基于SOM-K-means的天涯BBS水军帖的聚类分析

发布时间：2018-07-15 22:04

【摘要】：互联网的出现和普及给人们的生活带来了翻天覆地的变化，各种各样的网络论坛极大地拓展了人们信息交流和分享的渠道。然而，网络水军混杂其中，大大降低了网络信息的真实性和有效性，干扰了舆论的真实走向，甚至引发网络的信任危机。因此，了解水军群体的特征、如何识别水军帖对治理网络水军现象具有重要意义。本文从网络水军活跃的天涯BBS入手，抓取若干热帖的特征属性数据，首先进行数据预处理，通过变量相关性分析提出用户主页的注册日期、登录次数和粉丝数为三个有效聚类变量。然后，以自组织映射神经网络（SOM）聚类为预处理，找出合理聚类数N的区间值和最终聚类中心，将其作为K-means聚类的聚类数和初始聚类中心，实现SOM K-means两阶段聚类分析，以提高聚类的精确性。从聚类的视角发现水军群体的注册日期明显集中于个别日期，其粉丝数、帖子数较少，各变量数据的相似性明显，用户ID命名规则异常等特点，，并提出从注册日期的集中性、各变量数据一致性、前两页顶帖用户的注册日期集中程度和用户ID命名规则等四个方面识别水军帖，并以实例说明了该规则能够有效应用于水军帖的识别。最后，提出本研究的不足之处和未来研究方向。综上，本文按照数据的获取、处理、分析和应用的思路来组织内容，从聚类视角研究BBS帖子用户的属性特征，重点总结水军群体的特征表现，为水军帖识别提供一种有效方法。
[Abstract]:The emergence and popularization of the Internet have brought about tremendous changes in people's lives, and various network forums have greatly expanded the channels for people to exchange and share information. However, the intermingling of network forces greatly reduces the authenticity and effectiveness of network information, interferes with the true trend of public opinion, and even leads to network trust crisis. Therefore, it is of great significance to understand the characteristics of watermen and how to identify them. This paper starts with the active BBS of the network army, grabs the characteristic attribute data of some hot posts, first carries on the data preprocessing, proposes the registration date of the user's home page through the variable correlation analysis. The number of login and the number of fans are three effective clustering variables. Then, based on the self-organizing mapping neural network (SOM) clustering, the interval value and the final clustering center of the reasonable clustering number N are found out, which is used as the clustering number and the initial clustering center of K-means clustering, and the SOM K-means two-stage clustering analysis is realized. In order to improve the accuracy of clustering. From the point of view of clustering, it is found that the registration date of water army group is obviously concentrated on individual date, the number of fans and posts is less, the similarity of variable data is obvious, the rule of user ID naming is abnormal and so on, and the centrality of registration date is put forward. The data consistency of each variable, the registration date concentration of users in the first two pages and the naming rules of user ID are four aspects to identify the waterfront posts, and an example is given to show that the rules can be effectively applied to the identification of Shuijun posts. Finally, the deficiency and future research direction of this study are put forward. In summary, according to the idea of data acquisition, processing, analysis and application, this paper organizes the content, studies the attribute characteristics of users of BBS posts from the perspective of clustering, and summarizes the characteristics of water army groups, which provides an effective method for the identification of water posts.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：C913.4

【相似文献】