Cross-fusion activates deep modal integration for multimedia recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Retrieval

Background:

• Recommendation systems are crucial for online platforms, impacting consumer behavior and transaction efficiency.
• Existing multimedia algorithms struggle to deeply integrate product and user interaction data for effective recommendations.
• Improving the fusion of multimodal information is key to advancing recommender system performance.

Purpose of the Study:

• To propose a novel cross-fusion-activated multi-modal (CFMM) integration method for recommender systems.
• To achieve deep fusion of product and user information by integrating diverse data modalities.
• To enhance the performance of recommendation systems through improved data integration and a novel fusion loss function.

Main Methods:

• Developed a cross-fusion module for deep fusion of features from different modalities.
• Introduced a fusion loss function to optimize the recommendation network.
• Conducted extensive experiments and ablation studies on three real-world datasets.

Main Results:

• The proposed CFMM method demonstrated superior recommendation performance compared to existing algorithms.
• Achieved a maximum improvement of 3.8% in key metrics like Recall@20, NDCG@20, and Precision@20.
• Validated the effectiveness of the cross-fusion module and fusion loss function through ablation studies.

Conclusions:

• The CFMM method successfully achieves deeper integration of multimodal information for enhanced recommendations.
• The approach offers significant improvements in recommendation accuracy and efficiency.
• Future work can further enhance performance by extending multimodal information interaction algorithms.