Abstract:

With the evolution of the era of big data, rumors in the Internet have sprung up. In current rumor identification methods, supervised learning models require a large amount of labeled data during training, and the manual labeling of network rumors takes a long time. Therefore, this paper proposes a semi-supervised graph convolutional neural network, which can effectively use unlabeled data. The model trains on labeled nodes, updates the weight matrix shared by all nodes, and propagates the information of labeled nodes to unlabeled nodes. At the same time, it solves the problem of weak generalization of the supervised learning model and the instability of the unsupervised learning model. Compared with three network rumor identification methods based on SVM algorithm, logistic regression algorithm and BiLSTM model, respectively, the method in this paper reaches 86.1% and 85.3% in the two evaluation indicators of recall rate and F1 value, which further improves the accuracy and stability of network rumor identification. The method can effectively reduce the cost of manual tagging, identify rumors in social media and online news, and provide new ideas for the management of network rumors.

Key words: rumor identification, semi-supervised learning, graph convolutional neural network

摘要：

随着大数据时代的演进，互联网中的谣言成井喷状涌现。目前网络谣言鉴别方法中，基于监督学习的模型在训练过程中需要大量标注数据，同时网络谣言的人工标注用时较长，故提出采用半监督学习的图卷积神经网络，可有效利用无标注数据。通过在有标注节点上训练模型，更新所有节点共享的权重矩阵，将有标注节点信息传播给无标注节点，同时解决监督学习模型泛化能力不强和无监督学习模型不稳定的问题。与基于SVM算法、逻辑回归算法和BiLSTM模型的三种网络谣言鉴别方法相比，该方法在召回率、F1值两个评价指标上分别达到86.1%、85.3%，进一步提升了网络谣言鉴别的准确性和稳定性。该方法可有效减少人工标注代价，鉴别社交媒体和网络新闻中的谣言，为网络谣言的治理提供新思路。

关键词: 谣言鉴别, 半监督学习, 图卷积神经网络