Multitask Learning for Mental Health Conditions with Limited Social Media Data

Multi-Task Learning for Mental Health using Social Media Text Adrian Benton Johns Hopkins University adrian@cs.jhu.edu Margaret Mitchell Microsoft Research∗ mmitchellai@google.com Dirk Hovy University of Copenhagen mail@dirkhovy.com Abstract We introduce initial groundwork for esti- mating suicide risk and mental health in a deep learning framework. By model- ing multiple conditions, the system learns to make predictions about suicide risk and mental health at a low false positive rate. Conditions are modeled as tasks in a multi- task learning (MTL) framework, with gen- der prediction as an additional auxiliary task. We demonstrate the effectiveness of multi-task learning by comparison to a well-tuned single-task baseline with the same number of parameters. Our best MTL model predicts potential suicide at- tempt, as well as the presence of atypical mental health, with AUC > 0.8. We also find additional large improvements using multi-task learning on mental health tasks with limited training data. 1 Introduction Suicide is one of the leading causes of death worldwide, and over 90% of individuals who die by suicide experience mental health conditions.1 However, detecting the risk of suicide, as well as monitoring the effects of related mental health conditions, is challenging. Traditional methods rely on both self-reports and impressions formed during short sessions with a clinical expert, but it is often unclear when suicide is a risk in particular.2 Consequently, conditions leading to preventable suicides are often not adequately addressed. ∗Now at Google Research. 1https://www.nami.org/Learn- More/Mental-Health-Conditions/Related- Conditions/Suicide#sthash.dMAhrKTU.dpuf 2Communication with clinicians at the 2016 JSALT work- shop (Hollingshead, 2016). Automated monitoring and risk assessment of patients’ language has the potential to complement traditional assessment methods, providing objec- tive measurements to motivate further care and ad- ditional support for people with difficulties related to mental health. This paves the way towards ver- ifying the need for additional care with insurance coverage, for example, as well as offering direct benefits to clinicians and patients. We explore some of the possibilities in the deep learning and mental health space using written so- cial media text that people with different mental health conditions are already producing. Uncov- ering methods that work with such text provides the opportunity to help people with different men- tal health conditions by leveraging a task they are already participating in. Social media text carries implicit information about the author, which has been modeled in nat- ural language processing (NLP) to predict au- thor characteristics such as age (Goswami et al., 2009; Rosenthal and McKeown, 2011; Nguyen et al., 2014), gender (Sarawgi et al., 2011; Ciot et al., 2013; Liu and Ruths, 2013; Volkova et al., 2015; Hovy, 2015), personality (Schwartz et al., 2013; Volkova et al., 2014; Plank and Hovy, 2015; Park et al., 2015; Preot¸iuc-Pietro et al., 2015), and occupation (Preotiuc-Pietro et al., 2015). Similar text signals have been effectively used to predict mental health conditions such as depression (De Choudhury et al., 2013; Copper- smith et al., 2015b; Schwartz et al., 2014), suici- dal ideation (Coppersmith et al., 2016; Huang et al., 2015), schizophrenia (Mitchell et al., 2015) …

Content truncated to 500 words