Rumour Veracity Assessment in Social Media

近年来，作为快速信息传播渠道的社交媒体越来越多。然而，社交媒体也成为病毒式恶作剧和在线虚假信息的温床。因此，对于记者、紧急服务和政府机构来说，能够在接近真实的时间内检测快速传播的错误/虚假信息变得越来越重要。这引起了越来越多的研究自动错误/虚假信息检测，事实检查和内容验证。然而，目前的准确度水平达不到实际采用所需的准确度，因为训练数据很小，并且方法性能往往会在看不见的错误/虚假信息上显着降低。因此，本研究项目将通过整合来自多个社会来源的信息，并建立一个能够进行交叉验证的知识网络，对监督和非监督的真实性评估方法进行新的研究。这一高层次的目标产生了以下目标：RO 1：开发一种非监督的特征提取方法，用于识别有效的谣言真实性分类的判别特征; RO 2：通过整合来自外部来源的证据，通过跨媒体分析和交叉链接，开发基于图形神经网络的监督方法; RO 3：实施分布式概率推理工具，以分析社交媒体中虚假/错误信息的动态传播模式; RO 4：在基准数据集中使用新方法。

Recent years have witnessed a growing proliferation of social media that serve as channels for rapid information dissemination. However, social media have also become a hotbed of viral hoaxes and online disinformation. It thus becomes increasingly important for journalists, emergency services and government agencies to be able to detect in near real time fast-spreading mis/disinformation. This has given rise to a growing body of research on automatic mis/disinformation detection, fact checking and content verification. Present accuracy levels, however, fall short of the accuracy required for practical adoption as training data is small and method performance tends to degrade significantly on unseen mis/disinformation. This research project will thus carry out novel research on supervised and unsupervised methods for veracity assessment by integrating information from multiple social sources and building a knowledge network that enables cross verification.This high-level aim gives rise to the following objectives:RO1: Develop an unsupervised feature extraction approach for identifying discriminative features for effective rumour veracity classification;RO2: Develop supervised approaches built on graphical neural networks by incorporating evidence from external sources, through cross-media analysis and cross-linking;RO3: Implement distributed probabilistic inference tools to analyse dynamic spreading-patterns of dis/misinformation in social media;RO4: Validate the new methods in benchmarking datasets.