基于预训练语言模型的早期微博谣言检测  

Early Microblog Rumor Detection Based on Pre-trained Language Model

作　　者：冯茹嘉 张海军[1] 潘伟民[1] FENG Rujia;ZHANG Haijun;PAN Weimin(College of Computer Science and Technology,Xinjiang Normal University,Urumqi 830000)

机构地区：[1]新疆师范大学计算机科学技术学院,乌鲁木齐830000

出　　处：《计算机与数字工程》2023年第5期1075-1080,1184,共7页Computer & Digital Engineering

基　　金：2019年度自治区创新环境(人才、基地)建设专项(人才专项计划--天山雪松计划)“面向高校课堂的多模态数据情感倾向性分析的关键技术研究”(编号:2019XS08);国家自然科学基金-新疆联合基金重点项目“网络谣言检测与舆论引导算法研究”(编号:U1703261)资助。

摘　　要：该文基于微博数据以实现早期谣言检测为目的,挖掘微博内容的深层语义信息。为提高谣言检测效率,该文使用预训练语言模型对部分参数进行了预训练,提高了模型的训练速度,同时解决了一词多义情况的词向量表示问题,优化了深层语义信息的表达能力。结合BiGRU-MHA模型学习微博内容的深层语义信息,最后输出微博事件的分类结果。实验结果表明,XLNet+BiGRU-MHA模型的F1值达到95.5%,在任何时间阶段内均处于领先地位。Based on microblog data,this paper aims to realize early rumor detection and mining the deep semantic information of microblog content.In order to improve the efficiency of rumor detection,a pretrained language model is used to pre-train some parameters and improve the training speed of the model.At the same time,the problem of word vector representation in the case of polysemy is solved,and the expression ability of deep semantic information is optimized.Combined with Bigru-MHA model to learn the deep semantic information of microblog content,and finally output the classification results of microblog events.The experimental results show that the F1 value of XLNET+Bigru-MHA model reaches 95.5%,which is in the leading position in any time period.

关 键 词：谣言检测 预训练模型 深度学习 

分 类 号：O141.4[理学—数学]