have become a ubiquitous platform to share users’ thoughts, opinions, activities, and preferences with their friends and acquaintances, the shared content across these platforms can be used to diagnose different health problems, including insomnia. Only a few recent studies have examined the prediction of insomnia from Twitter data, and we found research gaps in predicting insomnia from word usage patterns and correlations between users’ insomnia and their Big 5 personality traits as derived from social media interactions.

Objective

The purpose of this study is to build an insomnia prediction model from users’ psycholinguistic patterns, including the elements of word usage, semantics, and their Big 5 personality traits as derived from tweets.

Methods

In this paper, we exploited both psycholinguistic and personality traits derived from tweets to identify insomnia patients. First, we built psycholinguistic profiles of the users from their word choices and the semantic relationships between the words of their tweets. We then determined the relationship between a users’ personality traits and insomnia. Finally, we built a double-weighted ensemble classification model to predict insomnia from both psycholinguistic and personality traits as derived from user tweets.

Results

Our classification model showed strong prediction potential (78.8%) to predict insomnia from tweets. As insomniacs are generally ill-tempered and feel more stress and mental exhaustion, we observed significant correlations of certain word usage patterns among them. They tend to use negative words (eg, “no,” “not,” “never”). Some people frequently use swear words (eg, “damn,” “piss,” “fuck”) with strong temperament. They also use anxious (eg, “worried,” “fearful,” “nervous”) and sad (eg, “crying,” “grief,” “sad”) words in their tweets. We also found that the users with high neuroticism and conscientiousness scores for the Big 5 personality traits likely have strong correlations with insomnia. Additionally, we observed that users with high conscientiousness scores have strong correlations with insomnia patterns, while negative correlation between extraversion and insomnia was also found.

Conclusions

Our model can help predict insomnia from users’ social media interactions. Thus, incorporating our model into a software system can help family members detect insomnia problems in individuals before they become worse. The software system can also help doctors to diagnose possible insomnia in patients.

Finding Relevant Psychoeducation Content for Adolescents Experiencing Symptoms of Depression: Content Analysis of User-Generated Online Texts

Background

Symptoms of depression are frequent in youth and may develop into more severe mood disorders, suggesting interventions should take place during adolescence. However, young people tend not to share mental problems with friends, family, caregivers, or professionals. Many receive misleading information when searching the internet. Among several attempts to create mental health services for adolescents, technological information platforms based on psychoeducation show promising results. Such development rests on established theories and therapeutic models. To fulfill the therapeutic potential of psychoeducation in health technologies, we lack data-driven research on young peoples’ demand for information about depression.

Objective

Our objective is to gain knowledge about what information is relevant to adolescents with symptoms of depression. From this knowledge, we can develop a population-specific psychoeducation for use in different technology platforms.

Methods

We conducted a qualitative, constructivist-oriented content analysis of questions submitted by adolescents aged 16-20 years to an online public information service. A sample of 100 posts containing questions on depression were randomly selected from a total of 870. For analysis, we developed an a priori codebook from the main information topics of existing psychoeducational programs on youth depression. The distribution of topic prevalence in the total volume of posts containing questions on depression was calculated.

Results

With a 95% confidence level and a ±9.2% margin of error, the distribution analysis revealed the following categories to be the most prevalent among adolescents seeking advice about depression: self-management (33%, 61/180), etiology (20%, 36/180), and therapy (20%, 36/180). Self-management concerned subcategories on coping in general and how to open to friends, family, and caregivers. The therapy topic concerned therapy options, prognosis, where to seek help, and how to open up to a professional. We also found young people dichotomizing therapy and self-management as opposite entities. The etiology topic concerned stressors and risk factors. The diagnosis category was less frequently referred to (9%, 17/180).

Conclusions

Self-management, etiology, and therapy are the most prevalent categories among adolescents seeking advice about depression. Young people also dichotomize therapy and self-management as opposite entities. Future research should focus on measures to promote self-management, measures to stimulate expectations of self-efficacy, information about etiology, and information about diagnosis to improve self-monitoring skills, enhancing relapse prevention.

