Characterizing LLM-Empowered Personalized Story-Reading and Interaction for Children: Insights from Multi-Stakeholder Perspectives

TL;DR

The paper investigates how large language models can support personalized, child centered story reading and interaction. Using a design-based approach, the authors develop StoryMate, a GPT-4 powered, RAG-enhanced tool with four design goals and corresponding features to tailor reading content, conversation, attention engagement, and knowledge extension to children. A formative study informs the design, and a technology probe study with 12 children, 14 parents, and 13 experts reveals generally positive experiences alongside multi-dimensional personalized requirements, including content topics, language complexity, guidance timing, and interface preferences. The work offers design recommendations for scalable, ethically considerate deployment of LLM-powered personalized storytelling tools that balance child autonomy with guided inquiry and consider family and school contexts. It also highlights opportunities and challenges in integrating LLMs into children's reading practices and contributes empirical insights into stakeholder perceptions of such technologies in real-world settings.

Abstract

Personalized interaction is highly valued by parents in their story-reading activities with children. While AI-empowered story-reading tools have been increasingly used, their abilities to support personalized interaction with children are still limited. Recent advances in large language models (LLMs) show promise in facilitating personalized interactions, but little is known about how to effectively and appropriately use LLMs to enhance children's personalized story-reading experiences. This work explores this question through a design-based study. Drawing on a formative study, we designed and developed StoryMate, an LLM-empowered personalized interactive story-reading tool for children, following an empirical study with children, parents, and education experts. Our participants valued the personalized features in StoryMate, and also highlighted the need to support personalized content, guiding mechanisms, reading context variations, and interactive interfaces. Based on these findings, we propose a series of design recommendations for better using LLMs to empower children's personalized story reading and interaction.

Figures (8)

• Figure 1: StoryMate's interfaces. a) Library interface, which shows customized reading content. b) Greeting interface, wherein a robot guides children's self-introduction. c) Customizable story-reading mode interface, through which children can set read modes. d) Dashboard interface, which records children's reading activities, content, etc.
• Figure 2: The interaction process of StoryMate. Receiving children's information in stage 1, StoryMate retrieves story-situated information (stage 2). For story-based questions, we use story (A), child's information (B), and summarized story narratives (D) as input. For knowledge-extending questions, we add matched knowledge (C). Then, in stage 3, StoryMate interacts with children and actively updates conversation status (E). A conversation example is shown at the right end.
• Figure 3: An example of the architecture of the Next Generation Science Standards (NGSS) national2013next. We manually organize the DCIs alongside their corresponding Performance Expectations to build our knowledge source.
• Figure 4: The technical process behind StoryMate's knowledge infusion. We use a fine-tuned retriever to identify the most educationally appropriate piece of knowledge and keyword. The word-knowledge pair is then integrated into the prompt.
• Figure 5: The prompt design for StoryMate to greet children. The input prompt consists of four components: 1) Task Summary, 2) Generation Requirements, 3) Format Setting, and 4) Conversation History.