The Power of Images: A Computational Investigation of Political Mobilization via Social Media

社交媒体被一些学者描述为“弱者的武器”。现在，世界上任何拥有移动设备和网络连接的人都可以与无限大的潜在受众进行交流。该项目提议对倡导组织、社会抗议运动和极端主义激进组织的社交媒体交流中的图像进行首次大规模研究。该团队将实时收集社交媒体交流信息，并利用计算机科学最近的“深度学习”进步，为数百万张图像自动标记物体(人群、动物、武器等)、实体(特定演员)和引发的情感(恐惧、悲伤、幽默等)。该团队试图评估，在其他条件相同的情况下，这些图像是否能促进政治信息的参与和传播。技术摘要心理学研究发现，图像更容易让个人进行认知处理，并且比文字对意识的影响更大。对于传统上被边缘化的群体来说，图像应该是特别重要的工具，这些群体依靠局外人的战略来促进对问题的认识和他们的观点。然而，这是一个新的领域，因为没有理论和数据库来研究政治社交媒体传播的动员效果。研究小组将为范围广泛的政治团体构建新的社交媒体数据库。该项目将把新的标记政治图像集贡献给现有的计算机科学图像存储库。该小组还提议为社会科学家举办关于先进图像分类方法的国家讲习班，并将通过会议报告和同行评议出版物向学者和从业人员传播研究成果。

General AbstractSocial media have been described by some scholars as a "weapon of the weak." Any person in the world with a mobile device and network connection can now communicate with a potential audience of unlimited size. This project proposes the first large-scale study of the roll of images in the social media communications of advocacy organizations, social protest movements, and extremist militant groups. The team will collect social media communications in real time and exploit recent "deep learning" advances from computer science to automatically label millions of images for objects (crowds, animals, weapons etc.), entities (specific actors), and emotions evoked (fear, sadness, humor, etc.). The team seeks to assess whether, all else equal, these images promote political message engagement and diffusion.Technical AbstractResearch in psychology finds that images are easier for individuals to process cognitively, and have a larger impact on consciousness than do words. Images should be especially important tools for traditionally marginalized groups that rely on outsider strategies to promote issue awareness and their views. This is new ground however, as there is no theory and no databases for studying the mobilizing effects of political social media communications. The research team will construct new social media databases for a wide range of political groups. The project will contribute new labeled political image collections to existing computer science image repositories. The team also proposes to offer national workshops on advanced image classification methods for social scientists, and will disseminate research findings to scholars and practitioners through conference presentations and peer-reviewed publications.