Automated Visual Analysis for the Study of Social Media Effects: Opportunities, Approaches, and Challenges
Automated Visual Analysis for the Study of Social Media Effects: Opportunities, Approaches, and Challenges
复制标题
DOI：
10.1080/19312458.2023.2277956
复制
发表时间：
2023-11
影响因子：
11.4
通讯作者：
Yilang Peng;Irina Lock;Albert Ali Salah
中科院分区：
文献类型：
--
作者：
Yilang Peng;Irina Lock;Albert Ali Salah
为了提高我们对社交媒体效应的理解，将日益流行的视觉媒体纳入我们的研究至关重要。在本文中，我们讨论了自动视觉分析研究社交媒体效应的理论机会，并概述了可以促进这一研究的现有计算方法。具体来说，我们强调了现有计算机视觉工具的输出与媒体效果研究相关的理论概念之间的差距。我们提出了多种方法来弥合自动视觉分析中的这一差距，例如证明现有工具中特定视觉特征的理论意义，开发监督学习模型来测量感兴趣的视觉属性，以及应用无监督学习来发现有意义的视觉主题和类别。最后，我们讨论了计算通信研究中自动化可视化分析的未来方向，例如开发基准数据集，旨在反映更有理论意义的概念，并结合大型语言模型和多模态通道来提取见解。
ABSTRACT To advance our understanding of social media effects, it is crucial to incorporate the increasingly prevalent visual media into our investigation. In this article, we discuss the theoretical opportunities of automated visual analysis for the study of social media effects and present an overview of existing computational methods that can facilitate this. Specifically, we highlight the gap between the outputs of existing computer vision tools and the theoretical concepts relevant to media effects research. We propose multiple approaches to bridging this gap in automated visual analysis, such as justifying the theoretical significance of specific visual features in existing tools, developing supervised learning models to measure a visual attribute of interest, and applying unsupervised learning to discover meaningful visual themes and categories. We conclude with a discussion about future directions for automated visual analysis in computational communication research, such as the development of benchmark datasets designed to reflect more theoretically meaningful concepts and the incorporation of large language models and multimodal channels to extract insights.