A Reflective Storytelling Agent for Older Adults: Integrating Argumentation Schemes and Argument Mining in LLM-Based Personalised Narratives
In an innovative approach to enhance the interaction between older adults and digital companions, recent research has introduced a reflective storytelling agent utilizing large language models (LLMs). This new technology aims to provide purposeful narrative experiences tailored to the needs and preferences of older adults, addressing existing challenges such as hallucinations and transparency in LLM outputs.
Research Overview
The study, documented in arXiv:2605.10531v1, investigates the potential of knowledge-driven LLMs in supporting engaging storytelling. The research integrates several advanced methodologies, including:
- Knowledge Graphs
- User Modelling
- Argumentation Theory
- Argument Mining
This multi-faceted approach aims to enhance narrative generation by ensuring that stories are both relevant and coherent, ultimately fostering a more meaningful interaction between older adults and their digital companions.
Study Phases
The research was conducted in two distinct phases aimed at refining the storytelling agent and evaluating its effectiveness:
Phase I: Participatory Design
The initial phase involved a participatory design process that included 11 domain experts. Their insights provided a foundation for the iterative refinement of the storytelling system. The system was designed to generate narratives based on structured user models that represent health-promoting activities and individual motivations.
Phase II: User Evaluation
In the second phase, 55 older adults were engaged to evaluate the persona-based narratives generated by the system. Participants assessed stories across four prompts and two creativity levels, focusing on key attributes such as:
- Perceived Purpose
- Usefulness
- Cultural Relatability
- Inconsistencies
Additionally, the system computed hallucination-risk indicators to further assess the quality and reliability of the generated narratives.
Key Findings
The evaluation revealed several significant insights:
- Approximately two-thirds of the narratives recognized personally relevant purposes, showcasing the system’s ability to tailor stories to individual needs.
- Argument-based purposes were identified in about half of these narratives, indicating a successful integration of argumentation theory.
- Cultural recognizability was a critical factor influencing participants’ willingness to engage with the storytelling functionality.
- Minor inconsistencies were generally accepted as long as the narratives remained understandable and personally relevant.
- Narratives that exhibited higher hallucination-risk indicators were more frequently perceived as inconsistent.
- Higher argument-quality indicators correlated with better clarity and meaningfulness ratings from participants.
Conclusion
This study positions argument mining as a valuable tool for reflective inspection in health-oriented LLM storytelling for older adults. By comparing formal grounding signals with human evaluations, researchers can enhance narrative quality and ensure that digital companions effectively meet the needs of older users. The findings underscore the potential of personalized storytelling to enrich the lives of older adults, fostering a deeper connection with technology in an increasingly digital world.
Related AI Insights
- Enhance LLMs Structural Attention with Slash Method
- Elementary OS vs Linux Mint: Best User-Friendly Linux Distro
- EmbodiSkill: Adaptive Skill Evolution for Embodied Agents
- PrimeKG-CL: Benchmark for Continual Learning on Biomedical Graphs
- Medicare’s ACCESS Model Revolutionizes AI in Healthcare
- EGL-SCA: Advanced Graph Reasoning with Dual-Space Framework
- Verifiable Process Rewards Boost Agentic Reasoning in AI
- GuardAD: Enhancing Autonomous Driving Safety with Markov Logic
- E-TCAV: Efficient Concept-Based Neural Network Interpretability
- IndustryBench: Benchmarking LLMs for Safe Industrial QA
