摘要
文章通过解决深层文献分类中的效果优化和效率提升问题,有效实现中文医学文献的中图法自动分类。首先采用层次分类思想,构建基于BERT模型的两层分类器集群,以实现在深层文献分类中分类效果的优化;其次,基于微服务的方式对处理流程进行优化以实现分类速度的提升。在此基础上,设计单层分类器和两层分类器集群分类效果与分类速度的对照实验,对比分析两种模型的分类效果以及分类速度差异。实验结果表明:两层分类器集群较比单层分类器取得了更好的分类效果,其F1值得到4.39%的提升;基于微服务的方式可以有效提升分类器的预测速度。为实现快速、准确的文本自动分类提供了一种解决方案。
By solving the problem of effect optimization and efficiency improvement in deep literature classification, the article effectively realizes the automatic classification of Chinese medical literature by Chinese Library Classification.First, we propose a two-layer classifier cluster based on BERT to optimize the classification effect in the deep-level literature classification, the idea is derived from the hierarchical classification.Secondly, we optimize the processing flow based on microservice to achieve the improvement of the classification speed.On this basis, we design comparative experiments of effect and efficiency on single-layer classifier and two-layer classifier cluster, then analyze the difference in effect and speed.Experimental results show that the two-layer classifier cluster which we constructed has achieved better classification results than the single-layer classifier, and its F1-score has been improved by 4.39%.The microservice-based approach can effectively improve the prediction speed of the classifier.A solution is provided for realizing fast and accurate automatic text classification tasks in the future.
作者
赵旸
张智雄
刘欢
Zhao Yang;Zhang Zhixiong;Liu Huan
出处
《图书馆学研究》
CSSCI
北大核心
2021年第21期49-55,61,共8页
Research on Library Science
基金
中国科学院文献情报能力建设专项子项目“基于科技文献知识的人工智能(AI)引擎建设”(项目编号:E0290906)的研究成果之一
关键词
两层分类器集群
层次分类法
BERT
文献分类
微服务架构
two-layer classifier cluster
hierarchical classification
BERT
literature classification
microservice architecture
