基于点击流数据的电子商务用户购物行为分析

发布时间：2018-05-27 04:47

本文选题：网购行为 + 点击流数据　；参考：《广东工业大学》2017年硕士论文

【摘要】：信息技术的飞速发展使得传统的实体商店已不是营销者提供商品和服务的唯一渠道,网上虚拟商店应运而生,即电子商务。人们开始发现线上用户行为挖掘的力量。然而,电子商务平台是一个虚拟平台,商家无法像实体店一般直接通过提问的方式了解用户的需求,并对应地向用户提供个性化的服务,虽然当前对于线上用户行为的挖掘和分析已经很多,但是研究结果对于感知用户内心真实需求的效果却很差。因此,想要更深入地剖析在线用户的行为,就必须把握他们内心的偏好和需求。点击流数据是用户在线上进行网购行为的直接产物,它在一定程度上反映了用户的行为轨迹。本文在用户行为和需求理论,以及相关的点击流数据分析理论的基础上,针对上述研究做了如下几项工作:第一,阐述用户行为和需求理论,以及相关的点击流数据分析的研究现状,并提出本文的意义。第二,点击流数据的获取思路主要有两种方法:基于服务器端和基于客户端。本文结合这两种方法的优缺点,设计了一个网购行为实验,同时设计了一个Chrome插件在客户端获取点击流数据,并在实验前后对实验对象进行预调查和深度访谈。第三,对实验数据进行预处理,并对处理后的数据进行初步的可视化展示。根据实验数据的预处理结果,对页面状态之间的转换进行马尔科夫描述,得到转移概率矩阵,并通过一阶马尔科夫模型对其进行简单地预测,得到实验对象在第18步转移后转移到购买成功页面的概率趋于稳定,并且概率为0.34%。第四,基于浏览频率和停留时间两个度量,提出两个综合性指标:以停留时间为主要指标和以浏览频数为主要指标。基于这两个综合性指标使用K-means算法对数据进行聚类分析,聚类结果显示用户可分为是否具有明确购买意向,而他们因为商品参数、销量和品牌等因素的影响具有不同的表现。同时,对实验后的深度访谈结果进行分析,根据用户实验前后的行为差异将用户分为意向-行为一致型和意向-行为差异型。此外,根据他们提供的网购行为偏好将用户分为价格敏感型、品牌偏好型、注重评价型、销量偏好型、信用偏好型以及注重性能型。
[Abstract]:With the rapid development of information technology, traditional physical stores are not the only channel for marketers to provide goods and services. People are beginning to discover the power of online user behavior mining. However, E-commerce platform is a virtual platform, merchants can not understand the needs of users directly like physical stores, and provide personalized services to users accordingly. Although there has been a lot of research on online user behavior mining and analysis, the results of the research on the perception of the real needs of users is very poor. Therefore, to further analyze the behavior of online users, we must grasp their inner preferences and needs. Click-stream data is the direct product of online shopping, which reflects the behavior of users to a certain extent. Based on the theory of user behavior and requirement, and the theory of click-stream data analysis, this paper has done the following work: first, the theory of user behavior and demand. And the research status of click-stream data analysis, and the significance of this paper. Secondly, there are two main ways to obtain click-stream data: server-based and client-based. Combining the advantages and disadvantages of these two methods, this paper designs an online shopping behavior experiment, and designs a Chrome plug-in to obtain click-stream data on the client side, and carries out pre-investigation and in-depth interviews on the subjects before and after the experiment. Thirdly, preprocess the experimental data and visualize the processed data. According to the preprocessing results of experimental data, the transformation between page states is described by Markov, and the transition probability matrix is obtained, and the first order Markov model is used to predict the transition probability matrix. The probability of the subjects moving to the successful page after the 18th step is stable, and the probability is 0.34. Fourthly, based on the two measures of browsing frequency and residence time, two comprehensive indexes are proposed: the residence time as the main index and the browsing frequency as the main index. K-means algorithm is used to cluster the data based on these two comprehensive indexes. The clustering results show that the users can be divided into whether they have clear purchase intention or not, but they have different performance because of the influence of commodity parameters, sales volume and brand. At the same time, the results of in-depth interviews after the experiment were analyzed. According to the behavior differences of users before and after the experiment, users were divided into two types: intention-behavior consistent type and intention-behavior difference type. In addition, users are classified into price sensitive type, brand preference type, evaluation type, sales preference type, credit preference type and performance-oriented type according to their online shopping preferences.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F713.55;F724.6

【相似文献】