基于特征选择和模型融合的网络购买行为预测研究

发布时间：2018-04-09 23:42

本文选题：网络购买行为　切入点：预测　出处：《北京交通大学》2017年硕士论文

【摘要】：网络购物已成为人们日常生活中必不可缺的一部分。网络购物中顾客和商家不需要面对面交易,这使得商家不能很好地把握消费者的想法和需求。但是顾客的购物行为的任何一个细节却服务器记录着,这使得通过分析这些行为数据来了解消费者的偏好甚至实现预测其购买行为成为可能。因此本文提出了使用大数据分析方法——机器学习算法从大量的消费者历史网购行为数据中学习出隐含在其中的购买模式获得模型,当新的顾客购物行为数据被输入到该模型中时,即可实现对顾客购买行为的预测。本文首先对网络购买行为的影响因素和预测研究进行了文献综述,深入了解网络购买行为的本质并发现目前基于大数据分析的网络购买行为研究仍处于起步阶段。所以本文以阿里巴巴举办的大数据竞赛作为研究背景,并将用户在阿里巴巴电子商务平台上真实的购物行为数据作为研究数据,通过使用机器学习算法对网络购买行为进行建模。首先使用Sql Server在原始数据的基础上构造了 322个特征,并基于Extra-trees算法提取出对于预测购买行为最有帮助的10大特征。然后本文选择了两种常用的机器学习算法:逻辑斯特回归和支持向量机,将这10个特征分别输入两个算法得到两个预测模型。最后本文基于Soft-voting的方法对以上两个算法进行融合。实验证明,融合后的模型较单一的模型具有更好的预测效果。本文的研究以数据为驱动,旨在实证说明使用消费者的历史购物行为预测其未来购买行为的可行性。本文的预测模型可以被用于购物网站的推荐系统中,实现用户界面的完全个性化,激发顾客的购买欲望,提高电子商务平台的转化率。
[Abstract]:Online shopping has become an essential part in people's daily life. The online shopping customers and businesses do not need face-to-face transactions, the business is not very good grasp of consumer's ideas and needs. But any details of a customer's shopping behavior is a record of the server, through the analysis of these data to understand consumer behavior the prediction of their buying behavior preference even possible. Therefore this paper proposes the use of large data analysis methods, machine learning algorithms from the consumer behavior of online shopping history of a large number of data obtained in the model of implicit learning mode of purchasing them, when the customer shopping behavior data are input to the model, can realize the prediction of customer purchase behavior. Firstly, influence on online purchasing behavior factors and prediction research conducted a literature review, in-depth understanding of the network The essence of purchase behavior and found that the current analysis of large data network based on purchasing behavior research is still in its initial stage. So the Alibaba held a big data race as the research background, and the shopping user behavior data in the Alibaba real e-commerce platform as the research data, the learning algorithm to model the network purchasing behavior using the machine. The first to use Sql Server 322 features are constructed based on the original data, and Extra-trees algorithm to extract the features of 10 help for the prediction based on the purchase behavior. Then this paper chooses two kinds of machine learning algorithms: logistic regression and support vector machine, and the 10 features are input to the two algorithm two model. Finally fusion of the above two algorithms based on the Soft-voting method. The experimental results show that after fusion The prediction of the model is single model has better. This study based on data driven, empirical to illustrate the history of consumer shopping behavior to predict the feasibility of future purchase behavior. The prediction model can be used to recommend shopping site, is completely personalized user interface now, stimulate the customers desire to buy, to improve the conversion rate of e-commerce platform.

【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F713.55

【参考文献】