Utilizing Instagram Data to Identify Usage Patterns Associated With Schizophrenia Spectrum Disorders

Background and Objectives: Prior research has successfully identified linguistic and behavioral patterns associated with schizophrenia spectrum disorders (SSD) from user generated social media activity. Few studies, however, have explored the potential for image analysis to inform psychiatric care for individuals with SSD. Given the popularity of image-based platforms, such as Instagram, investigating user generated image data could further strengthen associations between social media activity and behavioral health. Methods: We collected 11,947 Instagram posts across 68 participants (mean age = 23.6; 59% male) with schizophrenia spectrum disorders (SSD; n = 34) and healthy volunteers (HV; n = 34). We extracted image features including color composition, aspect ratio, and number of faces depicted. Additionally, we considered social connections and behavioral features. We explored differences in usage patterns between SSD and HV participants. Results: Individuals with SSD posted images with lower saturation (p = 0.033) and lower colorfulness (p = 0.005) compared to HVs, as well as images showing fewer faces on average (SSD = 1.5, HV = 2.4, p < 0.001). Further, individuals with SSD demonstrated a lower ratio of followers to following compared to HV participants (p = 0.025). Conclusion: Differences in uploaded images and user activity on Instagram were identified in individuals with SSD. These differences highlight potential digital biomarkers of SSD from Instagram data.

The relationship between linguistic expression in blog content and symptoms of depression, anxiety, and suicidal thoughts: A longitudinal study

Data generated within social media platforms may present a new way to identify individuals who are experiencing mental illness. This study aimed to investigate the associations between linguistic features in individuals' blog data and their symptoms of depression, generalised anxiety, and suicidal ideation. Individuals who blogged were invited to participate in a longitudinal study in which they completed fortnightly symptom scales for depression and anxiety (PHQ-9, GAD-7) for a period of 36 weeks. Blog data published in the same period was also collected, and linguistic features were analysed using the LIWC tool. Bivariate and multivariate analyses were performed to investigate the correlations between the linguistic features and symptoms between subjects. Multivariate regression models were used to predict longitudinal changes in symptoms within subjects. A total of 153 participants consented to the study. The final sample consisted of the 38 participants who completed the required number of symptom scales and generated blog data during the study period. Between-subject analysis revealed that the linguistic features "tentativeness" and "non-fluencies" were significantly correlated with symptoms of depression and anxiety, but not suicidal thoughts. Within-subject analysis showed no robust correlations between linguistic features and changes in symptoms. The findings may provide evidence of a relationship between some linguistic features in social media data and mental health; however, the study was limited by missing data and other important considerations. The findings also suggest that linguistic features observed at the group level may not generalise to, or be useful for, detecting individual symptom change over time.

A Systematic review of the validity of screening depression through Facebook, Twitter, Instagram, and Snapchat

BACKGROUND

The aim of this study was to determine the validity of using social media for depression screening.

METHOD

Article searches on PubMed and PsycINFO from database inception to August 20, 2019 were completed with a search string and filters.

RESULTS

15 articles made the inclusion criteria. Facebook, Twitter, and Instagram profiles of depressed people were distinguishable from nondepressed people shown by social media markers. Facebook studies showed that having fewer Facebook friends and mutual friends, posting frequently, and using fewer location tags positively correlated with depressive symptoms. Also, Facebook posts with explicit expression of depressive symptoms, use of personal pronouns, and words related to pain, depressive symptoms, aggressive emotions, and rumination predicted depression. Twitter studies showed that the use of "past focus" words, negative emotions and anger words, and fewer words per Tweet positively correlated with depression. Finally, Instagram studies showed that differences in follower patterns, photo posting and editing, and linguistic features between depressed people and nondepressed people could serve as a marker.

LIMITATIONS

The primary articles analyzed had different methods, which constricts the amount of comparisons that can be made. Further, only four social media platforms were explored.

CONCLUSION

Social media markers like number and content of Facebook messages, linguistic variability in tweets and tweet word count on Twitter, and number of followers, frequency of Instagram use and the content of messages on Instagram differed between depressed people and nondepressed people. Therefore, screening social media profiles on these platforms could be a valid way to detect depression.

