社交网络环境下的多标签分类研究

发布时间：2018-07-25 18:58

【摘要】：随着社交网络的快速发展,出现了像Facebook、Twitter和YouTube这样成功拥有海量用户的社交网站。社交网络作为一种共享知识、与朋友联系互动的媒介,在我们生活中起到越来越重要的作用。标签分类是社交网络中的一项重要应用,例如在社交网络中的用户具有兴趣标签和好友关系标签。此外,用户也可以给社交网络中的各种文本、图片、视频信息打标签。在传统标签分类中,网络数据由单个标签表示。但随着各种社交网络应用的丰富,网络数据的形式也越来越多样化,单个标签已无法满足社交网络数据复杂和多语义的特性。因此,社交网络环境下的多标签分类研究得到了越来越多的关注。基于此,本文将针对社交网络结构分析、社交网络环境下的多标签分类以及多标签在推荐系统中的应用三个方面进行研究。本文的主要工作如下：(1)介绍了社交网络环境下多标签分类的产生背景和研究意义,分析了社交网络结构分析、多标签分类以及推荐系统的研究现状和研究缺陷,并详述了相关领域的概念、分类、关键参数和经典算法。(2)提出一种基于链接寿命的社交网络结构分析方法。将链接寿命加入社交网络结构分析中,研究链接寿命对于社交网络结构中重要的基础参数(包括度,网络直径和平均聚类系数等)的影响。实验表明,加入链接寿命后,社交网络的演化结构和传统的研究有很大的不同,特别是,链接寿命的微小变化会导致网络直径的剧烈变化。(3)在上述社交网络结构的基础上,提出了两种半监督的多标签分类算法。在两种经典的关系型分类器的基础上,加入must-link约束和不确定性概率,研究must-link约束对于多标签分类的影响。实验表明,该方法在大规模社交网络上比经典关系型分类器具有更好的分类精度和效率,尤其当已知标签数量很少的时候。(4)在上述算法计算得出的社会标签的基础上,提出了一种多源评价聚合算法。首先基于评分者的社会标签计算他们的权威程度,然后将权威程度加入多源评价聚合过程中,来更加准确的评估实体的真实得分。实验表明,该方法能有效消除推荐系统中的严格推荐者和宽松推荐者带来的干扰噪音,并且无需任何关于严格和宽松推荐者比例的先验信息。
[Abstract]:With the rapid growth of social networks, social networking sites such as Facebook Twitter and YouTube have become successful with a large number of users. As a medium for sharing knowledge and interacting with friends, social networks play an increasingly important role in our lives. Label classification is an important application in social networks, such as users with interest tags and friends tags in social networks. Users can also tag text, pictures, and video messages on social networks. In traditional label classification, network data is represented by a single tag. However, with the abundance of various social network applications, the forms of network data are becoming more and more diverse. A single label can no longer satisfy the complex and multi-semantic characteristics of social network data. Therefore, more and more attention has been paid to the classification of multiple tags in the social network environment. Based on this, this paper will focus on three aspects: the analysis of social network structure, the classification of multi-label in social network environment and the application of multi-label in recommendation system. The main work of this paper is as follows: (1) the background and significance of multi-label classification in social network environment are introduced, and the research status and defects of social network structure analysis, multi-label classification and recommendation system are analyzed. The concepts, classification, key parameters and classical algorithms of related fields are also described in detail. (2) A social network structure analysis method based on link life is proposed. The link life is added to the analysis of the social network structure to study the influence of the link life on the important basic parameters (including degree, network diameter and average clustering coefficient) in the social network structure. The experimental results show that the evolutionary structure of social network is very different from the traditional research after adding link life, especially, the small change of link life will lead to the drastic change of network diameter. (3) based on the above social network structure, Two semi-supervised multi-label classification algorithms are proposed. On the basis of two classical relational classifiers, the influence of must-link constraints on multi-label classification is studied by adding must-link constraints and uncertainty probability. Experiments show that this method has better classification accuracy and efficiency than the classical relational classifier on large-scale social networks, especially when the number of known labels is small. (4) on the basis of the social labels calculated by the above algorithms, A multi-source evaluation aggregation algorithm is proposed. Firstly, based on the social label of the raters, the degree of authority is calculated, and then the degree of authority is added to the aggregation process of multi-source evaluation to evaluate the real score of the entity more accurately. Experiments show that the proposed method can effectively eliminate the interference noise caused by strict and loose referrals in the recommendation system and does not require any prior information about the proportion of strict and loose referrals.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP393.09

【相似文献】