Understanding Influence Operations via Images

Online social media platforms can be (mis)used to spread state-sponsored online influence operations. This could involve the manipulation of digital content to shape public opinion. Prior works have predominantly focused on the role of textual content in the spread of influence operations. In this work, we study the role of image coordination patterns across multiple influence operations on a social media platform (Twitter). We focus on two tasks. First, we identify image coordination by constructing image-image similarity graphs based on only image features. Second, we characterize images involved in influence operations to unveil politically and non-politically motivated images. We find varying degrees of coordination through images in different influence operations by the same state. Further, we also observe the use of politically motivated similar images extensively in these influence operations.