Elamur Research⌕
☰
home / papers / 2014 / Quantifying Mental Health Signals in Twitter

Quantifying Mental Health Signals in Twitter

Summary

Mental health research has far less quantifiable data than most physical health fields. Survey-based social media studies are also small and limited to conditions with an online-administrable questionnaire. The authors instead build labeled Twitter datasets automatically. They search a large multi-year health tweet collection for self-reported diagnoses (patterns like "I was diagnosed with X"), manually keep only genuine statements, and download those users' recent timelines. A random sample of Twitter users serves as an approximate control group. They cover four conditions: depression, bipolar disorder, PTSD and seasonal affective disorder (SAD).

On these data they run three analyses:

  • LIWC comparisons against controls,
  • classifiers that separate diagnosed from control users,
  • open-vocabulary language models.

The LIWC results reproduce known depression findings and extend them to the other three disorders. The classifiers distinguish diagnosed users for every condition. The language models beat LIWC and pattern-of-life features, which suggests Twitter holds mental-health signal not yet described in the clinical literature. The approach is cheap and fast, and it complements surveys rather than replacing them.

Key points

  • Filtering: each candidate user's newest tweets (up to 3200) were fetched. Users were kept only if they had at least 25 tweets and at least 75% of their tweets were English, as judged by the Compact Language Detector.
  • Dataset size:
GroupUsersTweets
Bipolar394992k
Depression4411.0m
PTSD244573k
SAD159421k
Control5728 (from 10k sampled)13.7m
  • Depression yield: only half of the depression diagnosis matches were annotated. The authors estimate roughly 800–900 depression users are recoverable, versus 117 in the survey-based study of De Choudhury et al. (2013d).
  • LIWC features: Swear, Anger, PosEmo, NegEmo, Anx, plus pronoun groups merged by grammatical person: Pro1 (I+We), Pro2 (You), Pro3 (SheHe+They).
  • Cross-disorder correlations: scores for depression, PTSD and bipolar correlate with each other, while SAD stands apart. The overlap of users shared between groups could explain only correlations up to about r = 0.05.
  • Coverage limits: Alzheimer's (its patients rarely use Twitter) and eating disorders produced few or no self-stated diagnoses.

Interesting ideas

  • I1. Self-disclosure as a labeling source: regex matching on first-person statements, followed by a quick human check for jokes and quotes, gives large user-level labels cheaply. This could bootstrap datasets for any rare condition or self-identified attribute where surveys are slow or expensive.
  • I2. Replicate before claiming new findings: the authors first showed that the noisy, automatically built data reproduces established LIWC findings for depression, then reported new results. This is a cheap sanity check for any weakly supervised corpus.
  • I3. Literature features vs. open vocabulary: comparing theory-driven features (LIWC) with data-driven language models shows whether the text holds signal beyond existing theory. A gap in favour of the open models points to undiscovered phenomena worth studying.
  • I4. Correlating analytics across tasks: checking how scores from different models and disorder groups correlate revealed similar language among the often-comorbid disorders. The authors also checked that shared users could not explain the effect. Such correlation checks help interpret what black-box classifiers actually capture.
  • I5. Keep a large random control sample: a big random user sample serves as a reusable negative class across several target groups. The authors note it is likely contaminated, but it remains a practical baseline when true controls cannot be verified.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

Cited by

117 papers of our library cite it (of the 140884 whose bibliographies we read); 7 of them have a page here, 110 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/w14-3207

@inproceedings{coppersmith2014quantifying,
  title = {{Quantifying Mental Health Signals in Twitter}},
  author = {Coppersmith, Glen and Dredze, Mark and Harman, Craig},
  booktitle = {Proceedings of the Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality},
  year = {2014},
  doi = {10.3115/v1/w14-3207},
  url = {https://doi.org/10.3115/v1/w14-3207}
}

