基于时间序列的微博谣言检测

Microblog Rumors Detection Based on Time Series

微博的低门槛造就了谣言产生的低成本,致使微博成为谣言信息的温床。因此,快速有效地检测谣言对微博至关重要。论文提出基于时间序列的微博谣言检测方法。为了提高谣言事件检测的性能,针对时间序列划分方法进行研究,提出基于聚类的微博事件划分方法,根据微博在时间上的聚合程度构建时间序列。同时基于GRU网络构建事件分类模型,自动学习特征用于谣言检测。实验结果表明,检测准确率达到96.7%,验证了该方法在谣言检测问题上的有效性。

The low threshold of Microblog creates the low cost of rumor generation,which makes Microblog become the hotbed of rumor information. Therefore,the rapid and effective rumors detection is crucial to Weibo. This paper proposes a time series based rumor detection method for Microblog. In order to improve the performance of rumor event detection,this paper studies the time series division method,proposes a clustering based microblog event division method,and constructs a time series according to the degree of microblog aggregation in time. Meanwhile,event classification model is built based on GRU network,and automatic learning features are used for rumor detection. Experimental results show that the detection accuracy reaches 96.7%,which verifies the effectiveness of this method in rumor detection.