基于大模型的轻量级智能出版知识服务:理论基础与实现路径
Lightweight Intelligent Publishing Knowledge Services Based on Large Models:Theoretical Foundations and Implementation Paths
许洁 1袁小群 1朱瑞 2孟繁永3
作者信息
- 1. 语义出版与知识服务实验室,北京,100005;武汉大学出版研究院,武汉,430064
- 2. 武汉大学出版研究院,武汉,430064
- 3. 语义出版与知识服务实验室,北京,100005
- 折叠
摘要
以ChatGPT为代表的大型预训练模型(简称大模型)广泛应用于信息抽取、自动摘要、问答、纠错、续写等,为出版行业带来新机遇.然而,由于大模型训练门槛高,出版行业利用大模型存在困难.武汉大学牵头的语义出版与知识服务实验室研发了基于大模型的轻量级智能出版知识服务平台,为出版业低成本、高效率地利用大模型开展知识服务提供了解决方案.该平台采用"大模型+知识检索"和"预训练+微调"两条路径来运用大模型开展智能出版知识服务.实现了真正意义上的低代码、轻量化运行,减少了出版单位的负担,为降本增效、高质量发展提供有效支撑.
Abstract
Large pre-trained models represented by ChatGPT have found extensive applications in information extraction,automatic summarization,question-answer,error correction,and content generation etc.,bringing new opportunities for the publishing industry.However,the high training threshold of large models has posed challenges for their adoption in the publishing industry.The Semantic Publishing and Knowledge Service Laboratory led by Wuhan University has developed a lightweight intelligent publishing knowledge service platform based on large models,providing a solution for the publishing industry to use large models in knowledge services at low cost and high efficiency.The model adopts two approaches,"large model+knowledge retrieval"and"pre-training+fine-tuning"to apply large models in intelligent publishing knowledge services.It realizes true low-code and lightweight operation,reducing the burden on publishing units,and providing effective support for cost reduction and efficiency enhancement,and high-quality development.
关键词
大模型/预训练/智能出版/知识服务/出版大模型Key words
Large model/Pre-training/Intelligent publishing/Knowledge services引用本文复制引用
出版年
2024
NETL
NSTL
万方数据