Investigating the Use of Electronic Well-being Diaries Completed Within a Psychoeducation Program for University Students: Longitudinal Text Analysis Study

Background

Psychoeducation has the potential to support students experiencing distress and help meet the demand for support; however, there is a need to understand how these programs are experienced. Web-based diaries are a useful activity for psychoeducation because of their therapeutic benefits, ability to capture naturalistic data relevant to well-being, and appropriateness for text analysis methods.

Objective

This study aims to examine how university students use electronic diaries within a psychoeducation program designed to enhance mental well-being.

Methods

The Science of Happiness course was administered to 154 undergraduate students in a university setting (the United Kingdom). Diaries were collected from the students for 9 weeks. Baseline well-being data were collected using the Short Warwick-Edinburgh Mental Wellbeing Scale (SWEMWBS). The percentage of negative and positive emotion words used in diaries (emotional tone) and use of words from five life domains (social, work, money, health, and leisure) were calculated using the Linguistic Inquiry and Word Count 2015 software. Random effects (generalized least squares) regression models were estimated to examine whether time, diary characteristics, demographics, and baseline well-being predict the emotional tone of diaries.

Results

A total of 149 students participated in the diary study, producing 1124 individual diary entries. Compliance with the diary task peaked in week 1 (n=1041, 92.62%) and was at its lowest in week 3 (n=807, 71.81%). Compared with week 1, diaries were significantly more positive in their emotional tone during week 5 (mean difference 23.90, 95% CI 16.89-30.90) and week 6 (mean difference 26.62, 95% CI 19.35-33.88) when students were tasked with writing about gratitude and their strengths. Across weeks, moderate and high baseline SWEMWBS scores were associated with a higher percentage of positive emotion words used in diaries (increases compared with students scoring low in SWEMWBS were 5.03, 95% CI 0.08-9.98 and 7.48, 95% CI 1.84-13.12, respectively). At week 1, the diaries of students with the highest levels of baseline well-being (82.92, 95% CI 73.08-92.76) were more emotionally positive on average than the diaries of students with the lowest levels of baseline well-being (59.38, 95% CI 51.02-67.73). Diaries largely focused on the use of social words. The emotional tone of diary entries was positively related to the use of leisure (3.56, 95% CI 2.28-4.85) and social words (0.74, 95% CI 0.21-1.27), and inversely related to the use of health words (−1.96, 95% CI −3.70 to −0.22).

Conclusions

We found evidence for short-term task-specific spikes in the emotional positivity of web-based diary entries and recommend future studies examine the possibility of long-term impacts on the writing and well-being of students. With student well-being strategies in mind, universities should value and encourage leisure and social activities.

Suitability of Text-Based Communications for the Delivery of Psychological Therapeutic Services to Rural and Remote Communities: Scoping Review

BACKGROUND

People living in rural and remote areas have poorer access to mental health services than those living in cities. They are also less likely to seek help because of self-stigma and entrenched stoic beliefs about help seeking as a sign of weakness. E-mental health services can span great distances to reach those in need and offer a degree of privacy and anonymity exceeding that of traditional face-to-face counseling and open up possibilities for identifying at-risk individuals for targeted intervention.

OBJECTIVE

This scoping review maps the research that has explored text-based e-mental health counseling services and studies that have used language use patterns to predict mental health status. In doing so, one of the aims was to determine whether text-based counseling services have the potential to circumvent the barriers faced by clients in rural and remote communities using technology and whether text-based communications, in particular, can be used to identify individuals at risk of psychological distress or self-harm.

METHODS

We conducted a comprehensive electronic literature search of PsycINFO, PubMed, ERIC, and Web of Science databases for articles published in English through November 2020.

RESULTS

Of the 9134 articles screened, 70 met the eligibility criteria and were included in the review. There is preliminary evidence to suggest that text-based, real-time communication with a qualified therapist is an effective form of e-mental health service delivery, particularly for individuals concerned with stigma and confidentiality. There is also converging evidence that text-based communications that have been analyzed using computational linguistic techniques can be used to accurately predict progress during treatment and identify individuals at risk of serious mental health conditions and suicide.

