基于知识关联特征的网络内容识别——以健康谣言为重点

Content Detection Based on Knowledge Relation: Cases of Health Rumors

社会普遍信息化使互联网成为最便捷的知识获取途径,但随着信息发布和传播的门槛越来越低,谣言频现成为互联网时代的社会问题,与个人生活紧密相关的健康话题则成为了谣言重灾区。通过分析微信和今日头条两平台中谣言内容特征,结合新闻传播学和计算机科学的现有研究成果,提出如何通过知识关联特征进行谣言内容识别的研究问题。在分析问题的过程中,针对今日头条谣言库的减肥谣言高频词团进行社会网络分析和可视化处理,发现共现关系比相似关系更有助于发现知识类谣言的常用搭配和关键话题。最后,从规范性研究的角度,提出结合知识图谱和标签两项技术建构新生谣言的识别机制。

Ubiquitous informatization in the society enables the Internet to be the most efficient access for acquiring knowledge.However,with the decrease of barriers for releasing and distributing information,frequent occurrence of rumors has become a social problem.Moreover,health topics are closely related to individuals' lives,so they are the worst-hit area of rumors.Based on content analysis of rumors in WeChat and Toutiao.com,the research question of how to detect rumors through the characteristics of knowledge relation is proposed.In order to explore and present the knowledge relation,a social networking analysis tool was adopted to visualize the cluster of high frequency words of weight-loss rumors that were refined from the rumor base at Bytedance.With the analysis through visualization,it is found that the co-occurrence relation is better than the similarity relation at discovering common phrases and popular topics of knowledge-based rumors.Lastly,in a normative perspective,a mechanism for detecting newborn rumors is constructed,which should integrate technologies of knowledge graph and labeling.