Scholarlink

<p>In the field of steel e-commerce, quality objection texts typically reflect consumers’ negative feedback regarding the dimensions, specifications, appearance, and performance of steel products. By quickly classifying these quality objection texts, the customer service team of a steel e-commerce platform can swiftly identify and categorize specific consumer complaints, thereby accelerating the resolution process. However, traditional text classification methods require a large amount of labeled data, and their performance is suboptimal in data-scarce scenarios. Therefore, this paper aims to effectively address this issue by combining prompt engineering and large language model (LLM). Specifically, we first integrate prompt engineering and the LLM to propose various data augmentation strategies, such as text generation and synonym replacement. Then, we design a prompt engineering framework for supervised fine-tuning (SFT) based on the LLM, which includes four key elements: role, task, requirements, and supplementary information. Additionally, we incorporate one-shot and few-shot techniques to effectively enhance the performance of SFT. Finally, extensive ablation experiments demonstrate the superiority and practical feasibility of our method.</p>

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

Text classification methods in the steel e-commerce industry using the large language model and prompt engineering

  • 期刊论文

  • 发表日期：

    2025年8月8日

查看全文
  • Qiaojuan Peng,
  • Xiong Luo,
  • Yuqi Yuan,
  • Fengbo Gu,
  • Hailun Shen,
  • Ziyang Huang

摘要

In the field of steel e-commerce, quality objection texts typically reflect consumers’ negative feedback regarding the dimensions, specifications, appearance, and performance of steel products. By quickly classifying these quality objection texts, the customer service team of a steel e-commerce platform can swiftly identify and categorize specific consumer complaints, thereby accelerating the resolution process. However, traditional text classification methods require a large amount of labeled data, and their performance is suboptimal in data-scarce scenarios. Therefore, this paper aims to effectively address this issue by combining prompt engineering and large language model (LLM). Specifically, we first integrate prompt engineering and the LLM to propose various data augmentation strategies, such as text generation and synonym replacement. Then, we design a prompt engineering framework for supervised fine-tuning (SFT) based on the LLM, which includes four key elements: role, task, requirements, and supplementary information. Additionally, we incorporate one-shot and few-shot techniques to effectively enhance the performance of SFT. Finally, extensive ablation experiments demonstrate the superiority and practical feasibility of our method.