Decoding loneliness: Can explainable AI help in understanding language differences in lonely older adults?

Area of Science:

• Gerontology
• Computational Linguistics
• Artificial Intelligence

Background:

• Loneliness significantly impacts older adults' health.
• Effective, targeted interventions for loneliness are scarce.
• Speech data offers a personalized alternative to surveys for assessing loneliness.

Purpose of the Study:

• To explore Natural Language Processing (NLP) and AI for identifying linguistic markers of loneliness in older adults.
• To distinguish between lonely and non-lonely individuals using speech data.
• To validate the use of explainable AI (XAI) in loneliness detection.

Main Methods:

• Collected speech data from 97 older adults via semi-structured interviews.
• Utilized Linguistic Inquiry and Word Count (LIWC-22) for linguistic feature extraction.
• Developed an XAI model to classify loneliness based on speech sections like social relationships and loneliness.

Main Results:

• The XAI model achieved high accuracy (0.889) and recall (1.0) in detecting loneliness.
• Speech sections on social relationships and loneliness were most predictive.
• Key features included social themes, pronoun usage, and conversational fillers.

Conclusions:

• Explainable AI (XAI) effectively detects loneliness from unstructured speech.
• Speech analysis provides valuable insights into the lived experience of loneliness.
• This method offers a novel approach for identifying loneliness in older populations.