微博中影响力的研究

发布时间：2018-05-27 22:21

本文选题：社会媒体 + 微博　；参考：《中国科学技术大学》2014年硕士论文

【摘要】：随着Web2.0技术的发展,各种类型的社会媒体如雨后春笋般先后出现,而微博,作为其中重要的一种类型,已成为人们之间相互沟通、传播信息的重要途径。由于微博存在一些特殊的性质,近年来,针对微博的研究日益增多,其中影响力的研究成为热点问题之一,包括影响力度量和影响力最大化两大主题。尽管已经取得了一些成果,但是也都仍然存在一些问题,比如在已提出的影响力最大化算法中,不是运行时间过长,就是准确度不高,或者是假设条件太多。在影响力度量问题上,也仍存在一些新的角度可以去探讨。因此,本文将从这两个方面对微博中的影响力展开研究。本文先根据用户交互行为的双向性,从转发力度、评论强度、提及密度和关键字相似度四个因素对用户交互程度进行评估。然后,考虑到一个用户对其关注者的影响力贡献值是变化的,且依赖于双方的交互行为,提出一种新的影响力度量算法MBUserRank,并在此基础上,对影响力最大化问题展开讨论。接着,为了更好地近似激活概率并降低复杂度,修改MBUserRank算法并得到MBRank算法。通过MBRank对全部用户排序,选出一部分排名靠前的用户组成候选种子节点集合,在此基础上引入启发,提出新的影响力最大化算法MBGreedy及其改进算法MBCELFo最后,对生活中一个常见问题——影响力排名前k的用户是否一定会导致影响力最大化进行了探索,并对候选种子集合大小的选择进行了说明。通过在腾讯微博数据集上大量实验,得出如下结论：1)、新提出的影响力度量算法MBUserRank能够给出更贴近现实的排序结果；2)、新提出的影响力最大化算法MBCELF结合了原始贪心算法和启发式算法的优点,能够在较短时间内取得近似最优解；3)、影响力排名前k的用户不一定会导致影响力最大化。
[Abstract]:With the development of Web2.0 technology, various types of social media have appeared, and Weibo, as one of the important types, has become an important way for people to communicate and disseminate information. Due to some special properties of Weibo, the research on Weibo has been increasing in recent years, among which the research of influence has become one of the hot issues, including the measurement of influence and the maximization of influence. Although some achievements have been made, there are still some problems, such as too long running time, low accuracy, or too many assumptions in the proposed algorithm. There are still some new points of view on the measurement of influence. Therefore, this paper will study the influence of Weibo from these two aspects. Based on the bidirectional behavior of user interaction, this paper evaluates the degree of user interaction from four factors: forwarding intensity, comment intensity, reference density and keyword similarity. Then, considering that the value of a user's impact contribution to his followers is variable and dependent on the interaction between the two parties, a new influence measurement algorithm, MBUserRank, is proposed, and on this basis, the problem of maximizing the influence is discussed. Then, in order to approximate the activation probability and reduce the complexity, the MBUserRank algorithm is modified and the MBRank algorithm is obtained. All users are sorted by MBRank, and some of the top ranking users are selected to form candidate seed node sets. On the basis of this, a new influence maximization algorithm MBGreedy and its improved algorithm MBCELFo are proposed. This paper explores a common problem in life, whether the user in the top k of influence will lead to the maximization of influence, and explains the selection of the size of the candidate seed set. By experimenting extensively with Tencent Weibo datasets, The conclusions are as follows: 1: 1, the new influence measurement algorithm MBUserRank can give a more realistic ranking result, and the new influence maximization algorithm MBCELF combines the advantages of the original greedy algorithm and heuristic algorithm. It is possible to obtain approximate optimal solutions in a short period of time, and users with the top k of influence do not necessarily lead to maximum impact.
【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】：2014
【分类号】：TP393.092

【参考文献】