Understanding Depressive Symptoms and Psychosocial Stressors on Twitter: A Corpus-Based Study

Abstract

BackgroundWith a lifetime prevalence of 16.2%, major depressive disorder is the fifth biggest contributor to the disease burden in the United States.ObjectiveThe aim of this study, building on previous work qualitatively analyzing depression-related Twitter data, was to describe the development of a comprehensive annotation scheme (ie, coding scheme) for manually annotating Twitter data with Diagnostic and Statistical Manual of Mental Disorders, Edition 5 (DSM 5) major depressive symptoms (eg, depressed mood, weight change, psychomotor agitation, or retardation) and Diagnostic and Statistical Manual of Mental Disorders, Edition IV (DSM-IV) psychosocial stressors (eg, educational problems, problems with primary support group, housing problems).MethodsUsing this annotation scheme, we developed an annotated corpus, Depressive Symptom and Psychosocial Stressors Acquired Depression, the SAD corpus, consisting of 9300 tweets randomly sampled from the Twitter application programming interface (API) using depression-related keywords (eg, depressed, gloomy, grief). An analysis of our annotated corpus yielded several key results.ResultsFirst, 72.09% (6829/9473) of tweets containing relevant keywords were nonindicative of depressive symptoms (eg, “we’re in for a new economic depression”). Second, the most prevalent symptoms in our dataset were depressed mood and fatigue or loss of energy. Third, less than 2% of tweets contained more than one depression related category (eg, diminished ability to think or concentrate, depressed mood). Finally, we found very high positive correlations between some depression-related symptoms in our annotated dataset (eg, fatigue or loss of energy and educational problems; educational problems and diminished ability to think).ConclusionsWe successfully developed an annotation scheme and an annotated corpus, the SAD corpus, consisting of 9300 tweets randomly-selected from the Twitter application programming interface using depression-related keywords. Our analyses suggest that keyword queries alone might not be suitable for public health monitoring because context can change the meaning of keyword in a statement. However, postprocessing approaches could be useful for reducing the noise and improving the signal needed to detect depression symptoms using social media.

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

“…This suggests that the time-structured patterns of emotion expressed on Facebook may provide better differentiation between individuals with and without depression where they express similar levels of negative emotion words. The current study suggests that the poor hit rate in some keyword approaches to classifying depression in status updates, as described by Mowery et al [15], may be enhanced by including measures of moment-to-moment variability in emotion word use.…”

“…For instance, Moreno et al [3] demonstrated that status updates on Facebook displaying references to depression symptoms, such as hopelessness, positively correlated with self-reported depression symptoms. Others have extended this by describing the linguistic characteristics of depression in posts and developing coding-schemes to identify depression-indicative Tweets or status updates [2,4,7,15]. While specific topics, keywords, and linguistic features (notably negative emotions) are able to identify depression-indicative posts with high sensitivity, many of these features may also be present in posts that are nonindicative of depression (low specificity).…”

“…While specific topics, keywords, and linguistic features (notably negative emotions) are able to identify depression-indicative posts with high sensitivity, many of these features may also be present in posts that are nonindicative of depression (low specificity). For example, Mowery et al [15] found that over 70% of tweets identified in their sample containing words related to depression were not actually indicative of depression, that is, words included in the depression annotation scheme were being used in tweets unrelated to a mental health disclosure. Thus, although negative emotion words correlate with the presence of depressive symptoms, it is a noisy and imprecise metric.…”

“…As shown by Mowery et al [15], using a keyword approach to identifying depression in social media posts results in a large proportion of false-positives, reducing the specificity with which depression can be identified through the average emotion expressed over time. Context is one reason this may occur, where the use of a word may not directly link to an experienced emotion (e.g., "I went to visit Happy Valley").…”

“…Future research should seek to apply more sophisticated open vocabulary approaches or postprocessing of status updates [15,41] to provide greater detail and accuracy of the language use context.…”