一种k-means和som结合算法在电信客户细分中的应用-application of k - means and som combined algorithm in telecom customer segmentation.docxVIP

摘要3G时代的到来使得中国电信行业的竞争格局发生了很大的变化，在新的经营环境下，运营商该如何调整各自的经营战略，从而让自己的市场占有率得到有效的提高。此时，能否合理的实施客户细分并准确定位客户的市场需求就成为了关键。数据挖掘技术的研究和发展为电信运营商的客户细分提供了有效的方法。K-means算法是数据挖掘中发展较早，应用较为成熟的聚类算法之一，其可以用来发现数据集的分布模式，是数据挖掘领域中一种有效的聚类算法。K-means算法也有明显的不足之处，如在聚类之前必须给定最优K值，初始聚类中心的随机确定使得聚类的结果往往具有较大的不确定性等。本文提出了一种基于距离代价函数的S-K算法，即利用距离代价函数事先确定K-means算法最优K值，弥补了K值必需事先给定的不足；运用SOM神经网络得到初始聚类中心，从而弥补了K-means算法因初始聚类中心随机确定而导致的聚类结果具有很大不确定性的不足，并运用了电信RFM模型来对电信客户群进行更加有效的细分。通过对Iris标准数据集的测试，证明了该算法的有效性。文章的重点是将基于距离代价函数的S-K算法运用在国内某通信运营商的客户细分研究上，得到了较为满意的聚类结果，并以此为运营商提出了具体的具有针对性的营销策略和建议。关键词：K-meansSOM客户细分RFM数据挖掘IABSTRACTWiththeadventoftheeraof3G,thecompetitivesituationofthetelecommunicationindustryhasdrasticallychangedinChina.Howshouldtheoperatorsmakeuseofthisgoldenopportunitytomaketheirmarketshareincreaseeffectivelyandefficiently?Uptonow,thequestionwhetherthetelecomcompanycanconductrationalcustomersegmentationandaccuratelytargetcustomerdemandhasbecomeverycritical.Fortunately,researchanddevelopmentofDataminingtechnologycanbeadoptedasaneffectiveapproachtotheaforementionedquestion.K-meansalgorithmisoneoftheearlierclusteringalgorithmsandhasrelativelymorematureapplicationsindatamining;itcanbeusedtoidentifythedistributionpatternofthedatasetandthereforehasbeenrecognizedasaneffectiveclusteringalgorithmindatamining.OfcourseitshouldbeacknowledgedthatK-meansalgorithmhasdrawbacks,forexamples,theoptimalKvalueshouldbegivenbeforeclustering,therandomdeterminationoftheinitialclustercentermakestheresultsuncertain,etc.ThispaperproposedaS-Kalgorithmbasedonthedistancecostfunction,i.e.,usingadistancecostfunctionintheK-meansalgorithmfortheoptimalKvaluebeforeconductingtheclustering;theinitialclustercenterisobtainedusingSOMnetworksratherthanbeingrandomly.Inthemeanwhile,thetelecomRFMmodelisemployedtosegmenttelecomcustomersmuchmoreeffectively.Insequel,thevalidityoftheproposedalgorithmisconfirmedthroughthetestoftheIrisstandarddataset.ThehighlightofthearticleistheapplicationofS-Kalgorithm,whichisbasedonthedistancecostfunction,tosometelecomcompanytoim