基于用户交互关系的用户创作内容质量评估

发布时间：2018-03-22 21:22

本文选题：社会网络分析　切入点：链接分析　出处：《北京邮电大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着Web2.0的发展日渐成熟,越来越多的网站信息资源依赖于用户的参与、主导和建设。这些由用户自发创造,并在传播的过程中不断丰富和壮大的信息资源被称为用户创作内容(User-generated content, UGC)。目前,针对UGC的相关理论研究得到了各行业广泛的关注和重视,同时也取得了很多有价值的研究成果。但是由于UGC具有人人均可参与创作的特点,导致其通常含金量很低、信息质量参差不齐。在数据挖掘、舆情分析、热点发现、信息搜索等众多基于UGC的研究工作中,对于UGC自身内容质量的合理评估就变得十分必要。在本文中,我们对UGC用户间的交互关系进行分析,衡量用户的权威程度,并在此基础上结合主题相似度分析过程,提出了一种评估用户创作内容质量的方法。我们综合考虑了用户间静态的社交关系和动态的交互行为关系,构建用户关系网络,在社会网络分析过程中分析用户所处网络节点所具有的社会性质并整合多种信息构建用户特征向量。使用基于用户真实度的链接分析方法和神经网络回归模型对用户自身的权威度进行评分。用户的权威度代表其影响力大小,并能在一定程度上反映其发布内容的价值。同时,我们考虑了用户权威性的主题性,根据网络社区中用户创作内容多人参与的特点,使用作者-主题模型(AT)分析用户兴趣点的主题分布。最终结合用户在UGC生成过程中的实际表现,计算出一段多人参与的用户创作内容的内容质量。为验证结果的有效性,我们对综合型社交论坛(天涯论坛)进行了较大规模的数据采集,并在真实数据集合上针对不同的影响因素进行了多次对比实验。同时,我们设计出了比较有说服力的评估方法对结果进行分析。最终结果表明,本文提出的方法在综合考虑多种因素时对于UGC自身质量可以做到更加合理的评估。本论文研究得到国家自然科学基金项目资助,编号71231002和61202247。
[Abstract]:With the development of Web2.0, more and more website information resources depend on the participation, leading and construction of users. And in the process of dissemination, the rich and growing information resources are called User-generated content (UGC). At present, the related theoretical research on UGC has been widely concerned and paid attention to by various industries. At the same time, many valuable research results have been made. However, because UGC has the characteristics that everyone can participate in the creation, it usually has very low gold content and uneven information quality. In data mining, public opinion analysis, hot spot discovery, In many research work based on UGC, such as information search, it is very necessary to evaluate the content quality of UGC itself. In this paper, we analyze the interaction between UGC users and measure the authority of users. On the basis of this, a method of evaluating the quality of user-created content is proposed based on the process of subject similarity analysis. We consider the static social relationship and dynamic interactive behavior relationship between users, and construct the user relationship network. In the process of social network analysis, the social properties of the nodes in which users are located are analyzed and the user feature vectors are constructed by integrating various kinds of information. The link analysis method based on user truthfulness and the neural network regression model are used to construct the user feature vectors. The user's own authority is rated. The user's authority represents the size of his influence, And can reflect the value of the content to some extent. At the same time, we consider the theme of user authority, according to the characteristics of users in the network community, The author-topic model (ATT) is used to analyze the topic distribution of user's point of interest. Finally, the content quality of a segment of user's creative content is calculated according to the user's actual performance in the process of UGC generation. In order to verify the validity of the results, we conducted a large scale data collection on the comprehensive social forum (Tianya Forum), and conducted several comparative experiments on different influencing factors on the real data set. At the same time, We have designed a more persuasive evaluation method to analyze the results. The final results show that the method proposed in this paper can be used to evaluate the quality of UGC in a more reasonable way when a variety of factors are taken into account. This thesis is supported by the National Natural Science Foundation of China, No. 71231002 and 61202247.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09;TP391.1

【相似文献】