Opinion convergence-based sentiment prediction of image advertisement

In this study, a novel approach was proposed for the sentiment prediction of image advertisements. Unlike general multilabel problems with correct answers, image sentiment prediction is a multilabel problem that involves converging the opinions of various labelers. Therefore, an opinion convergence-based image sentiment prediction methodology was proposed to model the decision-making process of image sentiment prediction. Hypothetical labelers were generated by recombining training datasets to maximize the characteristic distance, and each prediction model was trained with each combination to represent each hypothetical labeler with distinct characteristics. The results of the experiment revealed that the proposed image sentiment prediction method outperformed other existing models with advanced architectures or considered various factors for improving the accuracy of image sentiment prediction tasks. Moreover, the effectiveness of the proposed method was verified through qualitative experiments.