層次法聚類算法在中國(guó)石油湖北公司客戶細(xì)分中的應(yīng)用

全文85頁(yè) 約30000字 論述翔實(shí)

摘 要

中國(guó)加入WTO后，承諾逐步開放成品油的批發(fā)和零售業(yè)務(wù)。目前，大量的國(guó)際巨頭石油財(cái)團(tuán)正紛至踏入中國(guó)市場(chǎng)，加劇了該市場(chǎng)的競(jìng)爭(zhēng)。對(duì)此，中國(guó)石油湖北公司為了能在競(jìng)爭(zhēng)激烈的成品油市場(chǎng)中站穩(wěn)腳跟，應(yīng)重新考慮新的競(jìng)爭(zhēng)戰(zhàn)略，識(shí)別和提升自身的核心競(jìng)爭(zhēng)能力?，F(xiàn)下，業(yè)內(nèi)普遍認(rèn)為有效的客戶關(guān)系管理是企業(yè)提升自身核心競(jìng)爭(zhēng)能力的關(guān)鍵因素之一，而準(zhǔn)確的客戶細(xì)分是客戶關(guān)系管理的前提。因此，準(zhǔn)確地細(xì)分客戶對(duì)公司成功實(shí)施客戶關(guān)系管理至關(guān)重要。
目前，中國(guó)石油湖北公司的業(yè)務(wù)管理系統(tǒng)存在以下兩大問題：1）客戶分類太過主觀，不科學(xué)；2）客戶數(shù)據(jù)繁多，但信息量不高。因此，如何準(zhǔn)確地細(xì)分客戶、辨別出公司中有價(jià)值的客戶是中國(guó)石油湖北公司亟待解決的問題。
本文的目的是對(duì)中國(guó)石油湖北公司現(xiàn)有的客戶進(jìn)行細(xì)分，識(shí)別出公司中最有價(jià)值的客戶。首先，本文研究了各種常用的層次法(hierarchical method)聚類算法，總結(jié)了各個(gè)算法的利弊，對(duì)各個(gè)算法的多方面性能作了一個(gè)總體的、較全面的比較；接著，整合了BIRCH算法與CHAMELEON算法的核心思想，以此作為客戶細(xì)分的數(shù)據(jù)挖掘技術(shù)，并用JAVA語言實(shí)現(xiàn)了該算法整合；然后，從基于購(gòu)買行為的角度，比較、分析了RFM法(Recency Frequency Monetary value)和MARCUS提出的客戶價(jià)值矩陣(Customer Value Matrix)的利弊，選用客戶價(jià)值矩陣作為客戶細(xì)分的分析框架；隨后，對(duì)中國(guó)石油湖北公司現(xiàn)有的客戶進(jìn)行聚類，并以該公司的客戶數(shù)據(jù)為實(shí)驗(yàn)對(duì)象，比較、分析了算法的整合較BIRCH算法與CHAMELEON算法的優(yōu)越性；最后，運(yùn)用客戶價(jià)值矩陣解釋、分析了算法整合的聚類結(jié)果，并針對(duì)最有價(jià)值的客戶群提出了客戶保持策略。

【關(guān) 鍵 字】：數(shù)據(jù)挖掘,聚類,客戶關(guān)系管理,客戶價(jià)值矩陣,客戶細(xì)分

ABSTRACT
After entry into the WTO, China has ensured to open the wholesale and retail businesses in oil market widely. Nowadays, the competition in oil market has become more and more fierce , for there are a large number of consortiums entering China. Therefore, the branch of Chinese Petroleum Corp. in Hubei Province should adopt new competition strategies, discern and improve its core competence in order to survive in such a competitive oil market. Currently, effective Customer Relationship Management(CRM) is deemed to be a key factor in improving core competence, meanwhile accurate customer classification is the prerequisite of CRM.
Currently, there mainly exist two weak points in the Business Management System of the branch of Chinese Petroleum Corp. in HUBEI Province:1)Customer Classification is too subjective and unscientific;2)The Database has massive customer datum with low informative. Thus, how to classify customers accurately and identify the highly valuable customers of the Corporation are the problems which should be solved without delay.
The main purpose of this thesis is to classify the existing customers and identify the highly valuable customers in the branch of Chinese Petroleum Corp. in Hubei Province. Firstly, this thesis analyzes all kinds of commonly used Clustering Algorithms of hierarchical methods, summarizes the advantages and disadvantages of these algorithms, and compares the performances among these algorithms in general. Secondly, this thesis integrates the core principles of BIRCH algorithm and CHAMELEON algorithm, which are used as the data mining technology, and realizes the integrated-algorithm in JAVA. Thirdly, this thesis compares the RFM(Recency Frequency Monetary Value) with Customer Value Matrix proposed by MARCUS, based on purchasing-behavior, and uses Customer Value Matrix as the framework of customer classification. Fourthly, this thesis clusters the existing customers in the branch of Chinese Petroleum Corp. in HUBEI Province, and compares the integrated-algorithm with the BIRCH algorithm and CHAMELEON algorithm in performance respectively, based on the customer datum. Finally, this thesis explains and analyzes the clustering result of integrated-algorithm with Customer Value Matrix, and proposes some customer-retaining tactics according to the highly valuable customers.
【Key words】：Data Mining, Clustering, Customer Relationship Management, Customer Value Matrix, Customer Segmentation
【Type of Thesis】：Applied Research
















