摘要
针对目前网络谣言鉴别研究,文本学习往往会受到文本读入内容过长导致长距离信息丢失或者是为了捕捉局部信息而依赖于长期输入表示从而影响鉴别结果。通过提出S-LSTM(sentence-state long short term memory networks)算法在保留字词节点信息的同时对句子进行聚合,从而保留句子的局部和全局信息,进而提升网络谣言鉴别的精确性和有效性。与TextGCN、Bi-GCN、Att_BiLSTM等几种深度网络谣言鉴别方法的对比中,该方法在两组模型测试上的准确率分别达到78.87%、90.30%,均取得了不错的效果,在考虑句子全局信息的情况下,其对谣言鉴别效果会有不错的提升。
Aiming at the current research on the identification of online rumors,text learning is often affected by the long-distance information loss due to the long-distance reading of the text or the long-term input representation in order to capture local information,which affects the identification result.This paper proposed the S-LSTM algorithm which used it to aggregate sentences while retaining the word node information,thereby retaining the local and global information of the sentence,thereby improving the accuracy and effectiveness of network rumors identification.In comparison with several deep network rumor identification methods such as TextGCN,Bi-GCN,and Att_BiLSTM,the accuracy of this method on the two sets of model tests reaches 78.87%and 90.30%,respectively,and achieves good results.The result proves that the rumor identification effect can be improved in the case of considering the global information of the sentence.
作者
庞源焜
张宇山
Pang Yuankun;Zhang Yushan(School of Statistics&Mathematics,Guangdong University of Finance&Economic,Guangzhou 510320,China)
出处
《计算机应用研究》
CSCD
北大核心
2022年第4期1064-1070,共7页
Application Research of Computers
基金
国家自然科学基金资助项目(61876207)
广东省基础与应用基础研究基金资助项目(2020A1515011405)。
关键词
谣言鉴别
S-LSTM
图神经网络
文本分类
rumor identification
S-LSTM
graph neural network
text classification
