基于Encoder-Decoder-Detector的多任务谣言检测模型

A multi-task rumor detection model based on Encoder-Decoder-Detector

随着社交媒体的迅速发展,互联网已成为人们获取信息的主要平台。它尽管给人们带来了不少便利,但却也带来了谣言泛滥的问题。近年来,研究人员致力于应对这一挑战,采用从特征工程到深度学习的各种方法。然而,现有工作中并未充分利用谣言本身的特征结构。文章提出了一种Encoder-Decoder-Detector(ED 2)多任务谣言检测模型,运用Encoder模块充分提取文本和用户特征。该模型采用GCN和LSTM模型对原文和评论文本进行编码,任务一是Decoder模块恢复谣言传播特征,任务二是利用Detector模块进行解析和判断谣言是否属实。在公开数据集上,ED 2模型实现了最佳表现。

With the rapid development of social media,the Internet has become the primary platform for people to access information.While it has brought great convenience to people’s lives,it has also brought the problem of rampant rumors.Research has been devoted to addressing this challenge,using various methods ranging from feature engineering to deep learning.However,current work has not fully utilized the structural features of rumors themselves.This paper proposes an Encoder-Decoder-Detector(ED 2)multitask rumor detection model that utilizes the Encoder module to extract text and user features fully.The GCN and LSTM models are employed to encode the original text and the comment text,respectively.Task one involves using the Dencoder module to restore the rumor propagation characteristics,and combining it with the Detector module to analyze and judge the veracity of the rumors.On publicly available datasets,the ED 2 model achieved the best performance.