DSGRec: dual-path selection graph for multimodal recommendation | JoVE Visualize

Area of Science:

Computer Science
Artificial Intelligence
Data Science

Background:

Multi-modal recommendation systems are crucial for digital streaming.
Existing graph-based methods using graph convolutional networks (GCNs) have limitations in interpreting user interaction signals and integrating multi-modal information.
There's a need for enhanced collaboration between user behavior and multi-modal data in recommendation systems.

Purpose of the Study:

To propose a novel dual-path selective graph recommender system (DSGRec) that addresses limitations in current multi-modal recommendation approaches.
To enhance the accuracy and personalization of recommendations by improving the integration of user interaction signals and multi-modal item information.
To facilitate positive collaboration between interactive data and multi-modal information for superior recommendation performance.

Main Methods:

Decomposition of interaction information into a dual-path selection architecture.
Introduction of behavior-aware multimodal signal augmentation to extract rich semantic information.
Implementation of hypergraph-guided cooperative signal enhancement to capture hybrid global information.
Utilizing a primary module for learning dual-path selection signals and two auxiliary modules with independent contrastive learning tasks for signal adjustment.

Main Results:

DSGRec demonstrates superior performance compared to state-of-the-art recommendation baselines.
The proposed method effectively models user behavior and integrates multi-modal information.
Experimental results on three benchmark datasets validate the effectiveness of the dual-path architecture and enhancement components.

Conclusions:

DSGRec offers a more effective approach to multi-modal recommendation by enhancing the collaboration between user behavior and multi-modal data.
The dual-path selection architecture and auxiliary modules significantly improve recommendation accuracy and personalization.
The method provides a promising direction for future research in graph-based multi-modal recommendation systems.