의도 요소를 고려한 프롬프트 기반 고품질 AI 면접 데이터 선별 시스템

초록

The training performance of a Large Language Models(LLM) significantly depend on the quality of the dataset used. As the quality of data within a specific domain directly impacts model efficacy, constructing a high-quality dataset is crucial for model improvement. Therefore, this paper proposes a prompt-based data refinement system for selecting high-quality data for analyzing AI interviews. The proposed system aims to enable precise data screening through the analysis of intent elements in interview answers, and to evaluate the validity of the data screening rationale and the degree of improvement in efficiency by adding individual prompts corresponding to each intent element. By presenting a novel data curation system that improves the reliability and accuracy of datasets in the AI interview domain, this research establishes the basis for improving the learning efficiency of analytical models specific to the AI interview domain.

키워드