Understanding User Engagement in Museum TikTok Videos by Exploring Multimodal Cues

TikTok has emerged as a valuable platform for cultural institutions to connect with global audiences, offering museums a valuable opportunity to reach out through organic, self-created content. These social media video posts contain rich and abundant information, including structured and unstructured data. This paper aims to explore user engagement in museum TikTok videos by analysing various multimedia cues to estimate user engagement scores. We gathered a multimodal dataset of TikTok posts from seven well-known museums in this research. To properly quantify user engagement on Tiktok, we developed a feasible scoring function to capture highly correlated metrics from the posts. We also propose a multimodal framework that allows a combination of high-level feature embeddings extracted from video and metadata. Experiments using a random forest regressor showed that the fusion of caption sentiments and hashtag features can outperform individual feature baselines in predicting user engagement. Nevertheless, visual-based features remain promising, providing considerable opportunity for future work. This study offers a glimpse into potential data-driven models and applications that can benefit museums on social media.