Original Research
Personalized Applications of Large Language Models in Pre-University Computer Science Education: Bridging Global Equity, Mitigating Bias, and Addressing Structural Challenges
- Abstract
- Full text
- Metrics
The rapid development of Large Language Models (LLMs) has offered new opportunities for personalized education while raising concerns about bias amplification, equity gaps, and the digital divide. This quasi-experimental study explores LLM applications in high school Computer Science (CS) education, focusing on bias mitigation and bridging to higher education. A sample of 616 students from a key high school in northern China was randomly assigned to an experimental group (LLM personalized instruction, n=308) or a control group (traditional instruction, n=308). The 8-week intervention built CS skills through progressive tasks from basic syntax to comprehensive projects, aligned with university domains (e.g., ACM CS2023). Methods integrated UNESCO (2025) and OECD (2025) guidelines, emphasizing low-bandwidth solutions and bias mitigation strategies such as prompt engineering and multi-model comparison. Data analysis combined quantitative approaches (t-tests, ANOVA, bridging index) and qualitative NVivo thematic coding to assess performance gains, subgroup equity, and bias indicators (MAB and MDB). Results showed significant improvements in the LLM group (p<0.05), with a 15% average bridging index and 80%±5% bias mitigation rate. However, urban-rural and gender biases still require further optimization. This study provides empirical insights into responsible LLM use in education and proposes policy frameworks and model optimization (unified super models vs. mixture of experts) to advance global equity and ethical AI integration.
Page 1 of
Download Count : -
Visit Count : 46
Keywords
Large Language Models; Computer Science Education; Personalized Learning; Bias Mitigation; Educational Equity; Digital Divide; Intelligent Tutoring
Author(s) Information
Publisher’s Note
The claims, arguments, and counter-arguments made in this article are exclusively those of the contributing authors. Hence, they do not necessarily represent the viewpoints of the authors’ affiliated institutions, or EUROKD as the publisher, the editors and the reviewers of the article.
Acknowledgements
Not applicable.
Funding
Not applicable
CRediT Authorship Contribution Statement
Yang Xia: Conceptualization, Methodology, Investigation, Resources, Data Curation, Writing - Original Draft, Writing - Review & Editing, Visualization, Supervision, Project administration, Funding acquisition.
Generative AI Use Disclosure Statement
Generative AI tools (primarily Baidu Wenxin Yiyan/Ernie Bot, supplemented by Alibaba Tongyi Qianwen and iFlyTek Xinghuo) were used in a limited and transparent manner during the preparation of this manuscript. Specifically, AI was employed for the following purposes only: Generating and iteratively refining Socratic dialogue prompts and personalized task examples used in the experimental intervention (detailed in the Methodology section); assisting with the development and testing of the multi-agent framework for prompt stability evaluation (including cosine similarity calculations in the Tools and Settings subsection); performing minor linguistic polishing, copyediting, and formatting improvements to enhance readability and academic tone (corresponding to the Writing – Review & Editing role in the CRediT statement). All AI-generated content was thoroughly reviewed, verified, revised, and edited by the author. No generative AI was used for data collection, statistical analysis, interpretation of results, qualitative coding, drawing conclusions, or the creation of original intellectual content. The author takes full responsibility for the accuracy, integrity, and originality of the final manuscript. This disclosure follows the APA journals policy on generative AI use and the journal’s requirements for transparency.
Ethics Declarations
World Medical Association (WMA) Declaration of Helsinki–Ethical Principles for Medical Research Involving Human Participants
This study was approved by the school's ethics committee (approval no. CZXZ-2025-001, dated August 15, 2025), complying with Chinese laws and UNESCO (2025) principles. All participants provided informed consent; participation was voluntary with no coercion.
Competing Interests
The authors declare no competing interests.
Data Availability
All raw data, analysis materials, and supplementary resources (e.g., questionnaires, code, statistical outputs, course materials) are archived on Zenodo for transparency and replicability, anonymized per China's Personal Information Protection Law (2021) and UNESCO (2025) guidelines. Access: https://doi.org/10.5281/zenodo.17571306. Contact the author for queries.