Human and large language model classifications of hospice questions: data and analysis code
Year: 2026
Type: dataset
Abstract: Data and analysis code for an exploratory study comparing how human respondents and large language models interpret the dominant intent of questions asked in hospice care. Seventeen Polish-language questions were reconstructed from recurrent hospice encounters by three experienced hospice psychologists. Each was assigned to one of four intent categories — emotional, fear-related and spiritual; family and close others; health and treatment; daily life and practical — by 138 human respondents, including nine hospice clinicians, and by eight large language model configurations from four developers, each evaluated in five independent sessions. The deposit contains the anonymised human classifications (138 × 17 = 2,346), the run-level model classifications (8 × 5 × 17 = 680), the question set with English translations and human consensus rates, a codebook, and the analysis code. A short verification script recomputes every headline result reported in the associated article directly from the deposited files. The human survey was anonymous. Response timestamps were removed, year of birth was replaced by an age group, and the free-text occupation item was recoded into the four categories reported in the article, to prevent re-identification within the recruitment networks.
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7204668153
