面向社会媒体的用户在线社交圈识别与分析

发布时间：2017-12-27 15:39

本文关键词：面向社会媒体的用户在线社交圈识别与分析　出处：《哈尔滨工业大学》2016年博士论文　论文类型：学位论文

【摘要】：社会媒体已经成为人们在互联网中信息传播的重要平台,用户在社会媒体中关注各种信息源并与好友进行互动。这些行为导致了社会媒体中形成了庞大的用户社交网络,同时,用户线下大量的真实社会关系也存在于这种在线社交网络中。在以往的社会科学中,有限的调研形式和稀缺的数据给人与人社会关系的研究造成了很大的困难。社会媒体易于获取的数据使人与人之间社交强关系和社交圈的研究变成可能,这种线上熟人社交圈中的用户具有非常高的同质性,同一社交圈中的用户会很大程度的影响彼此在社交网络中的行为。因此,用户在线社交圈的研究是社会媒体中用户分析相关研究的重要基础。每个用户都在社会媒体中拥有大量的好友,用户和这些好友之间都社交关系强度各不相同,并且很难加以区分。同时,每个用户在社会媒体中大多拥有不同的社交圈,这些社交圈互相独立,代表用户不同的社会关系,例如高中同学、大学同学等。用户社交圈是用户社会强关系的一种典型的表现形式,识别并分析这些社交圈可以反映用户不同的社会维度,然而,社交圈具有很强的主观性和私密性,每个用户只能了解自己社交圈的好友构成和社交圈的意义,因此研究人员很难在社会媒体中直接获取一个用户的社交圈好友和社交圈意义等相关数据。为了解决以上问题,本文从以下方面开展了对用户在线社交圈识别和分析的相关研究。1、基于用户关注关系的在线社交圈识别。每个社交圈都是用户的社会强关系,因此每个社交圈内部的成员之间都彼此连接紧密。根据这个原理,本文提出了一种基于在线凝聚聚类的用户社交圈识别算法,并在用户相似度计算中引入了用户之间的社交属性,可以准确的识别用户在社会媒体中的多个不同的社交圈。为了解决主观性的社交圈数据难以获取的问题,本文建立了激励用户标注自己的社交圈成员的众包平台,该方法有效的获取了可供用户社交圈相关研究的真实数据。2、基于用户多维特征的在线社交圈识别。社交圈内的成员不仅在网络结构上连接紧密,而且在个人资料、兴趣爱好等方面具有很强的同质性。现有的社交圈识别方法利用网络结构特征和个人资料特征都分别取得了很好的识别效果,然而这些方法很难把用户在不同维度的特征结合起来。本文提出了基于矩阵分解的潜在因子联合模型,模型可以通过学习用户不同的维度特征得到特征融合后的用户向量,实验证明,与利用用户单一维度的特征模型相比,该模型通过融合多种用户特征有效的提高了用户社交圈识别的准确率。3、基于多元线性回归的用户在线社交圈标签挖掘。作为用户的社交强关系,每个社交圈都有各自不同的社会意义。每个社交圈成员在社会媒体中都有自己的标签,同一社交圈内成员的一些共同标签可以代表这个社交圈的意义,然而社会媒体中用户标签的稀少甚至缺失造成了用户标签数据的缺失,进一步给社交圈的标签挖掘带来困难。本文提出了一种基于多元线性回归的社交圈标签识别方法,同时融合和用户标签本身的特征和社交圈内的网络结构特征,为每个标签在社交圈内计算一个权重,权重大的标签更可能作为社交圈的代表性标签。与相关方法相比,该方法解决了标签数据的稀疏问题,提升了社交圈标签的识别效果。4、基于用户在线社交圈的用户个人资料补全。用户个人资料是用户在社会媒体中的重要特征,大量的用户个人资料缺失使用户资料补全成为近年来的热点研究方向。已有的用户资料补全方法大多基于用户文本,文本特性的变化和噪声给这类方法带来很大干扰。用户社交圈是用户社交强关系的体现,不同的社交圈代表了用户不同的社会维度。基于这个原理,本文提出了基于非负矩阵分解模型的用户个人资料补全方法,方法通过用户的不同社交圈补全用户不同社会维度的个人资料,保证了用户个人资料的多样性,相比已有方法提高了用户个人资料补全的性能。综上所述,本文开展了用户在线社交圈识别与分析的一些相关研究工作。相关的技术适用于主流社会媒体的用户社交强关系和社交圈的分析。在研究中取得了一些初步的结论和成果,希望能对社会媒体中用户分析的相关工作有所裨益。
[Abstract]:Social media has become an important platform for people to communicate information in the Internet. Users pay attention to various sources of information in social media and interact with friends. These behaviors lead to the formation of huge user social networks in social media, and at the same time, a large number of real social relationships on the user line also exist in this online social network. In the past social sciences, limited research forms and scarce data have caused great difficulties in the study of human and human relations. Easy access to social media data in the study of social relations and strong social circle between people becomes possible, the online social circle of acquaintances in the user has very high homogeneity, will greatly influence the behavior of each other in a social network with a social circle of users. Therefore, the research of user online social circles is an important basis for the research of user analysis in social media. Every user has a large number of friends in the social media, and the intensity of social relationships between users and these friends is different and difficult to distinguish. At the same time, most users have different social circles in social media. These social circles are independent of each other, representing different social relationships of users, such as high school classmates, college students, etc. Users social circle is a typical form of user strong social performance, recognition and analysis of the social circle can reflect the different users of the social dimension, however, social circle is very subjective and privacy, each user can only understand their own social circle of friends and social circle, so it is difficult for researchers to in social media in direct access to a user's social circle of friends and social circle significance and related data. In order to solve the above problems, this paper has carried out the following research on the identification and analysis of user online social circles. 