Identifying Influential Users by Topic in Unstructured User-Generated Content

Abstract

Social media users generate a variety of content that can potentially influence their followers’ behaviors and preferences. However, most user-generated content, such as text and images, is unstructured, making it difficult to analyze how individuals impact the creation of such data as existing models primarily focus on the numerical aspects of behavioral data. In this article, we propose a method to identify influential users who significantly affect others’ interest in content topics, with a specific focus on text and images. Our study introduces a new variant of the topic model, incorporating a hierarchical structure and a vector auto-regressive approach. This method accounts for both the evolution of topic distribution and the social influence among users on their interest in content topics. The empirical application of our model to image-sharing social media data demonstrates that our model outperforms conventional topic models in terms of predictive accuracy and topic interpretability. Moreover, we illustrate how visualizing the estimated social influence within the network can provide valuable insights for seeded marketing campaigns and data-driven product development by identifying influential users in various content areas. This approach also offers a deeper understanding of evolving trends in content, preferences, and demand.