Suicide note Bayes filter

Many people come across suspected suicide notes regularly on the Internet, but we really have no ability to differentiate real from angst from satire. Not only is our false negative rate extremely high, we’re encumbered by groupthink failings such as believing someone else must’ve already contacted the authorities.

Train a Bayes filter to differentiate suicide notes that predate a successful attempt from ones that do not. Psychiatrists and medical personnel could feed in data from unreleased successes. Privacy would be maintained even if the database were compromised because of the nature of Bayesian filtering.

A spidering robot could scour the entire web: diaries, message boards, and comments for potentially legitimate suicide notes and alert the proper authorities to intervene. It may even pick up on the tone of writing not meant to be a suicide note.

Suicide is a public health issue, and automatically examining the Internet as a human would, only faster and with better tools is not at all dissimilar from walking the halls of a high school looking for notes left behind by people hoping for them to be read.

Examining the filter itself may lead us to many insights on suicide itself. Psychiatrists could also use the filter to examine their own patients’ writings to detect changes they might overlook.

To deal with the liabilities of a false negative, the spider will not allow human direction, nor will it keep logs. There should be no way to determine if the spider has visited a page, so no one will feel absolved of responsibility and the administrators will not be liable, nor should anyone be able to point the spider to a page for the same reasons.