WISE Causal Models: Wisdom Infused Semantics Enhanced Causal Models - A Study in Suicidality Diagnosis

Anonymous

Keywords: Human AI Collaboration, Causal Models, Mental Health

TL;DR: Learning better causal models through human knowledge bootstrapped model averaging

Abstract: The COVID-19 Pandemic has highlighted the gap between the number of mental health care seekers and care providers. Netizens have taken to internet-based platforms such as Reddit to express their experiences. Mental illness diagnosis processes have clinically accepted causal interpretations and semantics. Curiously, mental illness diagnosis accuracy is low relative to similar well-studied illnesses. Motivated by this discrepancy, we propose Wisdom Infused Semantics Enhanced (WISE) causal models, inspired by the wisdom of the crowd idea that learns from a collective agreement among causal models and their semantics for mental illness diagnoses. We use suicidality diagnosis task descriptions, datasets, and baseline methods to evaluate the effectiveness of WISE causal models. Our experiments show that learning WISE causal models improve performance on these tasks.