基于电力工单文本的WTF-IDF特征选择优化  

Feature selection optimization of WTF⁃IDF based on power text

作　　者：朱君 杨萌 杜振东 章劲秋 嵇友浪 刘国婷 ZHU Jun;YANG Meng;DU Zhendong;ZHANG Jinqiu;JI Youlang;LIU Guoting(State Grid Jiangsu Marketing Service Center,Nanjing 210000,China;Nanjing Yunwen Network Technology Co.,Ltd.,Nanjing 210000,China)

机构地区：[1]国网江苏省电力有限公司营销服务中心,江苏南京210000 [2]南京云问网络技术有限公司,江苏南京210000

出　　处：《电子设计工程》2021年第5期159-163,共5页Electronic Design Engineering

基　　金：国网江苏省电力有限公司科技项目(J2018020)。

摘　　要：文中通过对95598客服服务中心的电力工单历史数据中反映的热点问题进行建模,实现对海量工单文本内容快速、准确地分类,为管理人员发现与解决问题提供支撑。针对电力领域的特殊性,引入了通过新词识别手段构建的针对电力文本的领域词典。通过对传统的TF-IDF算法加以改进,提出了加入领域词典的WTF-IDF特征选择优化算法,旨在给予工单问题中重要的词更多的注意力,有助于机器更好地理解句子中的重点。实验表明,所提方法在电力文本分类方面比传统的TF-IDF取得了更高的准确率与更快的模型训练速度。This paper models the hot issues reflected in the data of electricity history work orders to improve the speed and accuracy of 95598 customer service on customer judgment.Aiming at the particularity of the power field,a domain dictionary for power text is constructed by means of new word recognition.Through the improvement of the traditional TF⁃IDF algorithm,the WTF⁃IDF feature selection optimization algorithm is proposed to add the important words in the work order problem,which helps the machine better understand the sentence in the focus.Experiments show that the proposed method achieves higher accuracy and faster model training than traditional TF⁃IDF in terms of power text classification.

关 键 词：领域词典 新词识别 95598电力工单 WTF-IDF 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]