【作者】 杨洁；

【导师】 刘维；

【作者基本信息】 扬州大学 ， 计算机科学与技术， 2024， 硕士

【摘要】 随着信息的快速传播和社交媒体的普及,虚假信息和谣言在网络空间中蔓延,谣言检测技术可以有效识别和阻止虚假信息的传播,维护信息的真实性和可信度。然而,现有的谣言检测方法存在谣言模态单一、谣言准确率不高等问题。因此,本文以用户发布的帖子为谣言对象,分析帖子的文本信息、视觉信息以及情感信息,构建了多种谣言检测模型。主要研究内容包括以下几个方面:(1)提出了一种基于情感信息和共注意力机制的谣言检测方法。现有方法大多关注谣言的文本信息和传播结构信息,而忽略了情感信息的重要性。此外,现有方法简单地将谣言各模态之间的特征进行级联,未能充分挖掘谣言各模态之间的关系。因此,本文提出了一种基于情感信息和共注意力机制的谣言检测方法(SCNN),该方法首先关注了谣言的文本信息,在此基础上增加了情感信息作为谣言内容的补充。方法还考虑了谣言对应的视觉特征。此外,该方法设计了一个共注意力机制模块,有效地将谣言的多模态特征进行融合。实验结果表明,提取谣言的情感信息作为谣言内容的补充、充分挖掘谣言各模态特征之间的潜在关系有助于提升谣言检测效果。(2)提出了一种基于图像多特征和帖子情感信息的谣言检测方法。现有方法在提取谣言图像特征时大多只关注图像的像素域特征,忽略了频域特征的重要性。图像的频域特征的分析有助于辨别谣言图片是否经历了频繁的压缩以及图像是否被篡改。因此本文提出了一种基于图像多特征和帖子情感信息的谣言检测方法(MVSNN)。该方法首先提取了谣言的文本特征、情感特征以及像素域特征,同时考虑到图像频域特征的分析有助于辨别谣言图片是否经历了频繁的压缩以及篡改,方法设计了一个卷积神经网络提取了图像的频域特征。真实数据集上进行实验,实验结果表明谣言检测的准确率有了进一步的提升。(3)设计并实现了一个基于情感分析的谣言检测系统。该系统提供新闻资讯推送、新闻情感分析、谣言检测结果的可视化、用户管理等功能。该系统经过测试,界面友好,运行稳定,在各种场景下都可以发挥重要作用,有助于维护信息的真实性和可信度,保护公众利益和社会秩序。更多还原

【Abstract】 With the rapid spread of information and the popularity of social media,false information and rumors spread in cyberspace.Rumor detection technology can effectively identify and prevent the spread of false information and maintain the authenticity and credibility of information.However,existing rumor detection methods have problems such as single rumor mode and low rumor accuracy.Therefore,this article takes user posts as rumor objects,analyzes the text information,visual information and emotional information of the posts,and builds a variety of rumor detection models.The main research contents include the following aspects:(1)A rumor detection method based on emotional information and co-attention mechanism is proposed.Most existing methods focus on the textual information and propagation structure information of rumors,while ignoring the importance of emotional information.In addition,existing methods simply cascade the features between rumor modalities and fail to fully explore the relationship between rumor modalities.Therefore,this paper proposes a rumor detection method(SCNN)based on emotional information and co-attention mechanism.This method first pays attention to the text information of the rumor,and then adds emotional information as a supplement to the rumor content.The method also takes into account the visual features corresponding to the rumors.In addition,this method designs a co-attention mechanism module to effectively fuse the multi-modal features of rumors.Experimental results show that extracting the emotional information of rumors as a supplement to the rumor content and fully exploring the potential relationships between the modal features of rumors can help improve the rumor detection effect.(2)A rumor detection method based on image multi-features and post emotional information is proposed.When extracting rumor image features,most existing methods only focus on the pixel domain features of the image,ignoring the importance of frequency domain features.Analysis of frequency domain features of images can help identify whether rumored images have undergone frequent compression and whether the image has been tampered with.Therefore,this paper proposes a rumor detection method(MVSNN)based on multiple features of images and emotional information of posts.This method first extracts text features,emotional features and pixel domain features of rumors.At the same time,considering that the analysis of image frequency domain features can help identify whether rumor pictures have experienced frequent compression and tampering,the method designs a convolutional neural network extraction The frequency domain characteristics of the image.Experiments were conducted on real data sets,and the experimental results showed that the accuracy of rumor detection has been further improved.(3)A rumor detection system based on sentiment analysis is designed and implemented.The system can provide users with functions such as news consultation push,news sentiment analysis,visualization of rumor detection results,and user management.The system has been tested and found to be user-friendly and stable in operation.It can easily analyze the results of rumor detection and has important social value and promotion significance in public opinion detection and handling of emergencies.更多还原