Artwork Recommendation Based on Multi-modal Preference Perception

Recommendation systems are widely applied across various domains, enabling users to efficiently discover content of interest within vast amounts of information. Artwork recommendation is a significant application area; however, existing research on specialized recommendation methods for artworks remains limited. Additionally, the potential of leveraging multimodal information in artwork recommendation is often overlooked. Effectively integrating multimodal information is crucial for comprehensively understanding user preferences and optimizing recommendation performance in the artwork domain. To address this gap, this paper proposes an innovative multimodal-integrated artwork recommendation approach. This method incorporates multidimensional artwork information, deeply explores user preferences, and mitigates the issue of interaction data sparsity by designing a tailored self-supervised task to enhance preference modeling. Experimental results on a self-constructed dataset demonstrate the effectiveness of the proposed method, achieving superior recommendation quality compared to baseline approaches.