薬学および生命科学における研究開発の国際ジャーナル

抽象的な

COMPUTATIONAL ANALYSIS OF HINDI SUICIDE NOTES USING SNARE

Subhash Chandra, Carole E. Chaski

Suicide notes play a pivotal role in death investigation. SNARE (Suicide Note Assessment REsearch) software classifies texts as a suicide note or control text-type with accuracies from 80%-88%, depending on text length, from a database of about 1,000 English sources. Our objective was a pilot study to determine SNARE’s reliability among a non-English-speaking sample. Suicide notes were collected in New Delhi from 33 cases of suicide confirmed by autopsy findings, psychological autopsy, inquest paper, and crime-scene investigation. Thirteen out of fifteen legible notes were translated from Hindi into English and run through the SNARE algorithm. The software classified 8 of 13 translated texts as suicide notes (61.5%) and five as control texts (38.5%). All the misclassified notes were longer than 80 words. We concluded that even with the limitations of translation and lower accuracy of SNARE at high word count, the accuracy of 61.5% is greater than that of humans in differentiating between genuine and simulated suicide notes.