Methods to Enhance Data Labeling Efficiency Using Large Language Models

Abstract

The performance of artificial intelligence models heavily relies on extensive data labeling, which currently depends largely on manual effort, demanding significant time and cost. This study presents an automated data labeling method using Large Language Models (LLMs) to address this challenge. We developed an innovative approach that first summarizes data using text summarization libraries, followed by labeling through LLMs. This method considers the token-based cost structure of LLMs, maximizing data processing efficiency while optimizing costs. Experiments demonstrate that automated labeling using LLMs significantly reduces labeling time and overall costs compared to traditional manual methods. These results suggest that LLM-based data labeling can greatly enhance the efficiency of AI model development. Moreover, this research implies that LLM technology is not limited to language processing but can play a vital role in new areas of data processing and management. Such findings hold significant implications for future AI research and development, expanding the possibilities of LLM applications across various fields.