多因素融合下的微博话题可信度评估模型及实证研究

项目名称： 多因素融合下的微博话题可信度评估模型及实证研究

项目编号： No.71303179

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 管理科学

项目作者： 王平

作者单位： 武汉大学

项目金额： 19万元

中文摘要： 本项目研究并建立微博话题可信度量化评估模型。首先从理论上研究影响微博话题可信度评估的因素；然后分析微博之间的跟帖关系，并建立扩展的LDA话题模型进行微博话题抽取；最后以话题为粒度对微博可信度进行定量计算。计算微博可信度考虑以下三个因素：1）微博用户的可信度；2）微博和话题之间的关系；3）微博用户和微博之间关系。通过建立包含微博用户、微博、话题这三层的神经网络，并利用这三个因素计算神经网络每条边的权重，神经网络的输出值即为话题的可信度值。模型的有效性利用新浪微博数据进行验证。本项目的研究对于如何有效地挖掘微博数据、如何评估微博质量有着重要的理论意义，也具有广泛的应用前景。

中文关键词： 学术实体信息；社会媒体微博信息；主题抽取；可信度评估；影响因素

英文摘要： In the project, we will investigate and propose quantitative credibility model of microblog topic. In the first phase,we plan to analyze which factors will have impact on the credibility evaluation of microblog topic. For phase two, an extended LDA model, in which the posting relationsip between microblog is fully utlized, will be proposed in order to extract topics from microblog more accurately. In phrase three, we will investigate how to quantitatively calculate the credibility of microblog at topic granularity. Microblog credibility calculation will take the following factors into account:1)the credibility of microblog user;2)the relevance of microblog and microblog user;3) the relevance of microblog and extracted topic. At meanwhile, we will utilize neural network to model the relevance of these factors. The edges in the neural network are weighted by the relevance of these factors. Finally, the output of neural network are used as the credibility value of extracted topics. The credibility model of microblog can be evaluated by Sina microblog dataset.The project will have significantly theoretical contributions for effectively mining microblog data and evaluating the quality of microblog. The project will also have broader impacts on realistic applications.

英文关键词： Academic entity information；Social media microblogging information；Topic extraction；Credibility evaluation；Affecting factors