Learning in Context: Personalizing Educational Content with Large Language Models to Enhance Student Learning DOI: XXXXXXX.XXXXXXX ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Applied computing Education CCS: Human-centered computing User studies CCS: Information systems Personalization

Joy Jia Yin Lim, Daniel Zhang-Li, Jifan Yu, Xin Cong, Ye He, Zhiyuan Liu, Huiqin Liu, Lei Hou, Juanzi Li, Bin Xu · 2025

Read Learning in Context: Personalizing Educational Content with Large Language Models to Enhance Student Learning DOI: XXXXXXX.XXXXXXX ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Applied computing Education CCS: Human-centered computing User studies CCS: Information systems Personalization by Joy Jia Yin Lim, Daniel Zhang-Li, Jifan Yu, Xin Cong, Ye He, Zhiyuan Liu, Huiqin Liu, Lei Hou, Juanzi Li, Bin Xu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-15068.txt.

Introduction

Joy Jia Yin Lim leads a team of ten authors personalizing school content with language models in a 2025 education paper. Standard materials speak to no one in particular, and students disengage when examples ignore their backgrounds and interests. Their framework, PAGE, rewrites educational content around each student's context, such as major and hobbies.

PAGE was deployed inside a semester-long intelligent tutoring system and tested in a real course. Students receiving personalized materials learned measurably more than peers with standard content. They also reported higher engagement, stronger perceived relevance, and more trust in the system. Beyond the headline gains, the study draws design lessons for personalization tools: how much adaptation helps, where it can misfire, and how teachers might supervise it. The work shows personalization working in an authentic setting rather than a laboratory demo.

Catalog record

CREATORJoy Jia Yin Lim, Daniel Zhang-Li, Jifan Yu, Xin Cong, Ye He, Zhiyuan Liu, Huiqin Liu, Lei Hou, Juanzi Li, Bin Xu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-15068

