基于层次分类器的专利文本分类模型研究  

Research on Patent Text Classification Model Based on Hierarchical Classifier

作　　者：慎金花[1] 陈红艺 张更平[1] 秦乐洋 Shen Jinhua;Chen Hongyi;Zhang Gengping;Qin Leyang(Tongji University Library,Shanghai 200092;School of Economics and Management,Tongji University,Shanghai 200092;Alibaba,Hangzhou 311121)

机构地区：[1]同济大学图书馆,上海200092 [2]同济大学经济与管理学院,上海200092 [3]阿里巴巴,杭州311121

出　　处：《情报杂志》2023年第8期157-163,68,共8页Journal of Intelligence

摘　　要：数据为实验对象,通过文本预处理及文本特征表示后,基于KNN、支持向量机、Rocchio和朴素贝叶斯4种机器学习模型,分别探索IPC部、大类、小类和大组层次上的最佳分类模型及其组合。[研究结论]实验结果显示,层次结构可有效改善平面分类模型的性能,层次组合模型比层次单一模型拥有更高的分类准确率,各层次的最优分类模型分别是:支持向量机(部)、Rocchio+支持向量机(大类)、Rocchio+朴素贝叶斯+支持向量机(小类)、KNN+朴素贝叶斯+支持向量机+支持向量机(大组)。data as the research object,after text preprocessing and text feature representation,this study is based on KNN,support vector machine,Rocchio and Naive Bayes 4 machine learning models to explore the classification model with the highest accuracy at each layer.[Research conclusion]The experimental results show that the hierarchical structure can effectively improve the performance of the flat classification model.The hierarchical combination model has a higher classification accuracy than the hierarchical single model.The optimal classification models for each layer are:support vector machine(section),Rocchio+Support Vector Machine(main class),Rocchio+Naive Bayes+Support Vector Machine(subclass),KNN+Naive Bayes+Support Vector Machine+Support Vector Machine(main group).

关 键 词：专利分类 文本分类 层次分类 国际专利分类 发明专利 

分 类 号：G306[文化科学] TP391.1[自动化与计算机技术—计算机应用技术]