1. Online social circle recognition based on user concerns. Each social circle is a strong social relationship of the user, so the members of each social circle are connected to each other. According to this principle, a user social circle recognition algorithm based on online agglomerative clustering is proposed in this paper, and the social attributes between users are introduced into user similarity computation, which can accurately identify users in different social circles in social media. In order to solve the problem that the social circle data is difficult to obtain, a crowdsourcing platform is established to encourage users to annotate their social circle members. This method effectively gets the real data that can be related to users' social circles. 2. Online social circle recognition based on user multidimensional features. Members of the social circle are not only closely connected to the network structure, but also have strong homogeneity in personal data, interests, and so on. The existing social circle recognition methods have achieved good recognition results by using network structure and personal data characteristics. However, these methods are difficult to integrate users' characteristics in different dimensions. This paper presents a joint model of latent factors based on matrix decomposition, the model can be obtained by learning the characteristics of the user different user dimension vector, after feature fusion experiments show that compared with the characteristics of user model using single dimension, the model through the integration of a variety of user features can improve the accuracy of user circle recognition. 3. User online social circle label mining based on multiple linear regression. As a user's social relationship, each social circle has its own social significance. Each circle members have their own label in social media, some common label the same social circle members can represent the social circle, but the user tags in social media are even caused by the lack of lack of user tag data, and further to the social circle label mining difficult. This paper presents a circle label recognition method based on multiple linear regression, network structure and user and fusion tags itself characteristics and social circle, for each label to calculate a weight in the social circle, the right major label is more likely as the representative of social circle label. Compared with the related methods, this method solves the sparse problem of the label data and improves the recognition effect of the social ring label. 4, based on online social circle user personal data complement. User profile is an important feature in the social media users in the user profile is missing a number of user data completion has become a hot research direction in recent years. Most of the existing user data imputation method based on user text, change and noise characteristic of the text brings great interference to this kind of method. The social circle of the user is the embodiment of the strong social relationship of the user, and the different social circles represent the different social dimensions of the user. Based on this principle, this paper proposes customer personal data imputation method of non negative matrix factorization based methods by different users with different social circles complete social dimension of personal information users, to ensure the diversity of users' personal data, compared with the existing methods to improve the performance of user profile completion. To sum up, some related research work on the identification and analysis of user online social circles is carried out in this paper. The related technology is applicable to the analysis of social relations and social circles of users in the mainstream social media. Some preliminary conclusions and results have been obtained in the study, and it is hoped that it will be beneficial to the relevant work of user analysis in social media.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：C912.3;TP301.6

【参考文献】