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Learning in Context: Personalizing Educational Content with Large Language Models to Enhance Student Learning DOI: XXXXXXX.XXXXXXX ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Applied computing Education CCS: Human-centered computing User studies CCS: Information systems Personalization Joy Jia Yin Lim, Daniel Zhang-Li, Jifan Yu, Xin Cong, Ye He, Zhiyuan Liu, Huiqin Liu, Lei Hou, Juanzi Li, Bin Xu Abstract . Standardized, one-size-fits-all educational content often fails to connect with students’ individual backgrounds and interests, leading to disengagement and a perceived lack of relevance. To address this challenge, we introduce PAGE , a novel framework that leverages large language models (LLMs) to automatically personalize educational materials by adapting them to each student’s unique context, such as their major and personal interests. To validate our approach, we deployed PAGE in a semester-long intelligent tutoring system and conducted a user study to evaluate its impact in an authentic educational setting. Our findings show that students who received personalized content demonstrated significantly improved learning outcomes and reported higher levels of engagement, perceived relevance, and trust compared to those who used standardized materials. This work demonstrates the practical value of LLM-powered personalization and offers key design implications for creating more effective, engaging, and trustworthy educational experiences. 1. Introduction Standardized, one-size-fits-all educational content is failing to engage students ( wu2024comprehensiveexplorationpersonalizedlearning ) . As shown in Figure 1 , in both traditional classrooms and dominant online learning platforms, learners are often presented with generic materials that disregard their individual backgrounds, interests, and real-world contexts ( Collins and Halverson 2010 ; Collins and Halverson 2018 ) . This disconnect frequently leads to a "crisis of relevance" and "affective disengagement," where students struggle to see the value of abstract knowledge in their own lives ( Yang et al. 2019 ; Bondie et al. 2019 ) . While the vision of personalized education has been a long-standing goal in educational technology ( wu2024comprehensiveexplorationpersonalizedlearning ; Maghsudi et al. 2021 ; Taber 2008 ) , the resource-intensive nature of preparing personalized content have been a major barrier to its widespread adoption ( Bernacki and Walkington 2018 ) . This leaves a critical gap between the potential of personalized learning and the reality of current educational practices, resulting in diminished student motivation and learning outcomes ( Smith and Ragan 2004 ; Pekrun 2006 ; Pekrun and Perry 2014 ) . Figure 1. Comparison between standardized content for all students in traditional educational paradigms (including physical classrooms and online learning platforms) and personalized content for individual students. Recent advances in large language models (LLMs) present new possibilities to deliver personalized learning at scale ( Walkington and Bernacki 2014 ; Harackiewicz et al. 2016 ; Reber et al. 2018 ; Bernacki et al. 2021 ) , with their remarkable capabilities in contextual understanding, content generation, and knowledge synthesis ( Peláez-Sánchez et al. 2024 ; Gan et al. 2023 ) . While numerous studies have explored the potential of LLMs in various educational applications ( Waldeck 2007 ; Ramesh et al. 2024 ) , existing approaches often provide only superficial personalization, such as adjusting content complexity, rather than achieving deep contextual adaptation ( He-Yueya et al. 2024 ; Chen et al. 2023 ; Park et al. 2024 ) . This technological gap points to a significant design challenge for the community: How can we leverage the power of LLMs to create educational experiences that are not just algorithmically optimized, but are also genuinely resonant and meaningful to each individual? To address this challenge, we introduce PAGE , a retrieval-augmented framework for P ersonalized educ A tional content GE neration. Our research aims to explore the potential of LLMs in generating educational content with high ecological validity and pedagogical efficacy, by deep understanding and adapting to each student’s unique context. Building upon the demonstrated effectiveness of high-quality, human-authored educational materials, we investigate whether and how LLM-generated personalized content can further enhance student learning experience. We specifically focus on the following research questions: • RQ1 : To what extent does interacting with context-aware personalized content affect students’ knowledge acquisition compared to standardized materials? • RQ2 : How does the personalized content shape a student’s engagement, their perception of the material’s relevance, and their trust in the learning system? • RQ3 : What are the mechanisms through which this personalization influences a student’s learning process and their overall perception of the educational experience? To answer these questions, we conducted a comprehensive user study with 40 university students from diverse academic backgrounds. We integrated PAGE into a semester-long intelligent tutoring system ( Yu et al. 2024 ) and compared the effects of the personalized content generated by our system against human-authored standardized materials. We gathered both quantitative data on learning outcomes and rich qualitative insights into the students’ subjective experiences. Our contribution are as follows: (1) We introduce PAGE , a novel framework that generates personalized educational content by adapting standardized materials to a student’s unique background and interests. (2) We conduct comprehensive user study and provide robust evidence that our personalized content improves both learning outcomes and student engagement when compared to high-quality standardized materials. (3) We demonstrate the practical applicability of LLM-generated content in an authentic, real-world educational setting, distilling key design implications for creating trustworthy, context-aware learning experience. 2. Related Work This section briefly reviews the theoretical foundations of personalized content generation, the impact of LLMs for enhanced personalization, and the development of intelligent tutoring systems. 2.1. Personalized Content Generation Personalization aims to tailor educational content to the unique needs, goals, and contexts of individual students ( King and South 2017 ; Alamri et al. 2020 ; Belenky and Schalk 2014 ) . Early attempts explored a variety of applications, including template-based keyword insertion ( Høgheim and Reber 2015 ; Høgheim and Reber 2017 ) , character customization in educational media ( Pataranutaporn et al. 2022 ) , and static or rule-based student modeling ( Ku et al. 2007 ; López and Sullivan 1992 ) . While these methods demonstrated potential applicability, they often required considerable manual effort, lacked scalability, and struggled to capture the complexity and variability of student contexts ( Frenzel et al. 2012 ) . We aim to address these limitations by advancing scalable, deeper personalization to enable more adaptive learning experiences. 2.2. LLMs for Enhancing Personalization The recent advent of LLMs presents transformative potential for personalized education ( Wen et al. 2024 ; Razafinirina et al. 2024 ) . Some reframed standard problems (e.g., algebra word problems) using contexts derived from student interests ( Yadav et al. 2023 ) , while others focused on improving specific tasks such as vocabulary learning ( Leong et al. 2024 ) . Systems like PACE ( Liu et al. 2025 ) leveraged LLMs to create personalized conversational tutors for predefined learning styles. Despite the promising results, existing approaches have focused on domain-specific applications ( Leong et al. 2024 ) or particular aspects of personalization, such as surface-level contextualization or adaptation using predefined learning style models. Less attention has been paid to dynamically adapting standardized content to individual learning contexts. Our work aims to address this gap. 2.3. Intelligent Tutoring Systems (ITS) ITS represents a significant advancement in personalized education ( Nye et al. 2014 ; Bubeck et al. 2023 ) , as

Read the complete text