References

43 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Faiyaz Al Zamal, Wendy Liu, and Derek Ruths. 2012. Homophily and latent attribute inference: Inferring latent attributes of Twitter users from neighbors. In Proceedings of the International AAAI Conference on Weblogs and Social Media (ICWSM). find
  2. Jennifer Alvarez-Conrad, Lori A. Zoellner, and Edna B. Foa. 2001. Linguistic predictors of trauma pathology and physical health. Applied Cognitive Psychology, 15(7):S159–S170. find
  3. American Psychiatric Association. 2013. Diagnostic Statistical Manual 5. American Psychiatric Association. find
  4. Eiji Aramaki, Sachiko Maskawa, and Mizuki Morita. 2011. Twitter catches the flu: Detecting influenza epidemics using twitter. In Empirical Natural Language Processing Conference (EMNLP). find
  5. John W. Ayers, Benjamin M. Althouse, Jon-Patrick Allem, J. Niels Rosenquist, and Daniel E. Ford. 2013. Seasonality in seeking mental health information on google. American journal of preventive medicine, 44(5):520–525. find
  6. John W. Ayers, Benjamin M. Althouse, and Mark Dredze. 2014. Could behavioral medicine lead the web data revolution? Journal of the American Medical Association (JAMA), February 27. find
  7. Lisa F. Berkman, Thomas Glass, Ian Brissette, and Teresa E. Seeman. 2000. From social integration to health: Durkheim in the new millennium? Social Science & Medicine, 51(6):843–857, September. find
  8. Amber Boydstun, Rebecca Glazier, Timothy Jurka, and Matthew Pietryka. 2013. Examining debate effects in real time: A report of the 2012 React Labs: Educate study. The Political Communication Report, 23(1), February. [Online; accessed 25-February-2014]. find
  9. Kathleen T. Brady, Therese K. Killeen, Tim Brewerton, and Sylvia Lucerini. 2000. Comorbidity of psychiatric disorders and posttraumatic stress disorder. Journal of Clinical Psychiatry. find
  10. Patrick Callaghan. 2004. Exercise: a neglected intervention in mental health care? Journal of Psychiatric and Mental Health Nursing, 11:476–483. find
  11. Duncan G. Campbell, Bradford L. Felker, Chuan-Fen Liu, Elizabeth M. Yano, JoAnn E. Kirchner, Domin Chan, Lisa V. Rubenstein, and Edmund F. Chaney. 2007. Prevalence of depression-PTSD comorbidity: Implications for clinical practice guidelines and primary care-based interventions. Journal of General I… find
  12. Centers for Disease Control and Prevention (CDC). 2010. Behavioral risk factor surveillance system survey data. find
  13. Cindy Chung and James Pennebaker. 2007. The psychological functions of function words. Social communication, pages 343–359. find
  14. Glen A. Coppersmith, Craig T. Harman, and Mark Dredze. 2014. Measuring post traumatic stress disorder in Twitter. In Proceedings of the International AAAI Conference on Weblogs and Social Media (ICWSM). find
  15. Wendy D’Andrea, Pearl H. Chiu, Brooks R. Casas, and Patricia Deldin. 2011. Linguistic predictors of post-traumatic stress disorder symptoms following 11 September 2001. Applied Cognitive Psychology, 26(2):316–323, October. find
  16. Munmun De Choudhury, Scott Counts, and Eric Horvitz. 2013a. Major life changes and behavioral markers in social media: Case of childbirth. In Proceedings of the ACM Conference on Computer Supported Cooperative Work and Social Computing (CSCW). find
  17. Munmun De Choudhury, Scott Counts, and Eric Horvitz. 2013b. Predicting postpartum changes in emotion and behavior via social media. In Proceedings of the ACM Annual Conference on Human Factors in Computing Systems (CHI), pages 3267–3276. ACM. find
  18. Munmun De Choudhury, Scott Counts, and Eric Horvitz. 2013c. Social media as a measurement tool of depression in populations. In Proceedings of the Annual ACM Web Science Conference. find
  19. Munmun De Choudhury, Michael Gamon, Scott Counts, and Eric Horvitz. 2013d. Predicting depression via social media. In Proceedings of the International AAAI Conference on Weblogs and Social Media (ICWSM). find
  20. Munmun De Choudhury, Andres Monroy-Hernandez, and Gloria Mark. 2014. ” narco” emotions: Affect and desensitization in social media during the mexican drug war. find
  21. Munmun De Choudhury. 2013. Role of social media in tackling challenges in mental health. In Proceedings of the 2nd International Workshop on Socially-Aware Multimedia, pages 49–52. find
