基于多标签双词主题模型的短文本谣言分析研究  

Analysis of Short Text Rumors via Multi-Label Biterm Topic Model

作　　者：武庆圆[1] 何凌南[2] 

机构地区：[1]北京师范大学珠海分校管理学院,珠海519087 [2]中山大学传播与设计学院,广州510006

出　　处：《情报杂志》2017年第3期92-97,共6页Journal of Intelligence

基　　金：广东省软科学研究计划项目"基于多维引用深度分析的科技期刊学术影响力评价研究"(编号:2014A030304013)研究成果之一

摘　　要：[目的/意义]面对互联网与智能移动设备的兴起,谣言尤其是短文本类型的谣言发展速度十分迅猛。短文本谣言具有词语稀疏、语义提取困难等特点,这为精准识别谣言带来了挑战和困难。如何能够有效地鉴别进而控制谣言的传播是目前迫在眉睫的问题。[方法/过程]提出一个在文本与标签之间引入语义层的多标签双词主题模型,用于发现及探究网民发表在公共媒体平台上的短文本属于谣言或欺诈的倾向。该研究专门针对微信等短文本数据,并通过真实数据集对双词主题的提取和建模进行验证。[结果/结论]结果表明:上述模型可以有效鉴别谣言,帮助媒体加强和改进监管机制,遏制网络谣言、欺诈等现象。[ Purpose/Significance ] With the development of Intemet and smart mobile devices, the number of rumors increases very fast, especially the short text minors. Short text contains fewer words and is difficult to be analyzed semantically, which brings challenges and difficulties for rumor identification. Thus, it is valuable to effectively identify short text rumors and control the spread of them. [ Method/ Process] To measure and explore the rumor or fraudulence tendency of short text that is published by netizens through the public media platform, this paper proposes a multi-label biterm topic model which introduces a semantic layer between text and labels. The research particularly focuses on WeChat or other types of short text, and employs real-world datasets to verify the effectiveness of biterms＂ topic extraction and modeling. [ Results/Conclusion] Experimental results indicate that the model proposed could identify rumors effectively, help the media to strengthen and improve the supervision mechanism, and thus curb the network rumors, fraudulence and other phenomena.

关 键 词：多标签 双词 短文本 谣言 主题模型 微信 语义分析 

分 类 号：G202[文化科学—传播学]