A Taxonomy of Ethical Tensions in Inferring Mental Health States from Social Media

Social Media and Mental Health: Benefits, Risks, and Opportunities for Research and Practice

“…With research increasingly showing the valuable insights that social media data can yield about mental health states, greater attention to the ethical concerns with using individual data in this way is necessary (Chancellor et al 2019). For instance, data is typically captured from social media platforms without the consent or awareness of users (Bidargaddi et al 2017), which is especially crucial when the data relates to a socially stigmatizing health condition such as mental illness (Guntuku et al 2017).…”

“…Precautions are needed to ensure that data is not made identifiable in ways that were not originally intended by the user who posted the content as this could place an individual at risk of harm or divulge sensitive health information (Webb et al 2017;Williams et al 2017). Promising approaches for minimizing these risks include supporting the participation of individuals with expertise in privacy, clinicians, and the target individuals with mental illness throughout the collection of data, development of predictive algorithms, and interpretation of findings (Chancellor et al 2019).…”

Machine Learning in Mental Health

“…[ 33 , 105 , 211 , 219 ] as a mechanism for "diagnostic" ground truth, as this does not conform with clinical assessment tools such as the DSM [ 8 ]. The DSM provides a written manual for making accurate psychiatric diagnosis that is based on 60 years of empirical results [ 29 ]. Concerns about a lack of clinical grounding, theoretical contextualization, and psychometric validity were particularly prominent in the paper by Ernala et al [ 53 ].…”

“…For many of the social media studies papers, the pooling of "publicly" available data [e.g., 141 , 165 , 211 ] has often been described as not requiring explicit user consent. Recently, there is however increasing debate on whether the use of public data to predict, e.g., mental health states, may border on medical diagnosis and should be considered as human subjects research [ 29 ]. This is echoed in user research that suggests that social media users often do not have awareness that their online content is used for research, and express concerns about such use "without their consent" [ 58 ].…”

“…Even in less extreme cases, the false identification of a mental health condition could have severe implications for a person's self-esteem, reputation or employment (particularly for people working, i.e., in the police force, as pilots) [ 29 , 189 ]. This raises the question of who is responsible and accountable for errors and for making choices if and how individuals should be alerted to their own mental health status [ 29 ]. Within the broader literature on HCI in digital mental health, researchers have started to discuss the challenges involved in making people aware of machine-detected problems in ways that are sensible, and respond carefully to peoples' expectations, needs, or troubles (e.g .…”