A benchmark for cross-modal emotion-infused internet meme recommendation
Teng Zhang, Leihan Zhang, Qiang Yan · 2025
As an emerging phenomenon in internet culture, Internet memes have reshaped how people interact on social media. With their dissemination attributes and expressive capabilities, memes play crucial roles in discussions on social issues, cultural dissemination, marketing, and more. However, the rapid growth of internet memes has also brought about the issue of information overload, and traditional recommendation techniques may need to be revised in handling such multimodal content. This study aims to explore how to integrate emotional factors into meme recommendation. Firstly, we developed the Meme Recommendation benchmark, consisting of 6 communities, 3630 topics, and 148,886 records, along with evaluation metrics. Building upon this, we propose the IEF model, which integrates the emotional features of memes to capture users’ emotional needs effectively. By leveraging large-scale models to generate image captions as knowledge enhancement, the performance of the recommendation system is improved. Experimental results demonstrate the significant effectiveness of the IEF model in meme recommendation, providing new insights and approaches for recommendation system applications.