IBSS: The Spread and Impact of Moral Messages: Machine Learning, Network Evolution, and Behavioral Prediction

在2013年波士顿马拉松悲剧发生后不久，社交媒体上就出现了成千上万的亲社会行为，比如转发献血网站的链接，提供如何与亲人联系的信息，甚至提出为有需要的人提供食物和住所。这些行为远非孤立的行为，而是发生在社会网络中，在共同的同情和团结的道德信息中。这个跨学科的研究项目将研究人们如何应对公共危机，以及道德反应如何在社会网络中塑造这些反应。该项目将为道德心理学、网络社会学、计算机科学和其他领域提供新的理论见解和方法进步。它将加强对道德关注如何通过社会网络传播的理解，并探索处理人类道德决策和群体动力学的新理论框架。这些理论框架将指导人工智能技术的发展，用于建立道德的描述性模型，而情感分析和机器学习的新方法将用于评估网络中道德问题和社会影响的理论模型。通过审查影响道德信息传播和参与慈善捐赠等亲社会活动的因素，该项目可能有助于增加紧急情况下个人的福祉。该项目还将促进未来对社交媒体的公共性和持久性如何为理解和预测社会变化提供新途径的研究。跨学科的道德科学已经开发出了有效的衡量道德问题的方法，使用了许多不同的方法，如道德基础理论和施瓦茨的价值观循环。然而，这一领域的实证研究通常是在行为发生后通过问卷调查收集信息来评估道德判断。社会学在行为发生时做了更多的评估，但使用的方法更有限。计算机科学的最新创新提供了新的方法来收集有关自然环境中道德判断结构和大规模行为的信息，以及两者之间的关系。调查人员将采用这些新的基于计算机的方法来检查来自社交媒体的文本，以便在不依赖预设问卷的情况下检查道德关注和价值观的结构。他们将调查道德信息和行为传播的网络动态，他们将确定社交媒体中的道德内容如何在个人和社会尺度上预测现实世界的行为。研究人员将把机器学习和情感分析技术与道德认知和社会动态理论结合起来。他们将追求的问题包括，日常道德判断（在没有研究者提示的情况下做出的）与主流心理学道德理论的契合程度，以及是否有可能模拟和预测道德影响如何导致随后的亲社会或反社会行为。本项目由美国国家科学基金会跨学科行为与社会科学研究（IBSS）竞赛资助。

In the immediate aftermath of the 2013 Boston Marathon tragedy, hundreds of thousands of prosocial acts were evident on social media, such as reposted links for blood donation sites, information regarding how to get in touch with loved ones, and even offers to provide food and shelter for those in need. Far from isolated acts, these behaviors occurred within social networks, amid shared moral messages of empathy and solidarity. This interdisciplinary research project will examine how people respond to public crises and how moral reactions shape these responses in social networks. The project will contribute new theoretical insights and methodological advances in moral psychology, network sociology, computer science, and other fields. It will enhance understanding of how moral concerns spread through social networks and explore new theoretical frameworks dealing with human moral decision making and group dynamics. These theoretical frameworks will guide the development of artificial intelligence techniques for building descriptive models of morality, and the new methods of sentiment analysis and machine learning will be used to assess theoretical models of moral concerns and social influence in networks. By examining factors influencing the spread of moral messages and participation in prosocial activities, such as charitable giving, the project may help increase the well-being of individuals in emergency situations. The project also will facilitate future inquiry into how the public and persistent nature of social media may provide new ways to understand and forecast social change.The interdisciplinary science of morality has developed well-validated measures of moral concerns using a number of different approaches, such as Moral Foundations Theory and Schwartz's Values Circumplex. Empirical research in this field usually has assessed moral judgments via questionnaires gathering information well after actions have occurred, however. Sociology has done more to assess behavior as it occurs but has used even more limited measures. Recent innovations in computer science offer new ways to gather information about the structure of moral judgments and large-scale behavior in natural settings as well as the relationships between the two. The investigators will employ these new computer-based methods to examine texts from social media in order to examine the structure of moral concerns and values without relying on preset questionnaires. They will investigate the network dynamics of the spread of moral messages and behaviors, and they will determine how moral content in social media can predict real-world behavior at both individual and societal scales. The investigators will couple machine learning and sentiment analysis techniques with theories about moral cognition and social dynamics. Among questions they will pursue are how well everyday moral judgments (made without researcher prompting) correspond with dominant psychological theories of morality and whether it is possible to model and predict how moral influence can lead to subsequent prosocial or antisocial behavior. This project is supported through the NSF Interdisciplinary Behavioral and Social Sciences Research (IBSS) competition.