Using natural language processing to classify suicide notes

Abstract

We hypothesize that machine-learning algorithms (MLA) can classify completer and simulated suicide notes as well as mental health professionals (MHP). Five MHPs classified 66 simulated or completer notes; MLAs were used for the same task. Results: MHPs were accurate 71% of the time; using the sequential minimization optimization algorithm (SMO) MLAs were accurate 78% of the time. There was no significant difference between the MLA and MPH classifiers. This is an important first step in developing an evidence based suicide predictor for emergency department use.

“You’re as Sick as You Sound”: Using Computational Approaches for Modeling Speaker State to Gauge Illness and Recovery

“…Besides the annotation costs, the validity of the annotation is important. In several studies, expert opinions are not always reliable [42,45], but neither are patients' self-reports [5].…”

“…The goal of such an analysis is to gain deeper understanding of the psychological state of individuals committing suicide, as well as to help prevent suicide of such individuals. Pestian et al [45] envision a screening tool in place at a psychiatry emergency room that predicts the likelihood of an individual being in a suicidal state rather than merely depressed. The authors present preliminary results on the use of linguistic analysis when applied to suicide notes.…”

“…Free speech, albeit about a particular topic [5] 3. Proxy texts [45] 4. Specific tasks such as imitation, where subjects repeat words or sentence read to them [51].…”

“…Numerous experiments have been conducted on the automatic classification of the classic emotions, such as anger, happiness, sadness; secondary emotions such as confidence or annoyance; or simply positive from negative emotions, from acoustic, prosodic, and lexical information [39,50,3,9,26,32,33,1]. Motivation for these studies have come primarily from call center and Interactive Voice Response (IVR) applications, for which there is interest in distinguishing angry and frustrated callers from the rest, either to hand them off to a human attendant or to flag such conversations as problematic for off-line study [26,32] [36], romantic interest [45], "hotspots" in meetings (Wrede & Shriberg 2003), and confusion (Kumar et al. 2006).…”

Voice Parameter Analysis for the disease detection

“…Alpert et al [6] stated that clinical impressions were substantially related to acoustic parameters like fluency and prosody of depression patients. Pestian et al [21] designed a screening tool that predicts the likelihood of an individual being in depression or in suicidal state. The authors implied the results of linguistic analysis from an act of writing a suicide note from individuals who had committed suicide and normal individuals to extract features: word count, pronouns, anagrams, emotional words, etc.…”