As digital experiences increasingly blend text, image, and auditory information, the signals that drive attention and engagement are inherently multimodal. This talk distills practical insights from both structured and unstructured multimodal data. Specifically, Prof. Kanuri discusses how temporal variation in musical attributes shapes the resharing of short-form videos, revealing patterns that distinguish forgettable clips from viral ones. Building on these findings, he presents a fused, explainable wide-and-deep architecture that integrates text and visual cues to synthesize content and aid in influencer selection. Finally, he will introduce a novel method for discovering new moderators from multimodal data, which can help turn black-box predictions into actionable guidance. Evaluated across diverse datasets, these insights offer concrete takeaways for firms seeking to leverage multimodal data.
