天津移动关键绩效指标MIC和LASSO分析及规则集成预测
发布时间：2018-09-05 20:28
【摘要】：绩效管理已经从以年终分配为目的的绩效考核发展到以全面提升企业管理水平为目的的绩效管理。2002年以来中国移动公司已将关键绩效指标KPI管理模式的绩效管理置于核心位置：公司高层制定决策少不了KPI，月经营分析更是围绕KPI展开。以天津移动公司为例，其绩效考核虽已有一套较为完整的方法，但是员工得分普遍较低，得分低的一个重要原因就是KPI目标定的过高，因此不能充分发挥它的激励作用。所以本文首次利用各种相关系数及套索法LASSO，制定出KPI的合理目标，使绩效考核真正达到对客户经理的激励作用，从而提高移动公司的经济效益。 客户经理的绩效受三方面因素的影响：外部行业环境，自身能力和随机因素。外部行业环境可以通过公司的经营收入反映，通过计算相关系数之间的相关程度得到。如果相关系数的两个变量都是正态分布，则线性相关系数检验功效高；如果不是正态分布，就需考虑其它的相关系数。分别计算客户经理关键绩效指标完成值同公司运营收入之间的距离相关DCC(DistanceCorrelation Coefcient)、和HHG距离(Heller Heller Gorfine Distance)及最大信息指数MIC(Maximal Information Coefcient)等系数，综合比较这些系数，即可得到外部行业环境对客户经理绩效影响的结果。 员工的自身的个人能力因素可选取年龄、性别、职位、职级、所属团体、最高学历和工作天数7个指标为自变量，以2013年客户经理累计完成的集团信息化收入为因变量，利用LASSO方法找出对因变量影响最大的自变量。性别、职位职级、所属团体、最高学历都为分类变量，不能直接用于LASSO回归，因此，回归之前需要采用虚拟编码。LASSO方法循环迭代的步数可通过交叉证实(CrossValidation)得到最优值。从而得到对客户经理绩效影响最大的个人能力因素。 根据以上分析结果表明，客户经理绩效指标的完成值和公司运营收入关联性很大。因此，为了更好地制定绩效考核目标，就需要提前预测公司收入。目前由于联通、电信的强力竞争，移动公司客户流失严重，整个公司收入也在减少。选取天津滨海分公司签约的200家集团用户为例，将签约时间、2012年是否在网、2012客户规模、2012年集团成员统一付费通信收入、2012年增值业务收入指标作为自变量，，将2013年是否在网作为因变量，利用机器学习法中的先进规则集成(Rule Ensemble)法，进行分类拟合。再计算出各变量的重要性和绘制偏相关图形、交互作用图形等。最后预测出容易流失和需要重点攻关的集团用户，从而保证移动公司业务的收入稳定增长。
[Abstract]:Performance management has developed from performance appraisal aimed at year-end distribution to performance management aimed at improving the level of enterprise management. Since 2002, China Mobile has implemented the performance management of the key performance indicator KPI management model. At the core: the company's top decision making without KPI, monthly business analysis is around the KPI. Taking Tianjin Mobile Corporation as an example, although its performance appraisal has a set of relatively complete method, but the staff score is generally low, one of the important reasons of low score is that the goal of KPI is too high, so it can not give full play to its incentive role. So this paper makes use of all kinds of correlation coefficient and the lasso method LASSO, to formulate the reasonable goal of KPI for the first time, so that the performance appraisal can really achieve the incentive function to the customer manager, thus improving the economic benefit of the mobile company. Account manager performance is affected by three factors: external industry environment, self-competence and random factors. The external industry environment can be reflected by the company's operating income, and the correlation degree between the correlation coefficients can be calculated. If both variables of the correlation coefficient are normal distribution, the efficiency of linear correlation coefficient test is high; if the correlation coefficient is not normal distribution, other correlation coefficients should be considered. The distance correlation DCC (DistanceCorrelation Coefcient), HHG distance (Heller Heller Gorfine Distance) and the maximum information index MIC (Maximal Information Coefcient) between the completion value of the key performance index of the customer manager and the company's operating income were calculated, and these coefficients were compared synthetically. We can get the result of the effect of the external industry environment on the performance of the customer manager. The individual ability factors of employees can be selected as seven independent variables: age, gender, position, rank, affiliated group, highest educational background and working days, and the income of group informatization completed by the customer manager in 2013 is dependent variable. The LASSO method is used to find out the independent variables which have the greatest influence on dependent variables. Gender, rank and rank of position, group and highest education are all classified variables, which can not be directly used in LASSO regression. Therefore, the number of steps that need to be iterated by virtual coding. LASSO method before regression can be cross-verified by (CrossValidation) to obtain the optimal value. Thus, the personal ability factors which have the greatest influence on the customer manager's performance are obtained. According to the above analysis results, the completion value of the customer manager performance index is closely related to the operating income of the company. Therefore, in order to better establish performance appraisal goals, we need to predict the company's income ahead of time. At present, due to the strong competition of Unicom and telecom, mobile company customers are losing a lot, and the whole company's revenue is also decreasing. Taking 200 group users signed by Tianjin Binhai Branch as an example, the time of signing the contract, whether the customer size of 2012 is on the net, the unified payment communication revenue of 2012 group members, and the revenue index of value-added service in 2012 are taken as independent variables. In this paper, we use the advanced rules of machine learning method to integrate (Rule Ensemble) method and classify fit whether or not it is in the network as dependent variable in 2013. Then calculate the importance of each variable and draw partial correlation figure, interaction figure and so on. Finally, the group users who are easy to lose and need key problems are predicted to ensure the steady growth of mobile business revenue.