CONCLUSIONS

This review reveals a clear need for intensified research into the extent to which text-based counseling (and predictive models using modern computational linguistics tools) may help deliver mental health treatments to underserved groups such as regional communities, identify at-risk individuals for targeted intervention, and predict progress during treatment. Such approaches have implications for policy development to improve intervention accessibility in at-risk and underserved populations.

Digital health data-driven approaches to understand human behavior

Advances in digital technologies and data analytics have created unparalleled opportunities to assess and modify health behavior and thus accelerate the ability of science to understand and contribute to improved health behavior and health outcomes. Digital health data capture the richness and granularity of individuals' behavior, the confluence of factors that impact behavior in the moment, and the within-individual evolution of behavior over time. These data may contribute to discovery science by revealing digital markers of health/risk behavior as well as translational science by informing personalized and timely models of intervention delivery. And they may help inform diagnostic classification of clinically problematic behavior and the clinical trajectories of diagnosable disorders over time. This manuscript provides a review of the state of the science of digital health data-driven approaches to understanding human behavior. It reviews methods of digital health assessment and sources of digital health data. It provides a synthesis of the scientific literature evaluating how digitally derived empirical data can inform our understanding of health behavior, with a particular focus on understanding the assessment, diagnosis and clinical trajectories of psychiatric disorders. And, it concludes with a discussion of future directions and timely opportunities in this line of research and its clinical application.

Social media analytics in nutrition research: a rapid review of current usage in investigation of dietary behaviours

OBJECTIVE

Social media analytics (SMA) has a track record in business research. The utilisation in nutrition research is unknown, despite social media being populated with real-time eating behaviours. This rapid review aimed to explore the use of SMA in nutrition research with the investigation of dietary behaviours.

DESIGN

The review was conducted according to rapid review guidelines by WHO and the National Collaborating Centre for Methods and Tools. Five databases of peer-reviewed, English language studies were searched using the keywords 'social media' in combination with 'data analytics' and 'food' or 'nutrition' and screened for those with general population health using SMA on public domain, social media data between 2014 and 2020.

RESULTS

The review identified 34 studies involving SMA in the investigation of dietary behaviours. Nutrition topics included population nutrition health investigations, alcohol consumption, dieting and eating out of the home behaviours. All studies involved content analysis with evidence of surveillance and engagement. Twitter was predominant with data sets in tens of millions. SMA tools were observed in data discovery, collection and preparation, but less so in data analysis. Approximately, a third of the studies involved interdisciplinary collaborations with health representation and only two studies involved nutrition disciplines. Less than a quarter of studies obtained formal human ethics approval.

CONCLUSIONS

SMA in nutrition research with the investigation of dietary behaviours is emerging, nevertheless, if consideration is taken with technological capabilities and ethical integrity, the future shows promise at a broad population census level and as a scoping tool or complementary, triangulation instrument.

A Comparative Study of Online Depression Communities in China

Online communities have become a tool for researchers to understand and help individuals with depression. According to their operation mode in terms of management, communities can be divided into management depression communities (MDCs) and lacking-management depression communities (LDCs). This study aimed to investigate the characteristics and impact of LDCs in comparison with MDCs. All postings from the previous year were collected from the LDC and MDC. Keywords were extracted and coded to identify the themes, and a text classifier was built to identify the type of emotions and social support expressed in the postings. Community members were then clustered to explore their different participation patterns. We found that in the LDC, the expression of negative emotions was the most popular theme, there was a lack of information about the treatment of depression and a lack of social support providers, the level of engagement of providers was low, and support seekers did not receive attention. These results reveal the need for community management and can be used to develop more effective measures to support members of online depression communities.

Infodemiology and Infoveillance: Scoping Review

BACKGROUND

Web-based sources are increasingly employed in the analysis, detection, and forecasting of diseases and epidemics, and in predicting human behavior toward several health topics. This use of the internet has come to be known as infodemiology, a concept introduced by Gunther Eysenbach. Infodemiology and infoveillance studies use web-based data and have become an integral part of health informatics research over the past decade.

