Performance Evaluation and Factor Analysis of Multimodal Recommender Systems in Cold-Start User Scenarios

Key Points

• This research aims to verify the effectiveness of multimodal recommender systems in addressing the cold-start problem.
• Real-world dataset utilized to create a cold-start user environment.
• Performance of multimodal recommender systems compared with collaborative filtering methods.
• Analysis included evaluation of various MMRS approaches like FREEDOM and MGCN.
• MMRS approaches outperformed collaborative filtering methods like LightGCN in cold-start scenarios.
• Item-item graph-based approaches reflecting multimodal similarity showed superior performance.
• Refined strategies for graph application were identified as key factors for performance enhancement.

Abstract