【学位授予单位】：天津大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：F626
本文编号：2225378
[Abstract]:Performance management has developed from performance appraisal aimed at year-end distribution to performance management aimed at improving the level of enterprise management. Since 2002, China Mobile has implemented the performance management of the key performance indicator KPI management model. At the core: the company's top decision making without KPI, monthly business analysis is around the KPI. Taking Tianjin Mobile Corporation as an example, although its performance appraisal has a set of relatively complete method, but the staff score is generally low, one of the important reasons of low score is that the goal of KPI is too high, so it can not give full play to its incentive role. So this paper makes use of all kinds of correlation coefficient and the lasso method LASSO, to formulate the reasonable goal of KPI for the first time, so that the performance appraisal can really achieve the incentive function to the customer manager, thus improving the economic benefit of the mobile company. Account manager performance is affected by three factors: external industry environment, self-competence and random factors. The external industry environment can be reflected by the company's operating income, and the correlation degree between the correlation coefficients can be calculated. If both variables of the correlation coefficient are normal distribution, the efficiency of linear correlation coefficient test is high; if the correlation coefficient is not normal distribution, other correlation coefficients should be considered. The distance correlation DCC (DistanceCorrelation Coefcient), HHG distance (Heller Heller Gorfine Distance) and the maximum information index MIC (Maximal Information Coefcient) between the completion value of the key performance index of the customer manager and the company's operating income were calculated, and these coefficients were compared synthetically. We can get the result of the effect of the external industry environment on the performance of the customer manager. The individual ability factors of employees can be selected as seven independent variables: age, gender, position, rank, affiliated group, highest educational background and working days, and the income of group informatization completed by the customer manager in 2013 is dependent variable. The LASSO method is used to find out the independent variables which have the greatest influence on dependent variables. Gender, rank and rank of position, group and highest education are all classified variables, which can not be directly used in LASSO regression. Therefore, the number of steps that need to be iterated by virtual coding. LASSO method before regression can be cross-verified by (CrossValidation) to obtain the optimal value. Thus, the personal ability factors which have the greatest influence on the customer manager's performance are obtained. According to the above analysis results, the completion value of the customer manager performance index is closely related to the operating income of the company. Therefore, in order to better establish performance appraisal goals, we need to predict the company's income ahead of time. At present, due to the strong competition of Unicom and telecom, mobile company customers are losing a lot, and the whole company's revenue is also decreasing. Taking 200 group users signed by Tianjin Binhai Branch as an example, the time of signing the contract, whether the customer size of 2012 is on the net, the unified payment communication revenue of 2012 group members, and the revenue index of value-added service in 2012 are taken as independent variables. In this paper, we use the advanced rules of machine learning method to integrate (Rule Ensemble) method and classify fit whether or not it is in the network as dependent variable in 2013. Then calculate the importance of each variable and draw partial correlation figure, interaction figure and so on. Finally, the group users who are easy to lose and need key problems are predicted to ensure the steady growth of mobile business revenue.
【学位授予单位】：天津大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：F626
【参考文献】
相关博士学位论文 前1条
1 倪强;基于随机前沿和随机森林法的沿海开发区发展效率研究[D];天津大学;2011年
相关硕士学位论文 前5条
1 王彦霞;支持向量机在阵列电磁传播测井资料处理中的应用[D];燕山大学;2011年
2 蒋杭进;最大信息系数及其在脑网络分析中的应用[D];中国科学院研究生院（武汉物理与数学研究所）;2013年
3 周忠;铁通CD分公司绩效管理体系再设计[D];电子科技大学;2013年
4 李立桂;L市移动公司客户经理绩效指标研究[D];南京理工大学;2013年
5 战泉茹;基于最大信息系数的人脸特征选择[D];东北师范大学;2013年
本文编号：2225378
本文链接：https://www.wllwen.com/jingjilunwen/xxjj/2225378.html
