Manipulation-resistant Consensus Formation

研究关键词：在过去的十年中，聚合民意调查作为预测选举结果的一种手段已经失败和批评，特别是在英国和美国，特别是考虑到假新闻和外部阴谋的盛行。虽然充满了人口统计数据，这个指标是完全sans社会拓扑结构，也就是说，它没有兴趣的作用，社会关系和连接在影响个人。因此，我们问如何做社交网络影响意见的传播？当社区在这些网络中乱丢垃圾，并成为潜在的回声室时，会发生什么？为了防止各种形式的操纵，并能更好地通知代理，存在什么图结构和协议？Stewart et al（2019）展示了机器人的使用如何帮助少数党派/观点获得牵引力，我们想问网络的哪些先决条件使这些机器人无效？一个不结盟的政党利用这些机器人的成本效益如何？真实的代理人能意识到他们的存在并集体制定战略来对抗他们吗？将社交网络与代理人的地理空间分布联系起来，隔离如何影响意见？正如谢林（Schelling，1971）所指出的那样，观点可以导致种族隔离的出现，但现在观点动态与这种种族隔离之间的相互作用是什么？舆论动力学是否还有其他涌现的特性？我们希望验证我们的模型，例如使用麻省理工学院选举实验室（MIT Election Lab）关于过去美国选举的数据，以及TWS合作伙伴咨询的社交媒体平台的数据。研究的背景-最近的政治选举-主要在英国和美国，但越来越多地在其他许多国家-受到大量假新闻和不民主的压力，特别是在社交媒体上。机器人和假冒代理人充斥着数字空间，而不公正划分选区可以被视为对网络边缘的操纵。同样，社交媒体上的意见形成和传播也是一个非常普遍的问题，也会影响真实的选举，了解在线社区和回声室的作用也会激发研究。近年来，Facebook等大型科技公司因未能缓和和审查有害/攻击性/虚假信息和团体而受到严厉批评，因此有一个可以识别这种两极分化的社交网络的指标将有助于立法者和科技公司解决这些问题。研究的目的和目标-提供额外的工具，度量和方法，以了解如何操纵意见及其形成。为了产生抗操纵和/或为代理提供更多信息环境的图。研究方法的新奇-从动态和系统的角度分析选举-与使用静态和局部测量的现状不同-考虑可以高度极化的社区结构。与当前的中心性测量不同-节点/代理重要性的量化器-我们提出的指标将明确考虑意见/党时，排名节点的基础上，他们的影响力。我们的目标也产生图生成模型，耦合的拓扑结构的图和意见的节点，而在以前的模型中，图的形成已经完全脱离意见。潜在的影响，应用，和利益-减少受诽谤的政党/个人的不民主影响。研究;数字经济，工程，全球不确定性，ICT [信息和通信技术]，数学科学外部合作伙伴- TWS Partners

Research Keywords: opinion diffusion and collective decision making, social network analysis, strategic interaction, computational social scienceIn the last decade aggregate polling as a means to predict election outcomes has had its share of failures and critiques, in particular in the UK and in the US, especially given the prevalence of fake news and external machinations. While full of demographic data, this metric is completely sans social topology, that is it has no interest in the role of social ties and connections in influencing individuals.We thus ask how do social networks impact the diffusion of opinions? What happens when communities litter these networks, and are potentially echo chambers? What graph structures and protocols exist in order to prevent various forms of manipulation and that can better inform agents?Stewart et al (2019) showed how the use of bots can help even minority parties/opinions to gain traction, we want to ask what preconditions of the networks make these bots ineffective? How cost-effective and what best way can a mal-aligned party utilise these bots? Can real agents realise their presence and collectively strategise to counter them?Connecting the social network to the geospatial distribution of agents, how does segregation impact opinions? Opinions can cause the emergence of segregation, as in Schelling (1971), but what is now the interplay between opinion dynamics and this segregation? Are there other emergent properties of the opinion dynamics?We wish to validate our models for example using data from the MIT Election Lab on past US elections as well as from the social media platforms the TWS Partners consult on.The context of the research - Recent political elections - primarily in the UK and the US, but increasingly across many other countries - which have been subjected to masses of fake news and un-democratic pressures, especially in social media. Bots and fake agents litter the digital space while gerrymandering can be seen as the manipulation of network edges.Equally opinion formation and diffusion on social media is an extremely prevalent issue that does also impact real elections, understanding the role of online communities and echo chambers also motivates the research. In recent years large tech companies such as Facebook have been under heavy criticism for failing to moderate and censor harmful/offensive/fake information and groups thus having a metric that can identify such polarised social networks would help law-makers and tech companies address such issues.The aims and objectives of the research - To provide additional tools, metrics and methodologies to understand how opinions and the formations thereof are manipulated. To produce graphs that are manipulation-resistant and/or provide a more informative environment for agents.The novelty of the research methodology - Analyse elections from a dynamics and systems perspective - unlike the status quo that uses static and local measures - that considers community structures that can be highly polarised.Unlike current centrality measures - quantifiers of a node/agent's importance - our proposed metrics would explicitly consider opinions/parties when ranking nodes based on their influence.We aim also to produce graph generation models that couples the topology of the graph and the opinions of its nodes, whereas in previous models the graph formation have been entirely divorced from opinions.The potential impact, applications, and benefits - Reducing undemocratic effects of maligned parties/individuals. Identify important agents and structures of agents that most influence opinions.Research;Digital economy, Engineering,Global uncertainties, ICT [Information and Communication Technologies], Mathematical SciencesExternal Partner - TWS Partners