Practical Exploration of Using Large Language Models to Automatically Generate Feedback for Undergraduate Thesis Introductions

This article introduces a 2026 master's research project that explores how to use large language models (LLMs) to provide automated feedback for the introduction section of undergraduate theses. The project aims to address the pain points of traditional manual feedback, such as limited time, inconsistent quality, and long waiting periods, and offers new ideas for the development of academic writing assistance tools. The research covers aspects like system development, technical implementation, application scenarios, and challenges, and has important reference value for educational technology.

Academic writing is a core skill in higher education. As the "face" of a thesis, the introduction affects the first impression of research value, but undergraduates often face issues like disorganized structure, unclear logic, and ambiguous problem statements. Traditional manual feedback relies on supervisors, which has limitations such as limited time, large variations in feedback quality, and long waiting periods, spurring the demand for automated feedback tools.

In recent years, LLMs (such as the GPT series and Claude) have made breakthroughs in natural language understanding and generation. They can understand complex academic text structures, identify logical flaws, and provide improvement suggestions. Compared with traditional rule-based tools, LLMs can understand contextual information, evaluate the coherence of arguments and clarity of problems, and provide targeted rather than template-based feedback, laying a technical foundation for intelligent academic writing assistance tools.

The 2026 master's research project focuses on using LLMs to automatically generate feedback for undergraduate thesis introductions, and has developed a system that receives students' introduction texts, analyzes them via LLMs, and returns structured feedback. The core innovation lies in combining the general capabilities of LLMs with the specific needs of academic writing, which requires solving problems such as technical implementation, prompt design, and feedback quality evaluation.

The system includes four key components: 1. Text input and preprocessing module (format cleaning, segmentation); 2. LLM interaction layer (communication with APIs, prompt engineering is a core challenge); 3. Feedback structuring processing (conversion into classification labels such as "structural suggestions" and priority marking); 4. Quality evaluation mechanism (comparison with expert feedback, student satisfaction surveys, improvement effect tracking, etc.).

The automated feedback system has broad application prospects: For students, instant personalized guidance reduces dependence on supervisors and cultivates independent revision abilities; For supervisors, it frees up energy to focus on high-level guidance and improves discussion efficiency; For educational institutions, it standardizes the quality of writing teaching and reduces educational disparities caused by uneven resource allocation.

The application faces four major challenges: 1. Accuracy and reliability (avoiding model "hallucination" and incorrect suggestions); 2. Balance between personalization and universality (adapting to different disciplines/supervisor requirements); 3. Educational ethics (preventing over-reliance from affecting independent thinking); 4. Privacy and data security (complying with academic ethics and data protection regulations).

This research represents an interesting direction for the application of AI in higher education, exploring paths to improve educational efficiency by combining LLM capabilities with academic writing needs. In the future, we can expect more intelligent and personalized tools that not only provide feedback but also guide academic thinking training, becoming "intelligent academic supervisors". This open-source project provides a reference implementation for related research/development and is worth learning from.