Clustering semantic spaces of suicide notes and newsgroups articles

Abstract

Historically, suicide risk assessment has relied on question-and-answer type tools. These tools, built on psychometric advances, are widely used because of availability. Yet there is no known tool based on biologic and cognitive evidence. This absence often cause a vexing clinical problem for clinicians who question the value of the result as time passes. The purpose of this paper is to describe one experiment in a series of experiments to develop a tool that combines Biological Markers (Bm) with Thought Markers (Tm), and use machine learning to compute a real-time index for assessing the likelihood repeated suicide attempt in the next six-months. For this study we focus using unsupervised machine learning to distinguish between actual suicide notes and newsgroups. This is important because it gives us insight into how well these methods discriminate between real notes and general conversation.