  22. Mark Dredze. 2012. How social media will change public health. IEEE Intelligent Systems, 27(4):81– 84. find
  23. Danica Vukadinovic Greetham, Robert Hurling, Gabrielle Osborne, and Alex Linley. 2011. Social networks and positive and negative affect. Procedia - Social and Behavioral Sciences, 22:4–13, January. find
  24. Carleen Hawn. 2009. Take Two Aspirin And Tweet Me In The Morning: How Twitter, Facebook, And Other Social Media Are Reshaping Health Care. Health Affairs, 28(2):361–368. find
  25. Qiwei He, Bernard P. Veldkamp, and Theo de Vries. 2012. Screening for posttraumatic stress disorder using verbal features in self narratives: A text mining approach. Psychiatry Research. find
  26. Adam D. I. Kramer, Susan R. Fussell, and Leslie D. Setlock. 2004. Text analysis as a tool for analyzing conversation in online support groups. In Proceedings of the ACM Annual Conference on Human Factors in Computing Systems (CHI). find
  27. Stephen J. Lurie, Barbara Gawinski, Deborah Pierce, and Sally J. Rousseau. 2006. Seasonal affective disorder. American family physician, 74(9). find
  28. Susan L. McElroy, Lori L. Altshuler, Trisha Suppes, Paul E. Keck, Mark A. Frye, Kirk D. Denicoff, Willem A. Nolen, Ralph W. Kupka, Gabriele S. Lev-erich, Jennifer R. Rochussen, A. John Rush Rush, and Robert M. Post Post. 2001. Axis I psychiatric comorbidity and its relationship to historical illness… find
  29. Margaret Mitchell, Jacqueline Aguilar, Theresa Wilson, and Benjamin Van Durme. 2013. Open domain targeted sentiment. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1643–1654. find our card
  30. World Health Organization. 2001. The world health report 2001 - Mental health: New understanding, new hope. Technical report, Genf, Schweiz. find
  31. Minsu Park, Chiyoung Cha, and Meeyoung Cha. 2012. Depressive moods of users portrayed in Twitter. In Proceedings of the ACM SIGKDD Workshop on Healthcare Informatics (HI-KDD). find
  32. Michael J. Paul and Mark Dredze. 2011. You are what you tweet: Analyzing Twitter for public health. In Proceedings of the International AAAI Conference on Weblogs and Social Media (ICWSM). find
  33. Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, and Matthieu Perrot Édouard Duchesnay. 2011. scikit-learn:… find
  34. Frank J. Penedo and Jason R. Dahn. 2005. Exercise and well-being: a review of mental and physical health benefits associated with physical activity. Current Opinion in Psychiatry, 18(2):189–193, March. find
  35. James W. Pennebaker, Cindy K. Chung, Molly Ireland, Amy Gonzales, and Roger J. Booth. 2007. The development and psychometric properties of LIWC2007. find
  36. Nairan Ramirez-Esparza, Cindy K. Chung, Ewa Kacewicz, and James W. Pennebaker. 2008. The psychology of word use in depression forums in English and in Spanish: Testing two text analytic approaches. In Proceedings of the International AAAI Conference on Weblogs and Social Media (ICWSM). find
  37. Philip Resnik, Anderson Garron, and Rebecca Resnik. 2013. Using topic modeling to improve prediction of neuroticism and depression. In Proceedings of the 2013 Conference on Empirical Methods in Natural, pages 1348–1353. find
  38. Stephanie S. Rude, Eva-Maria Gortner, and James W. Pennebaker. 2004. Language use of depressed and depression-vulnerable college students. Cognition & Emotion, 18(8):1121–1133, December. find
  39. H. Andrew Schwartz, Johannes C. Eichstaedt, Margaret L. Kern, Lukasz Dziurzynski, Stephanie M. Ramones, Megha Agrawal, Achal Shah, Michal Kosinski, David Stillwell, Martin E. P. Seligman, and Lyle H. Ungar. 2013. Personality, gender, and age in the language of social media: The open-vocabulary appro… find
  40. Yla R. Tausczik and James W. Pennebaker. 2010. The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29(1):24–54. find
  41. John W. Tukey. 1977. Box-and-whisker plots. Exploratory Data Analysis, pages 39–43. find
  42. Myrna M. Weissman, Roger C. Bland, Glorisa J. Canino, Carlo Faravelli, Steven Greenwald, Hai-Gwo Hwu, Peter R. Joyce, Eile G. Karam, Chung-Kyoon Lee, Joseph Lellouch, Jean-Pierre Lépine, Stephen C. Newman, Maritza Rubio-Stipec, J. Elisabeth Wells, Priya J. Wickramaratne, Hans-Ulrich Wittchen, and E… find
  43. Hadley Wickham. 2009. ggplot2: elegant graphics for data analysis. Springer. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/coppersmith-2014-quantifying-mental-health-signals-twitter/