电商用户消费行为预测与心理建模方法研究

发布时间：2018-05-24 05:45

本文选题：电商用户画像 + 行为序列　；参考：《中国科学技术大学》2017年硕士论文

【摘要】：近年来,电子商务的飞速发展给人们的生活带来了巨大便利,人们越来越倾向于从电商网站中直接在线获取所需的商品和服务。在促使用户改变消费习惯的同时,电商网站也在后台记录了海量用户行为日志。为提升用户体验,适宜地推荐商品,这些日志信息已经被广泛应用于电商网站的画像和推荐系统。为了构建完善有效的画像和推荐系统,首先需要对用户未来的消费情况做出预测。目前电商平台大都以统计和建模历史消费信息的方式达成该目标,而较少考虑用户历史行为序列中包含的比较和选择信息。这种方式在商品数量日益繁多的今天已不再适用。因此本文提出了一种基于选择模型的方法,将用户历史消费情况与行为序列特征相结合,以更好地表达用户偏好,预测其消费情况。更进一步,用户消费时的行为表现由其心理状态决定,研究用户消费心理将会有助于理解用户需求、提供多元智能化服务。基于调查问卷的传统用户消费心理研究方法不仅耗时耗力而且具有很强的主观性。而大数据时代的到来,使利用历史行为数据建模用户心理逐渐成为可能。基于此,本文对用户消费时的犹豫心理进行了研究,提出了一种数据驱动的犹豫心理建模方法。总结来看,本文的主要研究内容和贡献如下:1)对基于选择模型的用户消费预测方法进行了研究。本文首先引入机会成本的概念并使用一个序列效用函数预估用户每个Session中的最佳替代品;接下来在每个Session中的被购买商品和最佳替代品之间建立基于潜在因子的选择模型;更进一步,利用Session中的所有比较信息提出了将潜在因子和行为序列效用结合的选择模型;最后,使用天猫网站的真实数据集验证了提出算法的有效性。2)对数据驱动的用户犹豫心理建模和应用方法进行了研究。本文首先根据用户行为序列表现出的几种特征定义和计算每个Session的可观测犹豫指数Ds;然后考虑到Ds受用户和商品的共同影响,构建了用户-商品的犹豫矩阵分解模型将Ds分解为用户和商品的犹豫因子;接下来,探讨了几种利用用户犹豫心理建模的结果提供多元智能服务的方法;最后,同样在天猫网站的真实数据集上对提出的方法进行了验证。
[Abstract]:In recent years, the rapid development of electronic commerce has brought great convenience to people's life. People are more and more inclined to obtain the needed goods and services directly from e-commerce websites. While urging users to change their consumption habits, e-commerce websites also record massive user behavior logs in the background. In order to enhance the user experience and recommend products appropriately, the log information has been widely used in the portrait and recommendation system of e-commerce websites. In order to construct an effective portrait and recommendation system, it is necessary to predict the future consumption of users. At present, most e-commerce platforms achieve this goal by means of statistics and modeling of historical consumption information, while less consideration is given to the comparison and selection information contained in the historical behavior sequence of users. This approach is no longer applicable at a time when there are more and more goods. Therefore, this paper proposes a method based on selection model, which combines the historical consumption of users with the characteristics of behavior sequence to better express user preferences and predict their consumption. Furthermore, the behavior of users is determined by their psychological state. The study of consumer psychology will help to understand the needs of users and provide multiple intelligent services. The traditional research method of consumer psychology based on questionnaire is not only time-consuming and labor-intensive, but also highly subjective. With the advent of big data, it is possible to use historical behavior data to model user psychology. Based on this, this paper studies the hesitancy psychology of user consumption, and proposes a data-driven modeling method of hesitant psychology. In conclusion, the main contents and contributions of this paper are as follows: 1) the user consumption prediction method based on selection model is studied. This paper first introduces the concept of opportunity cost and uses a sequential utility function to estimate the best alternatives in each Session, and then establishes a selection model based on potential factors between the purchased goods and the best alternatives in each Session. Furthermore, using all the comparative information in Session, a selection model combining potential factor and behavioral sequence utility is proposed. The validity of the proposed algorithm is verified by the real data set of Tmall website. 2) the modeling and application methods of data-driven user hesitancy are studied. In this paper, we first define and calculate the observable hesitation index (DS) of each Session based on several characteristics of user behavior sequence, and then consider that Ds are influenced by both user and commodity. The user-commodity hesitation matrix decomposition model is constructed to decompose Ds into user and commodity hesitation factors. Then, several methods to provide multiple intelligent services based on the results of user hesitation modeling are discussed. The proposed method is also validated on the real data set of Tmall website.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F713.55;F713.36

【相似文献】