目錄
第一章 緒論 10
1.1 研究背景 10
1.2 問題描述 10
1.3 研究思路和方法 11
1.4 論文的主要內(nèi)容 12
第二章 數(shù)據(jù)挖掘與聚類的理論綜述 13
2.1數(shù)據(jù)挖掘 13
2.1.1數(shù)據(jù)挖掘的概念 13
2.1.2數(shù)據(jù)挖掘的分類 13
2.1.3數(shù)據(jù)挖掘的基本過程 14
2.1.4數(shù)據(jù)挖掘的目的 15
2.2聚類分析 16
2.2.1聚類分析的概念 16
2.2.2聚類方法的分類 16
2.2.3聚類技術(shù)在客戶關(guān)系管理中的應(yīng)用 18
2.3 聚類分析與數(shù)據(jù)挖掘的關(guān)系 19
第三章 層次法聚類算法研究 20
3.1層次法聚類算法的概念 20
3.2 六種常用的層次法聚類算法研究 21
3.2.1單鏈接法 21
3.2.2全鏈接法 23
3.2.3 CURE算法 24
3.2.4 ROCK算法 27
3.2.5 四種層次法聚類算法的共同缺陷 30
3.2.6 CHAMELEON算法 33
3.2.7 BIRCH算法 38
3.3 層次法聚類算法小結(jié) 42
3.4算法整合 42
3.4.1算法整合的描述 42
3.4.2算法的程序設(shè)計(jì) 43
3.4.3算法整合后的時(shí)間復(fù)雜度分析 44
第四章 客戶細(xì)分 46
4.1 客戶關(guān)系管理的概念 46
4.2 客戶細(xì)分與客戶保持的概念 46
4.3 現(xiàn)有客戶的重要性 46
4.4客戶細(xì)分框架的分析 47
4.5 客戶細(xì)分框架的選取 50
第五章 BIRCH算法與CHAMELEON算法的整合及其在客戶細(xì)分中的應(yīng)用 52
5.1 聚類目的 52
5.2數(shù)據(jù)預(yù)處理 52
5.2.1數(shù)據(jù)選擇 52
5.2.2數(shù)據(jù)清理 53
5.2.3數(shù)據(jù)轉(zhuǎn)換 53
5.2.4數(shù)據(jù)歸約 55
5.3算法的整合在客戶細(xì)分中的應(yīng)用 55
5.3.1聚類階段1——BIRCH算法 55
5.3.2聚類階段2——CHAMELEON算法 60
5.3.3聚類結(jié)果評(píng)價(jià) 61
5.3.4聚類結(jié)果檢驗(yàn)與解釋 62
5.3.5孤立點(diǎn)分析 64
5.4 CHAMELEON算法與算法整合的比較 65
5.4.1運(yùn)用CHAMELEON算法對(duì)公司現(xiàn)有的客戶進(jìn)行聚類 65
5.4.2 CHAMELEON算法與算法整合的聚類效果的比較 66
5.4.3 CHAMELEON算法與算法整合計(jì)算效率的比較 67
5.4.4 CHAMELEON算法與算法整合的性能小節(jié) 67
5.4 客戶分類 68
5.4.1客戶類型描述 68
5.4.2 最優(yōu)客戶 69
5.4.3 經(jīng)常性客戶 71
5.4.4 樂于消費(fèi)型客戶 72
5.4.5 不確定性客戶 74
5.5 80/20法則 75
5.6 客戶保持策略 75
5.6.1重點(diǎn)發(fā)展與大客戶的關(guān)系 75
5.6.2鎖定策略 77
第六章 結(jié)論 78
6.1 結(jié)論、工作與貢獻(xiàn) 78
6.2本文的局限性 79
致謝 81
參考文獻(xiàn) 82

部分參考文獻(xiàn)

【31】陳明亮, 袁澤沛, 李懷祖. 客戶保持動(dòng)態(tài)模型的研究. 武漢大學(xué)學(xué)報(bào)(社會(huì)科學(xué)版), Vol154, No16 Nov. 2001:675-684.
【32】劉義,萬迪昉,張鵬. 基于購(gòu)買行為的客戶細(xì)分方法比較研究 .管理科學(xué),2003, 16(1):69-72.
【33】Timothy Bock , Mark Uncles. A taxonomy of differences between consumers for market segmentation, Intern. J. of Research in Marketing , 2002 ,19:215–224.
【34】趙玉臣,在商品營(yíng)銷中如何防范老客戶的流失，平原大學(xué)學(xué)報(bào), 2004,21卷第四期.
【35】范云峰，企業(yè)客戶流失的原因及防范措施，中國(guó)流通經(jīng)濟(jì)，2003，第五期.
【36】Carl Shapiro Hal Varian著，張帆譯，信息規(guī)則——網(wǎng)絡(luò)經(jīng)濟(jì)的策略指導(dǎo)，中國(guó)人民大學(xué)出版社，2000：(91-117).