Raven: to Locate and Identify Online Extremist Multimedia

恐怖分子和仇恨团体利用在线社交网络进行宣传、招募人员和筹集资金，这是一个公认的国家安全问题。我们提出了一种名为Raven的自动智能网络爬虫系统来定位和识别网络上的极端主义多媒体。Raven系统由两个主要组件组成：Muninn，一个专门的网络爬虫，和Huginn，一个机器学习层。Muninn是对现有网络爬虫的改进，该爬虫是在EPSRC之前的一个项目中开发的，用于查找和下载恐怖主义视频并安全地存储它们。Huginn是系统的智能部分，也是拟建项目的主要创新点。使用深度学习技术并使用Python编写，它被训练以识别和分析极端主义多媒体，并将Muninn爬行组件驱动到新网站和Twitter/Telegram频道。虽然现有的爬虫通常会分析网络上的极端主义文本，但多媒体更具挑战性。Huginn最初将使用现有的视频数据集进行训练，并不断使用Muninn下载的其他视频进行更新。随着系统看到越来越多的极端主义多媒体，Muninn了解到更多的来源，Huginn变得更聪明，更准确。穆宁和休金共同努力，帮助公司、分析人士、执法部门、情报机构和研究人员更快、更大规模地发现并摧毁极端主义的多媒体，以应对敏捷的恐怖分子和仇恨团体。

Exploitation of online social networks by terrorists and hate groups for propaganda, recruiting, and fund raising is a widely recognised national security problem. We propose an automated intelligent web crawling system called Raven to locate and identify online extremist multimedia. The Raven system consists of two main components: Muninn, a specialised web crawler, and Huginn, a machine learning layer. Muninn is an enhancement of an existing web crawler that was developed in a previous EPSRC project to find and download terrorist videos and store them securely. Huginn is the intelligent part of the system and the main innovation in the proposed project. Using deep learning techniques and written in Python, it is trained to recognise and analyse extremist multimedia and drive the Muninn crawling component to new web sites and Twitter/Telegram channels. While existing crawlers routinely analyse online extremist text, multimedia is much more challenging. Huginn will be initially trained with an existing video dataset and continually updated with additional videos that Muninn downloads. As the system sees more extremist multimedia, Muninn learns about more sources and Huginn becomes smarter and more accurate. Muninn and Huginn work together to aid companies, analysts, law enforcement, intelligence agencies, and researchers to find and take down extremist multimedia on a faster and larger scale, in the face of agile terrorists and hate groups.