OBJECTIVE

The aim of this paper is to provide a scoping review of the state-of-the-art in infodemiology along with the background and history of the concept, to identify sources and health categories and topics, to elaborate on the validity of the employed methods, and to discuss the gaps identified in current research.

METHODS

The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines were followed to extract the publications that fall under the umbrella of infodemiology and infoveillance from the JMIR, PubMed, and Scopus databases. A total of 338 documents were extracted for assessment.

RESULTS

Of the 338 studies, the vast majority (n=282, 83.4%) were published with JMIR Publications. The Journal of Medical Internet Research features almost half of the publications (n=168, 49.7%), and JMIR Public Health and Surveillance has more than one-fifth of the examined studies (n=74, 21.9%). The interest in the subject has been increasing every year, with 2018 featuring more than one-fourth of the total publications (n=89, 26.3%), and the publications in 2017 and 2018 combined accounted for more than half (n=171, 50.6%) of the total number of publications in the last decade. The most popular source was Twitter with 45.0% (n=152), followed by Google with 24.6% (n=83), websites and platforms with 13.9% (n=47), blogs and forums with 10.1% (n=34), Facebook with 8.9% (n=30), and other search engines with 5.6% (n=19). As for the subjects examined, conditions and diseases with 17.2% (n=58) and epidemics and outbreaks with 15.7% (n=53) were the most popular categories identified in this review, followed by health care (n=39, 11.5%), drugs (n=40, 10.4%), and smoking and alcohol (n=29, 8.6%).

CONCLUSIONS

The field of infodemiology is becoming increasingly popular, employing innovative methods and approaches for health assessment. The use of web-based sources, which provide us with information that would not be accessible otherwise and tackles the issues arising from the time-consuming traditional methods, shows that infodemiology plays an important role in health informatics research.

Methods in predictive techniques for mental health status on social media: a critical review

Social media is now being used to model mental well-being, and for understanding health outcomes. Computer scientists are now using quantitative techniques to predict the presence of specific mental disorders and symptomatology, such as depression, suicidality, and anxiety. This research promises great benefits to monitoring efforts, diagnostics, and intervention design for these mental health statuses. Yet, there is no standardized process for evaluating the validity of this research and the methods adopted in the design of these studies. We conduct a systematic literature review of the state-of-the-art in predicting mental health status using social media data, focusing on characteristics of the study design, methods, and research design. We find 75 studies in this area published between 2013 and 2018. Our results outline the methods of data annotation for mental health status, data collection and quality management, pre-processing and feature selection, and model selection and verification. Despite growing interest in this field, we identify concerning trends around construct validity, and a lack of reflection in the methods used to operationalize and identify mental health status. We provide some recommendations to address these challenges, including a list of proposed reporting standards for publications and collaboration opportunities in this interdisciplinary space.

Are Aspects of Twitter Use Associated with Reduced Depressive Symptoms? The Moderating Role of In-Person Social Support

In a two-wave, 4-month longitudinal study of 308 adults, two hypotheses were tested regarding the relation of Twitter-based measures of online social media use and in-person social support with depressive thoughts and symptoms. For four of five measures, Twitter use by in-person social support interactions predicted residualized change in depression-related outcomes over time; these results supported a corollary of the social compensation hypothesis that social media use is associated with greater benefits for people with lower in-person social support. In particular, having a larger Twitter social network (i.e., following and being followed by more people) and being more active in that network (i.e., sending and receiving more tweets) are especially helpful to people who have lower levels of in-person social support. For the fifth measure (the sentiment of Tweets), no interaction emerged; however, a beneficial main effect offset the adverse main effect of low in-person social support.

