电子商务模式下的顾客行为特征提取及利润挖掘

发布时间：2018-05-24 19:01

本文选题：智能电子商务 + 数据挖掘　；参考：《天津大学》2010年博士论文

【摘要】：随着Internet技术的不断发展,电子商务系统给商家和客户带来了越来越多的信息,于是各种基于电子商务的个性化服务应运而生,个性化服务成为一个研究的热点,引起人们的广泛关注。从客户角度出发,客户更关心顺利找到自己需要的商品。电子商务系统可以模拟商店销售人员向用户提供商品推荐,帮助用户找到所需商品,从而顺利完成购买过程。从企业角度出发,发现高价值的商品组合,帮助企业优化客户,为企业创造更多的利润,是企业实施电子商务系统的最终目的。本文从这两个角度出发,对个性化服务的电子商务模型相关问题进行了深入研究。在文章开头介绍了数据挖掘的主要方法和研究热点。评价了相关问题的研究进展,简要介绍了进化计算理论基础、整体框架以及最新研究进展。以下是本文主要研究内容和创新性工作,主要包括: (1)利用系统分析理论和价值链理论,提出了基于个性化服务的智能化电子商务模式,并分析这种商务模式框架特点和优势。然后,从企业战略角度分析了该模式在电子商务市场环境下竞争优势,并与传统模式进行了比较分析。最后,结合典型案例实证分析了这种新型的电子商务模式的现实意义,针对这种电子商务模式特点制定一套竞争战略,并对战略规划和实施进行详细论述。 (2)提出了一种基于遗传算法的顾客购买行为特征提取算法。该算法分为两个阶段,第一阶段,采用Tanimoto相似度来度量顾客间购买行为,并设计遗传聚类算法对顾客群体进行划分,把具有相似购买行为顾客聚集为一类。然后,针对不同顾客群体的购买行为特征,设计一种基于遗传算法的多种群特征提取方法,从各个子群体中发现顾客的购买行为的知识。为了增强种群内部协同进化能力和规则质量,我们采用最近邻替代遗传策略(q-NNR)和局部搜索策略。我们使用实际零售数据集对整个算法进行了验证,并与经典的Apriori算法进行比较,实验结果表明该算法在不需要产生频繁项集的情况下,可以比较高效生成精简规则集,在规则形式方面也更加灵活。最后,我们对实验结果进行了详细的分析。 (3)利用关联分析模型,建立一个多目标优化模型。该模型把商品直接收益和由于交叉销售因素产生的间接利润作为两个独立的优化目标,并设计多目标遗传算法进行求解。为了增加种群多样性和提高算法搜索能力,加入个体修补、填充策略和局部搜索策略。最后,用实际零售数据集对该多目标优化模型和多目标遗传算法进行了验证。通过实验分析表明,这种多目标优化算法可以获得丰富信息,为决策者制定具有针对性营销策略提供比较全面的信息。
[Abstract]:With the development of Internet technology, e-commerce system brings more and more information to merchants and customers. Draw people's wide attention. From the point of view of customers, customers are more concerned about finding the goods they need. E-commerce system can simulate the store salesperson to provide the product recommendation to the user, help the user to find the needed goods, and thus complete the purchase process smoothly. From the point of view of enterprise, it is the ultimate goal of the enterprise to realize the electronic commerce system to find the high value commodity combination, to help the enterprise optimize the customer, and to create more profit for the enterprise. From these two angles, this paper makes a deep research on the e-commerce model of personalized service. At the beginning of the article, the main methods and research focus of data mining are introduced. The research progress of the related problems is evaluated, and the theoretical basis, the global framework and the latest research progress of evolutionary computing are briefly introduced. The following are the main contents and innovative work of this paper, including: 1) based on the system analysis theory and value chain theory, the intelligent e-commerce model based on personalized service is proposed, and the characteristics and advantages of this business model framework are analyzed. Then, this paper analyzes the competitive advantage of this model in the electronic commerce market environment from the angle of enterprise strategy, and compares it with the traditional model. Finally, this paper analyzes the practical significance of this new mode of electronic commerce based on typical cases, formulates a set of competitive strategy according to the characteristics of this mode of electronic commerce, and discusses in detail the planning and implementation of the strategy. A genetic algorithm based on genetic algorithm (GA) is proposed to extract the feature of customer purchase behavior. The algorithm is divided into two stages. In the first stage, Tanimoto similarity is used to measure the purchase behavior between customers, and genetic clustering algorithm is designed to divide the customer population into a class of customers with similar purchase behavior. Then, a multi-population feature extraction method based on genetic algorithm is designed to find the knowledge of the customer's purchase behavior from each sub-population according to the purchase behavior characteristics of different customer groups. In order to enhance the ability of coevolution and the quality of rules within the population, we adopt the nearest neighbor alternative genetic strategy (Q-NNR) and the local search strategy. We use the real retail data set to verify the algorithm and compare it with the classical Apriori algorithm. The experimental results show that the algorithm can efficiently generate the reduced rule set without generating frequent itemsets. There is also greater flexibility in the form of rules. Finally, we analyze the experimental results in detail. A multi-objective optimization model is established by using the correlation analysis model. The model takes the direct profit of commodities and the indirect profit caused by cross-selling as two independent optimization objectives and designs a multi-objective genetic algorithm to solve the problem. In order to increase population diversity and improve the search ability of the algorithm, individual patching, filling strategy and local search strategy are added. Finally, the multi-objective optimization model and multi-objective genetic algorithm are verified with real retail data sets. The experimental results show that the multi-objective optimization algorithm can obtain abundant information and provide more comprehensive information for decision-makers to formulate targeted marketing strategies.
【学位授予单位】：天津大学
【学位级别】：博士
【学位授予年份】：2010
【分类号】：F274;F713.36;F224

【参考文献】