基于SVM的迁移学习方法在互联网金融网页分类中的应用  

Application of SVM-Based Transfer Learning Method in Classification of Internet Financial Web

作　　者：谢林燕 刘纪伟 张玉 张峰 XIE Linyan;LIU Jiwei;ZHANG Yu;ZHANG Feng(National Computer Network Emergency Technology Coordination Center Hebei Branch,Shijiazhuang Hebei 050021,China)

机构地区：[1]国家计算机网络应急技术处理协调中心河北分中心,河北石家庄050021

出　　处：《通信技术》2023年第7期909-914,共6页Communications Technology

基　　金：河北省省级科技计划(20310701D);中央引导地方科技发展资金项目(216Z0701G);国家计算机网络应急技术处理协调中心青年科研基金项目(2020Q38)。

摘　　要：经过人工标注的互联网金融网页数据量偏小,且新兴金融业态繁多,给互联网金融网页的业态识别带来了困难。对此提出一种基于超平面位置调优支持向量机(Support Vector Machine,SVM)的迁移学习方法。首先利用迁移学习,将相关业态的已标注数据以及少量新业态的已标注数据迁移到新的金融业态中,然后通过对SVM算法模型进行调整,引入参数对分类超平面位置调优,实现分类准确率的提升。实验结果表明了该方法性能的有效性,采用该方法比现有方法准确率平均提高约0.8%,最高提高约1.7%。The data volume of manually labeled Internet financial web pages is small and there are many emerging financial industries,which brings difficulties to the industry identification of Internet financial web pages.In this regard,this paper proposes a transfer learning method based on hyperplane location tuning SVM(Support Vector Machine).First,the labeled data of related industries and a small amount of labeled data of new industries are transferred to the new financial industries through transfer learning.Then,the classification accuracy is improved by adjusting the SVM algorithm model and introducing parameters for hyperplane position tuning.The experimental results indicate the effectiveness of the performance of the method,the accuracy of the method is improved by about 0.8%on average and up to about 1.7%over the existing methods.

关 键 词：互联网金融 支持向量机 迁移学习 数据增强 业态识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]