Personalizing Reading Content for Children Utilizing Generative AI

Abstract

In recent years, a noticeable decline in reading motivation and literacy levels of children has resulted in significant challenges for the educational sector. This thesis explores the potential of Generative Artificial Intelligence (GenAI) to personalize reading content for children with the goal of improving reading motivation and supporting reading development. A review of scientific and commercial projects provides an overview of current Technology-enhanced Learning (TEL) approaches applying different GenAI-based personalization techniques, including context personalization based on interests, example choice through additionally generated learning material, and active personalization where children interact directly with GenAI systems to adapt the learning material independently. Based on these principles, a child-friendly Android tablet application was developed with the goal to personalize reading content according to the preferences and interests of the child. Besides content personalization, various reading challenges were integrated to enhance motivation and engagement. Three challenge types were designed, “remove unnecessary word”, “story questions”, and “order story chronologically”. Two personalization modes were implemented using predefined prompts to personalize content and generate additional challenges. The evaluation showed that both tested models, OpenAI’s GPT-4.1 and Google’s Gemini-2.0-Flash, generated personalized content that reflected the interest of the child while preserving the storyline. However, the generated text often contained unnatural word combinations, an issue that was particularly evident in the outputs of the Large Language Model (LLM) from Google. While text difficulty adaptation was successful, the personalized texts often became more difficult to read. The evaluation further revealed that “story questions” and “order story chronologically” challenges were reliably generated, whereas the “remove unnecessary word” challenge frequently produced unsolvable outputs, making it unsuitable for GenAI-based generation. The results demonstrate the potential of GenAI to personalize reading content while also identifying existing limitations and opportunities for improvement. Future work could explore implementing adaptive difficulty levels, dynamic GenAI model selection, additional challenge formats, the integration of gamification elements, and incorporating multi-modal features such as audio and video to create more engaging reading experiences.

Date of Award	2025
Original language	English
Supervisor	Jens Krösche (Supervisor)