Reflective Storytelling Agent for Older Adults Using LLMs and Argument Mining
A new study published on arXiv introduces a reflective storytelling agent designed to support older adults through purposeful narrative interactions with digital companions. Addressing common Large Language Model (LLM) limitations like hallucinations and lack of transparency, the system integrates knowledge graphs, user modeling, argumentation theory, and argument mining. The research involved two phases: a formative evaluation with eleven domain experts and a user study with fifty-five older adults who assessed persona-based narratives. Results indicated that participants recognized personally relevant purposes in approximately two-thirds of the generated stories. Cultural recognisability significantly influenced willingness to use the technology, while minor inconsistencies were tolerated if narratives remained meaningful. The study found that higher hallucination-risk indicators correlated with perceived inconsistencies, whereas high argument-quality indicators aligned with better clarity and meaningfulness. This work positions argument mining as a crucial reflective inspection mechanism, bridging formal grounding signals with human evaluations in health-oriented AI storytelling, ultimately aiming to enhance the reliability and personal relevance of digital companions for elderly care.
Wire timeline
Reflective Storytelling Agent for Older Adults Using LLMs and Argument Mining
A new study published on arXiv introduces a reflective storytelling agent designed to support older adults through purposeful narrative interactions with digital companions. Addressing common Large Language Model (LLM) limitations like hallucinations and lack of transparency, the system integrates knowledge graphs, user modeling, argumentation theory, and argument mining. The research involved two phases: a formative evaluation with eleven domain experts and a user study with fifty-five older adults who assessed persona-based narratives. Results indicated that participants recognized personally relevant purposes in approximately two-thirds of the generated stories. Cultural recognisability significantly influenced willingness to use the technology, while minor inconsistencies were tolerated if narratives remained meaningful. The study found that higher hallucination-risk indicators correlated with perceived inconsistencies, whereas high argument-quality indicators aligned with better clarity and meaningfulness. This work positions argument mining as a crucial reflective inspection mechanism, bridging formal grounding signals with human evaluations in health-oriented AI storytelling, ultimately aiming to enhance the reliability and personal relevance of digital companions for elderly care.
cs.AI updates on arXiv.org