Analyzing Emotional Expression in TikTok Videos Through Semantic Segmentation
Sean Chen (Drake University), Nicholas Haisler (Drake University), Marshall Tentis (Drake University), Md Reza (Drake University)
In this work, we explore how objects and environments in TikTok videos correlate with perceived emotions. We constructed a dataset of approximately 100 videos (50 labeled as happy, 50 as sad) based on human annotation. To analyze these videos, we extracted frames and applied Mask2Former, a high-resolution deep learning model for semantic segmentation, to identify and categorize objects within each frame. By examining the presence and distribution of objects and environmental features, we aim to uncover patterns that differentiate happy and sad videos. This study provides insight into the types of content TikTok users share when experiencing certain emotions, offering a deeper understanding of how social media serves as a platform for emotional expression.