EIJL: Popularity Prediction of Social Media Advertisements Based on Multimodal Emotional Interaction and Joint Learning

Predicting the popularity of social media advertisements holds significant value in brand marketing. However, current prediction methods rely heavily on user networks and temporal data, overlooking the role of ad content and user emotions. This study aims to predict ad popularity by integrating multimodal emotional interactions between ad content and user feedback. The research collected image-text advertisement data from social media and employed the CLIP multimodal model to extract multimodal features of the advertisements. By extracting emotional features from advertisement text and images, the study investigates the interaction between text emotion, image emotion, and ad likability on advertisement popularity. An emotional interaction layer was designed, and a multi-task joint learning method that uses classification prediction to assist regression prediction was adopted to predict advertisement popularity. The study found that the interaction between text emotion, image emotion, and ad likability negatively impacts ad sharing. The integration of multimodal information, multi-task joint learning, and the emotional interaction layer effectively enhanced the model’s ability to predict popularity, with a significant improvement in the MEA index. This research demonstrates that leveraging multimodal emotional interaction information to enhance neural network predictions of advertisement popularity is effective from the perspective of meme propagation. This approach provides valuable insights and directions for further optimizing popularity prediction.