Persuasion Strategies in Advertisements

Modeling what makes an advertisement persuasive, i.e., eliciting the desired response from consumer, is critical to study of propaganda, social psychology, and marketing. Despite its importance, computational modeling persuasion in computer vision still infancy, primarily due lack benchmark datasets that can provide persuasion-strategy labels associated with ads. Motivated by literature psychology marketing, we introduce extensive vocabulary strategies build first ad image corpus annotated strategies. We then formulate task strategy prediction multi-modal learning, where design a multi-task attention fusion model leverage other ad-understanding tasks predict The dataset also provides segmentation masks, which corresponding images on test split. publicly release our code at https://midas-research.github.io/persuasion-advertisements/.