Social Media Analytics for Mental Health Research

In this body of work, we ask how how social media data can (1) supplement and support research in mental health and (2) facilitate the development of new mental health services.  We studied whether social media data (Twitter) can be used to collect a real-time metric of emotion data and that this correlates with existing mental health metrics.   We also examined whether we could detect suicidal ideation on Twitter to support research in mental health (for example, to study stigma around suicidal ideation), and to investigate socio-technical issues in health services around suicidal ideation detection.

Quick Links:

Our contributions on the topic of social media supplementing existing research data for mental health:

  • In collaboration with the Black Dog Institute, we examined the relationship of social media data to mental health metrics around reactions to the suicide of Robin Williams and the announcement of the 2014 Federal Budget.  The analysis shows correlations but also highlights how the data is different, posing further research questions about methodology [1].
  • We show that research insights for mental health are possible in follow up work where we investigated the Sydney siege incident [2] showing that differences in emotional responses on the two days of the event could be explained by the social discourse.
  • We explore further methodological questions for a real-time monitoring signal with research about the 2017 Manchester terrorist attacks (in preparation, but sample graph presented here)
  • We show demonstrate the feasibility of using text classification methods to detect suicidal ideation, which can be used to collect data for downstream mental health research [3, 4], particularly for collecting data that also includes non-concerning ideation.

Our contributions on the topic of the socio-technical challenges for new mental health services (around suicidal ideation) include:

  • In [3] we argue that the annotation of social media data is not straightforward, even with domain experts.  There is a mutual gap in understanding between mental health and social media researchers that can make annotation instructions difficult.  Convergence on a suitable set of instructions for suicidal ideation is thus hard and a topic on ongoing research.
  • In [3], we argue that there are still many challenges for the deployment of new services for suicidal ideation, as highlighted by what classification results might be possible and the social sensitivities of the topic.
  • We demonstrate that dialogue context can be an informative feature for the detection of suicidal ideation [5].


  • This body of work was funded by the Black Dog Institute, Amazon, GNIP.  A number of our publications were co-authored with BDI.
  • Our emotion analysis module is being commercialised through Signal, and as a software library, is in use in other CSIRO Data61 projects outside of the LASC team (for example, the PlaceSME project).


  • Our work on detecting suicidal ideation was the genesis of a system we submitted to the shared task for the 2016 Computational Linguistics and Psychology workshop, in which we had the best system [4]


  1. Larsen, Mark; Boonstra, Tjeerd; Batterham, Philip; O’Dea, Bridianne; Paris, Cecile; Christensen, Helen. We Feel: Mapping emotion on Twitter. IEEE Journal of Biomedical and Health Informatics (J-BHI).. 2014; 19(4):1246-52.
  2. Wan, Stephen; Paris, Cecile. Understanding Public Emotional Reactions on Twitter. In: 2015 AAAI International Conference on Weblogs and Social Media (ICWSM); May 26th-29th, 2015; Oxford, UK. Palo Alto, California: AAAI; 2015. 715-716.
  3. O’Dea, Bridianne; Wan, Stephen; Batterham, Philip; Calear, Alison; Paris, Cecile; Christensen, Helen. Detecting Suicidality on Twitter. Internet Interventions. 2015; 2(2):183–188.
  4. Kim, Mac; Wang, Yufei; Wan, Stephen; Paris, Cecile. Data61-CSIRO systems at the CLPsych 2016 Shared Task. In: Proceedings of the Third Workshop on Computational Linguistics and Clinical Psychology; June 12-17, 2016; San Diego, CA, USA. Association for Computational Linguistics; 2016. 128-132.
  5. Wang, Y.; Wan, S. & Paris, C. (2016) The Role of Features and Context on Suicide Ideation Detection.  In the Proceedings of the Australasian Language Technology Association Workshop 2016, 2016, 94-102. (Award: ALTA 2016 Best Student Paper)