Socio-cognitive and group-based processes in adherence to online misinformation

我的研究将利用计算社会科学的方法，主要由认知和社会心理学理论驱动，但将这些见解应用于使用大规模计算分析的研究背景中。偶尔，我也打算使用更传统的、实验性的社会科学研究方法来分析需要更精细粒度和数据质量的概念。目前，错误信息是一个被大量研究的话题。由于其广泛性，我的论文将侧重于态度和基于人口的两极分化在放大错误信息中的作用。我将从个人信息处理类型（社会认知）的角度来研究这个问题，同时也从用户的社会身份和群体成员身份如何影响他们在网上与之互动的人和信息（基于群体）的角度来研究这个问题。我的研究将利用不同的子项目，其中两个目前正在进行中。第一个项目旨在解决围绕在线回声室和敌对交叉（群体间）互动之间相互作用的研究差距。使用来自Reddit平台的数据，我将首先基于共享的用户基础将平台上类似的论坛（称为subreddits）聚集在一起。我会将此限制在政治子reddit，以确保集群之间的域相关性和相互排他性。使用子reddits之间的相似性度量，我将能够检测不同的社区（例如右倾子reddits，左倾子reddits，中立子reddits等）。转到用户层面，我将检测每个用户的“家庭”社区（基于他们发帖最多的地方），并得出他们互动倾向的分布指标——例如，有些用户可能100%的时间在他们的家庭社区发帖，而其他人可能只在那里发帖85%的时间（在其他社区发帖15%的时间）。我将利用自然语言处理（NLP）衍生的指标来评估用户帖子的属性，例如毒性和身份攻击，并为每个用户汇总这些指标。获得这些数据后，我将能够回归NLP指标（在家庭社区和非家庭社区之间划分）的用户交互趋势。我们观察到的结果模式将是量化回音室和更广泛的互动趋势之间相互作用的第一次尝试，它将为未来的阈值研究提供信息，这些阈值可能需要在社交媒体上的群体间互动变得无敌意之前达到。第二个正在进行的项目旨在捕捉政治社会身份在坚持错误信息中的作用。这将是一个比第一个样本更小但粒度更高的研究，遵循社会实验的方法。在三项研究中，我将1)操纵对个人的不同程度的身份攻击，2)操纵分享事实核查的账户的政治身份，以及3)操纵促进信息处理分析思维的线索的社会性（即是否没有线索，个人层面的线索，还是群体层面的线索）。对于每一种操作，我将观察它们在处理风格（分析或直觉）上的作用，以及随后在真实或虚假信息的可信度上的作用。这个项目与第一个直接相关，因为我在这里观察到的结果将允许我进一步定位第一个研究的敌意和身份攻击模式。

My research will utilize methods from computational social science, being driven predominantly by cognitive and social psychological theory but applying these insights in a research context using large-scale computational analyses. Occasionally, I also aim to use more traditional, experimental social science research methods to analyse concepts which require a finer granularity and quality of data.Misinformation is a heavily researched topic at the moment. Due to its broadness, my thesis will focus on the role of attitude and demographic-based polarization in amplifying misinformation. I will examine this both from the lens of individual information processing types (socio-cognitive), but also from the perspective of how users' social identities and group memberships can affect the people and information they interact with online (group-based).My research will utilize different sub-projects, two of which are currently in progress. The first project aims to address a research gap around the interplay between online echo-chambers and hostile cross-cutting (intergroup) interactions. Using data from the Reddit platform, I will first cluster together similar forums (called subreddits) on the platform based on shared user-base. I will restrict this to political subreddits to ensure domain relevance and mutual exclusivity between clusters. Using similarity metrics between the subreddits, I will be able to detect distinct communities (for example groups of right-leaning subreddits, left-leaning subreddits, neutral subreddits, etc). Moving to the user level, I will then detect each user's "home" community (based on where they tend to post the most), and derive distribution metrics for their interaction tendencies - for example, some users may post on their home community 100% of the time, while others may only post there 85% of the time (and 15% of the time on other communities). I will utilize Natural Language Processing (NLP)-derived metrics to assess the attributes of user posts, such as toxicity and identity attack, and aggregate these metrics for each user. Having derived this data, I will then be able to regress the NLP metrics (split across home communities and non-home communities) on the user interaction tendencies. The resulting pattern we observe will be the first attempt at quantifying the interplay between echo chambers and wider interaction tendencies, and it will inform future research on thresholds which may need to be met before intergroup interactions become non-hostile on social media.The second ongoing project aims to capture the role of political social identities in adherence to misinformation. It will be a smaller-sample but higher-granularity research than the first, following a social experimental approach. Across three studies, I will 1) manipulate different levels of identity attack on individuals, 2) manipulate the political identity of accounts which share fact-checks to misinformation, and 3) manipulate the sociality of cues which promote analytical thinking in processing information (i.e. whether there is no cue, an individual-level cue, or a group-level cue). For each of these manipulations, I will observe their role on processing style (analytical or intuitive) and subsequently on believability in true or false information. This project is directly relevant to the first, as the results I observe here will allow me to further situate the hostility and identity attack patterns of the first study.