그래프 RAG를 활용한 문법 오개념 진단 및 서술형 평가 지원 시스템 프로토타입 연구
A Prototype Study of a Grammar Misconception Diagnosis and Short-Answer Assessment Support System Using Graph RAG
3
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study explores the Graph Retrieval-Augmented Generation (Graph RAG) methodology to address challenges in grammar education—diagnosing learner misconceptions and assessing short-answer responses—by controlling LLM hallucinations. A grammar Knowledge Graph (KG) with 56 concepts was constructed, serving as a base for two prototypes: a ‘Grammar Misconception Diagnosis Model’ and an ‘Short-Answer Assessment Support Model.’ Both utilize a 3-stage reasoning architecture(Key Term Extraction → KG Context Exploration → Feedback Generation). Experimental results demonstrated the diagnosis model effectively controlled LLM conceptual errors, providing accurate feedback. The assessment model identified and quantified conceptual errors in student responses, confirming its potential to evaluate understanding beyond simple keyword matching.
Abstract
This study explores the Graph Retrieval-Augmented Generation (Graph RAG) methodology to address challenges in grammar education—diagnosing learner misconceptions and assessing short-answer responses—by controlling LLM hallucinations. A grammar Knowledge Graph (KG) with 56 concepts was constructed, serving as a base for two prototypes: a ‘Grammar Misconception Diagnosis Model’ and an ‘Short-Answer Assessment Support Model.’ Both utilize a 3-stage reasoning architecture(Key Term Extraction → KG Context Exploration → Feedback Generation). Experimental results demonstrated the diagnosis model effectively controlled LLM conceptual errors, providing accurate feedback. The assessment model identified and quantified conceptual errors in student responses, confirming its potential to evaluate understanding beyond simple keyword matching.