Radical Right Extremism, Online Propaganda and Hybrid Human-Automated Content Removal

该博士将研究在众多社交媒体平台上发现的激进的权利图像数据（包括符号，表情符号，模因和颜色的使用），以帮助刑事司法人员在线预防极端主义。该项目的关键问题是：知识增强问题：激进右翼团体如何使用图像在网上传播宣传？这些信息的关键细微差别是什么？这些群体又是如何表现的呢？政策相关问题：刑事司法系统如何最好地整合人工和自动决策流程，以实现内容删除流程的目的？混合模式在帮助反官员删除内容和采取预防性策略打击网络极端主义方面的效果如何？鉴于社交媒体平台上发布的内容数量庞大，使用技术对于有效删除恐怖主义内容至关重要。同时，自动化决策也有其局限性。特别是，机器处理数据和代码;它们不赋予意义（Hildebrandt 2018）。这些挑战在极右翼的背景下加剧。与所谓的伊斯兰国相关的内容不同，大多数激进的右翼内容都没有品牌。此外，奥弗顿窗口发生了变化，一些强大的参与者-包括国家元首，主要政党，一些传统媒体组织和广泛的西方公众-认同这一内容（康威，2020）。因此，有效性的关键在于改善混合人工自动化决策的能力（货车der Vegt et al，2019）。除了有效性之外，解决与技术管理相关的各种伦理问题也很重要（Brownsword，2016）。其中包括：向用户提供关于哪些内容是允许的和哪些内容是不允许的公正警告，以便他们能够就其对平台的使用做出知情决定;在内容被删除的情况下，确保向用户提供足够的信息，以便他们能够在希望对决定提出上诉时提出上诉;审计自动决策的结果，以检查可能的算法偏差（Macdonald et al，2019）。将使用内容分析方法对该项目收集的数据进行分析。内容分析利用定性和定量方法来批判性地分析音频和视频材料（Finch和Fafinski，2012）。具体的想法，概念，术语，主题和其他图像特征将被确定，并进行比较，以便详细描述，解释和分析的材料。这些类别将通过对数据的仔细阅读而不是预先定义来生成，从而确保归纳方法。从内容分析中生成编码类别将产生编码手册：包含编码器说明的文档，以便数据分析过程是具体的，一致的和可重复的。当编码类别应用于数据时，将创建编码时间表，即包含与样本中每个项目相关的所有调查结果的文件。这将产生一个定量数据集，从中得出结论，并使用定性方法进行介绍。内容分析是一种方法，可以应用于小型和大型数据集，以及定量和定性研究。编码手册将确保研究的可重复性，以及研究结果的可靠性和可验证性。最后，数据驱动的方法确保了所有调查结果和结论的整体性、归纳性和对话性。

This PhD will examine radical right image data (including symbols, emblems, memes and use of colour) found on numerous social media platforms in order to aid criminal justice operatives in their preventative approach to extremism online. The key questions of the project are:Knowledge Enhancing Questions:How do radical right groups use images to spread propaganda online?What are the key nuances of these messages?And how do the groups perform othering?Policy Relevant Questions:How best can human and automated decision-making processes be integrated for the purpose of content removal processes by the criminal justice system?How effective can a hybrid model be for aiding counter-officials in content removal and preventative tactics in combatting online extremism?MethodologyGiven the sheer volume of content posted on social media platforms, the use of technology is essential for effective removal terrorist content. At the same time, automated decision-making has its limitations. In particular, machines work with data and code; they do not attribute meaning (Hildebrandt 2018). These challenges are exacerbated in the context of the far-right. Unlike content associated with the so-called Islamic State, most radical right content is not branded. Moreover, there has been a shift in the Overton Window, such that some powerful actors - including heads of state, major political parties, some traditional media organisations and broad swathes of Western publics - identify with this content (Conway, 2020). The key to effectiveness therefore lies in the ability to improve hybrid human-automated decision-making (van der Vegt et al, 2019).As well as effectiveness, it is also important to address the various ethical issues associated with technological management (Brownsword, 2016). These include: providing users with fair warning of what content is and is not permissible, so that they can make informed decisions about their use of the platform; where content is removed, ensuring that users are provided with sufficient information to be able to appeal against the decision should they wish to do so; and, auditing the outcomes of automated decision-making in order to check for possible algorithmic bias (Macdonald et al, 2019).The data scraped for this project will be analysed using a content analysis methodology. Content analysis utilises both qualitative and quantitative methods to critically analyse audio and visual material (Finch and Fafinski, 2012). Specific ideas, concepts, terms, themes and other image characteristics will be identified, and comparisons made, to allow a detailed description, explanation and analysis of the material. These categories will be generated through a careful reading of the data, as opposed to being pre-defined, thus ensuring an inductive approach. The generation of coding categories from the content analysis will result in a coding manual: a document containing instructions for the coder, so that the process of data analysis is specific, consistent and repeatable. As the coding categories are applied to the data, a coding schedule will be created, i.e. a document containing all the findings related to each item within the sample. This will result in a quantitative dataset, from which conclusions will be drawn and presented using qualitative methods.There are several benefits to using this approach. Content analysis is a methodology that can be applied to both small and large datasets, and to quantitative and qualitative research. The coding manual will ensure the repeatability of the research and that the findings are robust and verifiable. Finally, the data-driven approach ensures the holistic, inductive and dialogic nature of all findings and conclusions.