Popularity Prediction of SNS Accounts using Transformer-Based Models with Multimodal Features

Abstract

The widespread use of social networking services (SNS) has elevated the importance of gaining popularity for both businesses and individuals. In the field of popularity analysis, various predictive models have been proposed to deepen our understanding of popularity. However, many previous studies have focused on analyzing individual content types such as images or videos, leaving a gap in methods for popularity analysis targeting SNS accounts, which serve as the source of such content and represent a higher-level concept. Therefore, this study proposes a model for predicting the popularity dynamics of SNS accounts. We target SNS accounts, utilizing multimodal features such as account information, text, images, and videos as inputs. When conducting predictive experiments using real-world data, the results obtained are relatively lower compared to the baseline. However, we are able to demonstrate potential improvements and directions for this task.