电子商务中的评论挖掘及应用研究

发布时间：2018-06-18 18:55

  本文选题：人类行为学 + 标度律　； 参考：《电子科技大学》2014年硕士论文

【摘要】：电子商务中用户的评论意见为潜在消费者提供了重要的参考依据。虚假评论在一定程度上误导了消费者的消费倾向,使得消费者逐渐丧失对电子商务评价系统的信任。准确高效的对系统中的虚假评论进行检测和清除成为一个必须解决的关键问题。本论文针对上述问题,在深入分析电子商务中人类行为特征及演化规律的基础上,基于人类行为学理论提取用户评论的行为特征,采用模式挖掘的方式对个体虚假评论用户以及虚假评论群组进行检测。主要研究内容如下:1.基于实际数据,采用去趋势波动分析法首次对电子商务中人类行为活动标度律进行实证研究。研发发现人类购买和浏览行为的演化过程不同于无关联的泊松过程,具有较强的长程关联特性和自组织临界性。2.个体虚假评论用户检测。提出了基于用户行为聚类和基于用户经验演化两种检测算法。其中,基于用户行为聚类的检测算法通过对用户评分行为聚类来计算用户的从众强度,采用二阶段检测的方法对用户的信誉大小做出评估。基于用户经验演化的检测算法根据用户行为随时间的演化特性,利用稳定性和无序性两个行为特征来刻画用户经验大小,对用户可信度进行判断。通过实验验证,两种检测算法均具有较高的检测准确度,优于与之对比的基准检测算法,尤其在用户评分较为稀疏的现实评价系统中检测效果十分出色。3.虚假评论用户群组检测。基于攻击利益最大化假设,采用关联规则对系统中的虚假评论群组进行检测。算法不仅可以挖掘虚假评论群组规模大小,也可以有效的对指定规模的群组进行检测。4.实现上述三种检测算法,从数据导入,虚假评论用户及群组检测,结果查看等方面,实现检测系统,完成了对电子商务中虚假评论用户的检测任务。本课题的创新点在于不依赖具体的评论形式和内容,从分析用户评论行为出发,提出基于评论行为聚类和经验演化的两种虚假评论用户检测算法,在保证检测准确度的同时提高了检测效率,具有较强的实用性。
[Abstract]:The comments of users in e-commerce provide important reference for potential consumers. To some extent, false comments mislead consumers' propensity to consume and make consumers lose their trust in electronic commerce evaluation system. Accurate and efficient detection and removal of false comments in the system has become a key problem that must be solved. In this paper, based on the analysis of the characteristics and evolution of human behavior in electronic commerce, the behavioral characteristics of user comments are extracted based on the theory of human behavior. The users and groups of individual false comments are detected by pattern mining. The main research contents are as follows: 1. Based on the actual data, the de-trend volatility analysis method is used for the first time to make an empirical study on the scale law of human behavior in electronic commerce. It is found that the evolutionary process of human purchase and browsing behavior is different from the independent Poisson process and has strong long-range correlation and self-organized criticality. Individual false comment user detection. Two detection algorithms based on user behavior clustering and user experience evolution are proposed. Among them, the detection algorithm based on user behavior clustering calculates the user's herd strength by clustering the user's scoring behavior, and evaluates the user's reputation by using the two-stage detection method. The detection algorithm based on the evolution of user experience describes the size of user experience according to the evolution characteristics of user behavior over time and uses two behavioral characteristics of stability and disorder to judge the reliability of users. The experimental results show that the two detection algorithms have high detection accuracy and are superior to the benchmark detection algorithm, especially in the practical evaluation system with sparse user scores. False comment user group detection. Based on the hypothesis of maximization of attack benefits, association rules are used to detect false comment groups in the system. The algorithm can not only mine the size of the false comment group, but also detect the specified size group effectively. The above three detection algorithms are implemented from the aspects of data import, false comment user and group detection, result checking and so on. The detection system is implemented and the detection task of false comment user in electronic commerce is completed. The innovation of this paper is that it does not depend on the specific form and content of comments, and from the analysis of user comment behavior, two kinds of false comment user detection algorithms based on comment behavior clustering and empirical evolution are proposed. At the same time, it improves the detection efficiency and has strong practicability.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F713.36;TP391.1
，

本文编号：2036526