|
AI assisted grading framework for Thai-language written exam questions based on LLM and rule-based reasoning approach |
|---|---|
| รหัสดีโอไอ | |
| Title | AI assisted grading framework for Thai-language written exam questions based on LLM and rule-based reasoning approach |
| Creator | Chutipon Trirattananurak |
| Contributor | Sasiporn Usanavasin, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | Exam evaluation, Natural language processing (NLP), Large language models (LLMs), Thai language, Rule-based reasoning, Educational assessment, การประเมินข้อสอบ, การประมวลผลภาษาธรรมชาติ, แบบจำลองภาษาขนาดใหญ่, ภาษาไทย, การอนุมานเชิงกฎ, การประเมินผลทางการศึกษา |
| Abstract | Assessment plays a central role in education, providing insights into student comprehension, learning progress, and curriculum effectiveness. Written exams are particularly valuable as they capture higher-order cognitive skills such as reasoning, problem-solving, and expressive ability. However, manual grading of written responses is both time-consuming and prone to inconsistency due to factors such as evaluator fatigue, subjective interpretation, and variations in teaching assistant training. In the Thai context, these challenges are amplified by the language's complex grammar, tonal system, and context-dependent exceptions, which complicate evaluation and increase the likelihood of bias. Consequently, many educators rely heavily on multiple-choice examinations, which are easier to grade but insufficient for assessing deeper levels of understanding.Recent advancements in Natural Language Processing (NLP), especially the development of Large Language Models (LLMs), provide new opportunities to address these challenges. LLMs offer sophisticated contextual analysis and semantic understanding that go beyond the limitations of traditional rule-based or keyword-based grading systems. However, their performance may still be unpredictable without structured guidance. This thesis proposes a hybrid AI-assisted grading framework that integrates the contextual reasoning capabilities of LLMs with the transparency and consistency of rule-based reasoning. The framework is designed specifically for Thai-language written exam questions, aiming to reduce grading workload, enhance fairness, and support wider adoption of open-ended assessments in Thai education.Through experimental evaluation, the framework demonstrates its ability to align closely with human grading while minimizing inconsistency. The results highlight its potential as a practical solution for teachers facing heavy workloads, as well as a step toward more equitable and comprehensive assessment methods in Thai education. This research contributes both a novel methodology for exam evaluation and empirical insights into the broader use of AI in educational contexts. |
