Predicting social-media reactions towards refugees by incorporating geo-tagged personality data and distance matrices

在攻读博士学位期间，我打算借鉴两种被广泛使用的概念取向（接触理论和人格理论）来更好地理解偏见的本质。仇外行为最常发生在移民率低的地区。在文献中，这是由群体间接触理论解释的，这种现象表明，缺乏直接或间接的接触与偏见的增加有关（Allport, 2006）。解释偏见的另一种方法是基于人格差异。右翼威权主义（RWA）和社会优势取向（SDO）可以准确预测偏见。RWA是由高度的安全凝聚力驱动的，在开放性低、责任心高的人群中发现，而SDO是在那些认为世界充满竞争、亲和力低的人群中发现的（Sibley & Duckitt, 2009）。我的博士学位将把这两种理论结合成一个模型，使我们能够计算出与外部群体的实际互动相比，内部信念的重要性。社交媒体数据包括参与者的个性和地理信息。Twitter每天有超过5亿条推文，Facebook每分钟有50万条新评论（Edwards, 2016）。相当多的Twitter数据都带有地理标签，也提供了个性洞察。Quercia博士及其同事认为，通过分析Twitter用户的社交媒体行为，可以有效预测超过88%的外向性方差（Quercia, Kosinski, Stillwell & Crowcroft, 2011）。我决定在德国进行我的主要分析，因为最近德国庇护政治的范式转变导致了明确定位的难民中心的开放，仅在2017年就涌入了170万难民（联邦移民和难民办公室，2018）。此外，我在德国长大，这让我可以用德语查阅原始文献，并与相关政党进行沟通。其基本思想是检查有偏见的推文作者与下一个难民中心之间的距离是否可以预测推文的情绪。此外，通过遵循Quercia及其同事（Quercia et al., 2011）概述的程序，将被动地推断个性。我将以自动方式整理Twitter个人资料和以难民和相关术语为特色的帖子。如此大的样本量限制了人工分类，因此将使用语义分析将tweet分类为积极或消极。为了计算难民中心和推文作者所在位置之间的距离，将创建一个距离矩阵。一般的假设是，住在离难民中心更近的人与难民的接触率会更高，因此在推特上对难民的负面评价也会比以前少。Rentfrow博士可以通过大规模的调查来收集个性信息，并通过分析Twitter作者的个人资料来被动地评估个性特征。这两组信息将结合在一个统一的模型中，预测一条推文是积极的还是消极的可能性。距离矩阵，以及大五人格变量，将被包括作为解释变量。由此产生的模型将是第一个将群体间接触理论与意识形态双重过程模型结合起来的模型，这是解释偏见的两个最著名的理论。一个综合模型提高了我们对这两种理论的理解，并使我们能够计算出与外部群体的实际互动相比，内部信念的重要性。此外，这个创新的模型使我们能够检查相互作用。这个新模型为其他假说铺平了道路。例如，接触通常会减少偏见，但对于开放性高的人和开放性低的人来说，这可能是不同的。在这种情况下，较低的开放性很可能与减少接触对降低压力的影响有关

During my proposed PhD, I plan to draw from two widely-used conceptual orientations (contact theory and personality theory) to better understand the nature of prejudice. Xenophobic actions are most often observed in areas with low immigrant rates. In the literature this is explained by Intergroup Contact Theory, the phenomenon by which a lack of contact - direct or indirect - is associated with increased prejudice (Allport, 2006).Another approach in explaining prejudice is based on personality differences. Prejudice can be accurately predicted by Right-Wing Authoritarianism (RWA) and Social Dominance Orientation (SDO). RWA is motivated by a heightened security-cohesion and found in people with low openness and high conscientiousness, whereas SDO is found in people who view the world as competitive and have low agreeableness (Sibley & Duckitt, 2009).My PhD will combine both theories into one model, allowing us to calculate the importance of internal beliefs compared to actual interactions with the outgroup.Social media data includes personality as well as geographic information about participants. Twitter hosts over 500 million tweets a day, and Facebook has half a million new comments every minute (Edwards, 2016). A fair amount of Twitter data is geo-tagged, and offers personality insights, too. According to Dr. Quercia and colleagues, e.g. over 88% of extraversion variance can be effectively predicted by analyzing Twitter users' social media behavior (Quercia, Kosinski, Stillwell & Crowcroft, 2011).I decided to do my primary analysis in Germany, as a recent paradigm shift in its asylum politics led to an opening of clearly-located refugee centers and an influx of 1.7m refugees in 2017 alone (Federal Office for Migration and Refugees, 2018). Additionally, I grew up in Germany, allowing me to review primary literature in German and communicate with the involved political parties.The basic idea is to check if distance between authors of biased tweets and the next refugee center is predictive of the sentiment of the tweets. Additionally, personality will be passively inferred by following the procedure outlined by Quercia and colleagues (Quercia et al., 2011)I will collate Twitter profiles and posts featuring refugees and relevant terms in an automated manner. This large sample size limits manual categorization, and therefore semantic analysis will be used to classify tweets as positive or negative. A distance matrix will be created in order to calculate the distance between the refugee centers and the location of the tweet authors. The general hypothesis is that people living closer to refugee centers will have had contact with refugees at a higher rate, and thus tweet less negatively about refugees than previously.Personality information will be collected by integrating large-scale surveys available to Dr. Rentfrow with passively assessing personality profiles by analyzing the Twitter author's profiles.Both sets of information will be combined in a unified model, predicting the likelihood of a tweet being positive versus negative. The distance matrix, as well as the big-5 personality variables, will be included as explanatory variables. The resulting model will be the first to combine Intergroup Contact Theory with the Dual Process Model of Ideology - the two most prominent theories explaining prejudice.A combined model improves our understanding of both theories and allows us to calculate the importance of internal beliefs compared to actual interactions with the outgroup. Furthermore, this innovative model enables us to check for interactions. This new model paves the way for alternative hypotheses. For instance, contact in general reduces prejudice but this may be different for people with high versus low openness. In this scenario, low openness will likely be associated with reduced chances of the contact having an effect on lowering prej