Blog Data “The Ground Truth”
Project with Pattern Recognition and Data Analysis, Deakin University
In this study, we will profile users’ mental health by harnessing novel machine learning based analyses of social media conversations.
Features that capture depression levels will be determined by correlating ‘ground truth’ depression measures with data-up indicators derived from machine learning using members of depression online communities.
Using the features identified, machine-learning models will then be applied to each of 1,000 individual trajectories over a six-month period to determine individualised predictive features. Concurrently, machine-learning models will be applied to 1,000 bloggers in anxiety and suicide prevention online communities.
In the final phase, algorithms will be applied to 5,000 Facebook users to determine generalisability. In particular, the performance of the models will be tested in samples that include symptomatic individuals without diagnosis who have very little contact with mental health services.
Outcomes of the study will include greater understanding of actual, real-time ‘descent’ into illness state – something unachievable using conventional survey methods and the development predictive systems capable of driving decisions concerning the provision of support for mental illness through social media, and the provision of alerts to individuals, carers and medical practitioners. The project is fundamentally transformative, in that it will demonstrate, for the first time, the ability of social media to generate personalised risk data, and directly lead to personalised apps and sophisticated decision tools for clinician use.
The research will be conducted by two outstanding internationally teams – one in machine learning and the other in Internet psychiatric research, areas where Australia already leads the world.