Scholarlink

Automatic Labeling is a feature within Azure Machine Learning Data Labeling that allows users to use large language models like GPT-3.5 Turbo to label datapoints automatically. In this paper we cover the high-level architecture of this service. We also present test results assessing the accuracy of this Automatic Labeling against several publicly available text and image datasets from Hugging Face, representing problems of varying difficulty, and comparing the labels generated by Automatic Labeling to the ground truth labels included with the datasets. We used GPT-3.5 Turbo to label the text datasets and GPT-4V to label the image datasets. Our test results demonstrate effectiveness of the feature with promising accuracies ranging between 48% to 94% with a simple generic prompt.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automatic Data Labeling Using Large Language Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Adarsh Ramanathan,
  • Jason Liang,
  • Dan Sommerfield,
  • Hossein Azari

摘要

Automatic Labeling is a feature within Azure Machine Learning Data Labeling that allows users to use large language models like GPT-3.5 Turbo to label datapoints automatically. In this paper we cover the high-level architecture of this service. We also present test results assessing the accuracy of this Automatic Labeling against several publicly available text and image datasets from Hugging Face, representing problems of varying difficulty, and comparing the labels generated by Automatic Labeling to the ground truth labels included with the datasets. We used GPT-3.5 Turbo to label the text datasets and GPT-4V to label the image datasets. Our test results demonstrate effectiveness of the feature with promising accuracies ranging between 48% to 94% with a simple generic prompt.