面向图书自动分类的大语言模型提示学习研究

A Research on Prompt Learning of Large Language Models for Automated Book Classification

为避免高昂的人力成本,从少量样本中学习图书类目的特征与规律已成为图书馆学的热门问题。以图书为研究对象,利用大语言模型ChatGPT表征文本,构建大语言模型提示学习模型,以实现自动识别并分类图书的目的。针对当前方法需要大规模数据样本与大量训练时间的缺陷,构建“询问大语言模型-提示-生成”图书分类范式。在广州图书馆和郑州图书馆10个一级类目共114 823条图书数据集上进行实验验证。实验结果显示,此范式在精准率、召回率与F1等指标上获得最优分类结果。

In order to avoid high costs of manpower,learning the features and patterns of book categories from a small number of sampleshas become a brand-new problem in library science.This paper takes the books as our research target,uses the large language model(ChatGPT)to represent the texts and images,builds a large language model based prompt learning model for recognizing and classifying books.In view of the defects of the current method requiring large-scale data samples and training time,this paper proposes a new paradigm of“inquiry large language model-prompt-generation”.The model was verified experimentally on the dataset of 114823 books from Guangzhou Library and Zhengzhou Library,and the best results were obtained in terms of precision,recall and F1.