A survey on social media popularity prediction: concepts, technologies, applications, and challenges

Due to the rapid development of user scale, interactive forms, and social media content diversity on social platforms, social media popularity prediction, aiming to predict the popularity of user generated content on social network according to multimodal and statistical features, is a significant challenging task, which is essential for many related downstream applications, such as personalized recommendation systems, marketing management, and public opinion governance, etc. Abundant research works provide a better understanding of the key factors driving the diffusion of social media content, but lacking an unified taxonomy of social media popularity prediction methods. In this paper, we comprehensively review social media popularity prediction methods over the past decade. Specifically, we first formally define the social media popularity prediction task and introduce the type of multi-modality content. Subsequently, we provide a novel taxonomy architecture from methodology perspective, which categorize the existing works into the five groups and their subclasses in each group, and analyze their advantages and disadvantages. Meanwhile, we outline the mainstream public datasets and evaluation metrics in detail. Finally, we discuss the open application and future direction in this field. This paper comprehensively provides valuable references for readers to understand the cutting-edge researches.