Patterns and Longitudinal Changes in Negative Emotions of People with Depression on Sina Weibo

Background: This study investigates negative emotional patterns of people with depression on Sina Weibo™ and gives an in-depth analysis of how their negative emotions change over time. Materials and Methods: A text classifier using deep learning methods was built to identify people on Sina Weibo with depression and associated negative emotions. The longitudinal changes in negative emotions were assessed using time series and cluster analysis. Results: Results indicate that people with depression (n = 616) were more active and expressed more negative emotions on social media compared with control users (n = 3,176). Furthermore, negative emotions of people with depression were mostly about their depression issues, such as treatment, hopelessness, suicidal ideation, or self-injury. Both groups of users usually expressed negative emotions in the late evening and early morning hours. Finally, longitudinal changes in negative emotions illustrate that users with depression tended to have relatively high negative emotions in the month when they started using social media to reveal their depression issues and that they exhibited three main evolutionary patterns of negative emotions on social media. Conclusions: Findings from the study could be used to track and monitor negative emotional states of people with depression on social media, in addition to providing an in-depth understanding of how negative emotions change over time, so that better intervention strategies can be adopted to assist them.

Twitter-based measures of neighborhood sentiment as predictors of residential population health

Several studies have recently applied sentiment-based lexicons to Twitter to gauge local sentiment to understand health behaviors and outcomes for local areas. While this research has demonstrated the vast potential of this approach, lingering questions remain regarding the validity of Twitter mining and surveillance in local health research. First, how well does this approach predict health outcomes at very local scales, such as neighborhoods? Second, how robust are the findings garnered from sentiment signals when accounting for spatial effects? To evaluate these questions, we link 2,076,025 tweets from 66,219 distinct users in the city of San Diego over the period of 2014-12-06 to 2017-05-24 to the 500 Cities Project data and 2010-2014 American Community Survey data. We determine how well sentiment predicts self-rated mental health, sleep quality, and heart disease at a census tract level, controlling for neighborhood characteristics and spatial autocorrelation. We find that sentiment is related to some outcomes on its own, but these relationships are not present when controlling for other neighborhood factors. Evaluating our encoding strategy more closely, we discuss the limitations of existing measures of neighborhood sentiment, calling for more attention to how race/ethnicity and socio-economic status play into inferences drawn from such measures.

Detecting Signs of Depression in Tweets in Spanish: Behavioral and Linguistic Analysis

BACKGROUND

Mental disorders have become a major concern in public health, and they are one of the main causes of the overall disease burden worldwide. Social media platforms allow us to observe the activities, thoughts, and feelings of people's daily lives, including those of patients suffering from mental disorders. There are studies that have analyzed the influence of mental disorders, including depression, in the behavior of social media users, but they have been usually focused on messages written in English.

OBJECTIVE

The study aimed to identify the linguistic features of tweets in Spanish and the behavioral patterns of Twitter users who generate them, which could suggest signs of depression.

METHODS

This study was developed in 2 steps. In the first step, the selection of users and the compilation of tweets were performed. A total of 3 datasets of tweets were created, a depressive users dataset (made up of the timeline of 90 users who explicitly mentioned that they suffer from depression), a depressive tweets dataset (a manual selection of tweets from the previous users, which included expressions indicative of depression), and a control dataset (made up of the timeline of 450 randomly selected users). In the second step, the comparison and analysis of the 3 datasets of tweets were carried out.

RESULTS

In comparison with the control dataset, the depressive users are less active in posting tweets, doing it more frequently between 23:00 and 6:00 (P<.001). The percentage of nouns used by the control dataset almost doubles that of the depressive users (P<.001). By contrast, the use of verbs is more common in the depressive users dataset (P<.001). The first-person singular pronoun was by far the most used in the depressive users dataset (80%), and the first- and the second-person plural pronouns were the least frequent (0.4% in both cases), this distribution being different from that of the control dataset (P<.001). Emotions related to sadness, anger, and disgust were more common in the depressive users and depressive tweets datasets, with significant differences when comparing these datasets with the control dataset (P<.001). As for negation words, they were detected in 34% and 46% of tweets in among depressive users and in depressive tweets, respectively, which are significantly different from the control dataset (P<.001). Negative polarity was more frequent in the depressive users (54%) and depressive tweets (65%) datasets than in the control dataset (43.5%; P<.001).

CONCLUSIONS

Twitter users who are potentially suffering from depression modify the general characteristics of their language and the way they interact on social media. On the basis of these changes, these users can be monitored and supported, thus introducing new opportunities for studying depression and providing additional health care services to people with this disorder.

