Understanding the role of social media in promoting anti-migration sentiment and hate crime

该提案旨在利用推特、全球定位系统数据、机器学习、网络科学、纵向和因果推理方法，调查社交媒体在影响联合王国仇恨犯罪方面的作用。具体而言，该项目旨在了解社交媒体在影响反移民情绪，随后的暴力行为中的作用，以及这些模式如何与种族社区的时间接触有关。它有三个目标：1。识别Twitter反移民社区，并检查它们出现的数字和地理背景; 2.评估和纠正Twitter数据中的代表性偏见; 3.估计反移民Twitter内容对仇恨犯罪的因果影响。该项目将增进我们对在线反移民社区的规模和结构的了解，并提供证据说明在线内容如何能够加强和传播仇外情绪，导致仇恨行动。这些证据将有助于通过利用Rowe博士（主要主管）与联合国和世界银行之间正在进行的合作来设计反歧视的政策方案。在方法上，该项目将创新培训和部署机器学习模型，以确定反对和支持移民的社区，并将使用移动的电话数据来创建一个与时间有关的少数民族社区接触程度的衡量标准。背景在线社交媒体平台（OSMP）在塑造我们的社会方面发挥着关键作用。它已成为一个主要的通信渠道，使社会联系在世界各地的遥远位置[2]。它帮助企业扩大其地理范围，并发起大规模的营销活动，以促进他们的产品，增加销售和收入[3]。与此同时，OSMP一直受到严格审查，特别是在英国脱欧公投[4]和当前的COVID-19大流行[5]期间。OSMPs促成了新的社会过程，特别是大规模的错误信息，机器人农业，数字回音室[5]，影响了我们在数字和物理世界中的行为[6]。网络仇恨言论一直是激烈和两极分化的辩论的核心[7]。尽管公众日益关注并呼吁采取政策行动，但几乎没有经验证据表明仇恨的在线内容如何转化为现实生活中的行为。与此同时，移民一直被认为是全球最具分裂性的社会问题之一[8]。关于移民的种族主义和仇外内容在社交媒体上很突出。关于OSMPs的仇外叙述有助于形成移民政策和政治结果[4]。这种叙事传播仇恨情绪，导致社会更加两极分化[9]，这可能会蔓延到身体暴力[1]。虽然先前的研究使用OSMP数据来识别和消除反移民叙事[10]，但对它们出现的地理和数字背景知之甚少。例如，我们对接触不同城市环境的时间与当地反移民数字内容模式的关系知之甚少;这些模式如何在社区人口和社会经济特征之间变化;以及回声室在多大程度上导致社交媒体上反移民社区的演变。为了弥补这些差距，该项目力求利用推特和普惠制移动的电话数据，查明网上反移民社区及其出现的数字和地理背景;解决推特数据中的代表性偏见;评估推特上反移民内容对仇恨犯罪发生的影响。方法该项目将分为三个阶段（S），如图1所示，映射到项目目标。四个关键的数据来源将使用Twitter，GPS Huq，人口普查和仇恨犯罪数据。考虑到道德问题，数据将存储在受密码保护的本地服务器上，在安全的房间和

Aim and objectives This proposal aims to investigate the role of social media in influencing hate crime in the United Kingdom using Twitter, GPS data, machine learning, network science, longitudinal and causal inference approaches. Specifically, the project seeks to understand the role of social media in influencing anti-immigration sentiment, subsequent acts of violence, and how these patterns may relate to time exposure to ethnic communities. It has three objectives: 1. Identify Twitter anti-migration communities, and examine the digital and geographic context within which they emerge; 2. Assess and correct representativeness biases in Twitter data; and 3. Estimate the causal impact of anti-migration Twitter content on hate crime. The project will advance our understanding of the size and structure of online anti-migration communities and evidence on how online content can reinforce and spread xenophobic sentiment leading to hateful actions. Such evidence will help to design policy programmes to counter discrimination by leveraging on ongoing collaborations between Dr. Rowe (primary supervisor) and the United Nations and the World Bank [1]. Methodologically, the project will innovate training and deploying a machine learning model to identify anti- and pro-migration communities and will use mobile phone data to create a time-dependent measure of exposure to ethnic communities. Background Online social media platforms (OSMPs) play a pivotal role in shaping our society. It has become a main communication channel enabling social connections over distant locations across the world [2]. It has helped businesses to expand their geographical reach and launch mass scale marketing campaigns to promote their products, increasing sales and revenue [3]. At the same time, OSMPs have been under intense scrutiny, particularly during the Brexit referendum [4] and current COVID-19 pandemic [5]. OSMPs have enabled the engendering of new social processes, notably mass scale misinformation, bot farming, digital echo chambers [5], influencing our behaviours in the digital and physical world [6]. Online hate speech has been at the core of intense and polarised debate [7]. Despite growing public concern and calls for policy action, there is little empirical evidence on the ways in which hateful online content translates into real-life behaviour. At the same time, immigration is consistently identified as one of the most divisive social issues globally [8]. Racist and xenophobic content on immigration is prominent on social media. Xenophobic narratives on OSMPs have contributed to shaping migration policy and political outcomes [4]. Such narratives spread sentiments of hate, leading to more polarised societies [9] which can spill onto physical violence [1]. While prior research has used OSMP data to identify and characterise anti-immigrant narratives [10], less is known about the geographical and digital context within which they emerge. For instance, we know little about how time exposure to diverse urban environments relate to local patterns of digital anti-immigration content; how these patterns may vary across neighbourhood demographic and socio-economic features; and the extent to which echo chambers lead to the evolution of anti-migration communities on social media. To address these gaps, the project seeks to leverage on Twitter and GSP mobile phone data to identify online anti-immigration communities, and the digital and geographical contexts within which they occur; address representativeness biases in Twitter data; and assess the influence of anti-migration Twitter content on the occurrence of hate crime. Methodology The project will be divided into three stages (Ss) as outlined in Fig.1 mapping to the project objectives. Four key data sources will be used Twitter, GPS Huq, Census and hate crime data. Given ethical concerns, the data will be stored on password protected local server, accessed in a secure room and a