电网技术
電網技術
전망기술
Power System Technology
2015年
11期
3163-3168
,共6页
刘建%赵加奎%杨维%顾国栋%李娟%何维民%方红旺%欧阳红
劉建%趙加奎%楊維%顧國棟%李娟%何維民%方紅旺%歐暘紅
류건%조가규%양유%고국동%리연%하유민%방홍왕%구양홍
客服%质检%C4.5 算法%Adaboost 算法%成本矩阵%抽样算法
客服%質檢%C4.5 算法%Adaboost 算法%成本矩陣%抽樣算法
객복%질검%C4.5 산법%Adaboost 산법%성본구진%추양산법
customer service%quality inspection%C4.5 algorithm%Adaboost algorithm%cost matrix%sampling algorithm
为解决目前电力95598客服中心质检部门的质检抽样策略效率低、漏检率高等问题,提出利用改进的C4.5算法对客服话务和工单质量进行是否存在问题进行分类的质检抽样算法。该算法针对客服分类问题的非平衡性特点,将成本矩阵引入到C4.5算法中,利用Adaboost算法构建组合分类器来提升分类算法的准确性。在质检前利用质检抽样算法对所有客服话务和工单可能为问题话务或问题工单的概率进行预测,抽取有问题概率较大的客服话务和工单进行人工质检。实验结果表明,利用该质检抽样算法,可以将问题话务和工单的分类准确率从C4.5算法的66.67%提高到现在的78.49%,从而提高了被抽取的客服话务和工单中问题话务和问题工单的占比,证明该质检抽样算法是可行且有效的,能够提高质检效率、降低漏检率。
為解決目前電力95598客服中心質檢部門的質檢抽樣策略效率低、漏檢率高等問題,提齣利用改進的C4.5算法對客服話務和工單質量進行是否存在問題進行分類的質檢抽樣算法。該算法針對客服分類問題的非平衡性特點,將成本矩陣引入到C4.5算法中,利用Adaboost算法構建組閤分類器來提升分類算法的準確性。在質檢前利用質檢抽樣算法對所有客服話務和工單可能為問題話務或問題工單的概率進行預測,抽取有問題概率較大的客服話務和工單進行人工質檢。實驗結果錶明,利用該質檢抽樣算法,可以將問題話務和工單的分類準確率從C4.5算法的66.67%提高到現在的78.49%,從而提高瞭被抽取的客服話務和工單中問題話務和問題工單的佔比,證明該質檢抽樣算法是可行且有效的,能夠提高質檢效率、降低漏檢率。
위해결목전전력95598객복중심질검부문적질검추양책략효솔저、루검솔고등문제,제출이용개진적C4.5산법대객복화무화공단질량진행시부존재문제진행분류적질검추양산법。해산법침대객복분류문제적비평형성특점,장성본구진인입도C4.5산법중,이용Adaboost산법구건조합분류기래제승분류산법적준학성。재질검전이용질검추양산법대소유객복화무화공단가능위문제화무혹문제공단적개솔진행예측,추취유문제개솔교대적객복화무화공단진행인공질검。실험결과표명,이용해질검추양산법,가이장문제화무화공단적분류준학솔종C4.5산법적66.67%제고도현재적78.49%,종이제고료피추취적객복화무화공단중문제화무화문제공단적점비,증명해질검추양산법시가행차유효적,능구제고질검효솔、강저루검솔。
In allusion to defects such as low efficiency and high omission rate of sampling strategy in quality inspection department of electric 95598 Customer Service Center, a quality inspection sampling algorithm using modified C4.5 algorithm is proposed to classify service calls and work sheets with or without defects. Considering non-equilibrium character of service classification problem, the proposed algorithm introducing cost matrix to C4.5 algorithm and Adaboost algorithm is used to construct combined classifier to improve accuracy of classification algorithm. The proposed algorithm is used to predict incorrectness probability for each service call and work sheet, and only service calls and work sheets of high incorrectness probability are manually inspected. Experiments results show that classified accuracy of incorrect calls and work sheets is increased from 66.67% with C4.5 algorithm to 78.49% with quality inspection sampling algorithm, thus increasing ratio of incorrect service calls and work sheets in all samples. The proposed quality inspection sampling algorithm is feasible and valid, improving efficiency of quality inspection and reducing omission rate.