Analyzing Twitter as a Platform for Alzheimer-Related Dementia Awareness: Thematic Analyses of Tweets

Background

Dementia is a prevalent disorder among adults and often subjects an individual and his or her family. Social media websites may serve as a platform to raise awareness for dementia and allow researchers to explore health-related data.

Objective

The objective of this study was to utilize Twitter, a social media website, to examine the content and location of tweets containing the keyword “dementia” to better understand the reasons why individuals discuss dementia. We adopted an approach that analyzed user location, user category, and tweet content subcategories to classify large publicly available datasets.

Methods

A total of 398 tweets were collected using the Twitter search application programming interface with the keyword “dementia,” circulated between January and February 2018. Twitter users were categorized into 4 categories: general public, health care field, advocacy organization, and public broadcasting. Tweets posted by “general public” users were further subcategorized into 5 categories: mental health advocate, affected persons, stigmatization, marketing, and other. Placement into the categories was done through thematic analysis.

Results

A total of 398 tweets were written by 359 different screen names from 28 different countries. The largest number of Twitter users were from the United States and the United Kingdom. Within the United States, the largest number of users were from California and Texas. The majority (281/398, 70.6%) of Twitter users were categorized into the “general public” category. Content analysis of tweets from the “general public” category revealed stigmatization (113/281, 40.2%) and mental health advocacy (102/281, 36.3%) as the most common themes. Among tweets from California and Texas, California had more stigmatization tweets, while Texas had more mental health advocacy tweets.

Conclusions

Themes from the content of tweets highlight the mixture of the political climate and the supportive network present on Twitter. The ability to use Twitter to combat stigma and raise awareness of mental health indicates the benefits that can potentially be facilitated via the platform, but negative stigmatizing tweets may interfere with the effectiveness of this social support.

Exploring the Utility of Community-Generated Social Media Content for Detecting Depression: An Analytical Study on Instagram

Background

The content produced by individuals on various social media platforms has been successfully used to identify mental illness, including depression. However, most of the previous work in this area has focused on user-generated content, that is, content created by the individual, such as an individual’s posts and pictures. In this study, we explored the predictive capability of community-generated content, that is, the data generated by a community of friends or followers, rather than by a sole individual, to identify depression among social media users.

Objective

The objective of this research was to evaluate the utility of community-generated content on social media, such as comments on an individual’s posts, to predict depression as defined by the clinically validated Patient Health Questionnaire-8 (PHQ-8) assessment questionnaire. We hypothesized that the results of this research may provide new insights into next generation of population-level mental illness risk assessment and intervention delivery.

Methods

We created a Web-based survey on a crowdsourcing platform through which participants granted access to their Instagram profiles as well as provided their responses to PHQ-8 as a reference standard for depression status. After data quality assurance and postprocessing, the study analyzed the data of 749 participants. To build our predictive model, linguistic features were extracted from Instagram post captions and comments, including multiple sentiment scores, emoji sentiment analysis results, and meta-variables such as the number of likes and average comment length. In this study, 10.4% (78/749) of the data were held out as a test set. The remaining 89.6% (671/749) of the data were used to train an elastic-net regularized linear regression model to predict PHQ-8 scores. We compared different versions of this model (ie, a model trained on only user-generated data, a model trained on only community-generated data, and a model trained on the combination of both types of data) on a test set to explore the utility of community-generated data in our predictive analysis.

Results

The 2 models, the first trained on only community-generated data (area under curve [AUC]=0.71) and the second trained on a combination of user-generated and community-generated data (AUC=0.72), had statistically significant performances for predicting depression based on the Mann-Whitney U test (P=.03 and P=.02, respectively). The model trained on only user-generated data (AUC=0.63; P=.11) did not achieve statistically significant results. The coefficients of the models revealed that our combined data classifier effectively amalgamated both user-generated and community-generated data and that the 2 feature sets were complementary and contained nonoverlapping information in our predictive analysis.

Conclusions

The results presented in this study indicate that leveraging community-generated data from social media, in addition to user-generated data, can be informative for predicting depression among social media users.