The researchers interviewed 80 independent senior living residents, aged 66-94, by asking 20 questions from the UCLA Loneliness Scale. They also taped personal conversations with them that were manually transcribed and examined with natural language processing tools.
Senior author Ellen Lee said the study relied on quantitative natural language processing along with the usual loneliness measurement tools to look for not just certain words, but the patterns in the words. Among the tools used to analyze the transcripts was IBM’s deep learning Watson NLP software.
The results showed that the AI could gauge loneliness symptoms almost as accurately as self-reports and questionnaires. Women were also more likely than men to acknowledge feeling lonely.
The study was published in the American Journal of Geriatric Psychiatry.