Dictionary Multi-Modal Temporal Graph Learning.

Area of Science:

• Graph Deep Learning
• Temporal Graph Learning
• Multi-modal Learning

Background:

• Current temporal graph methods overlook multi-modal information, limiting their ability to model complex real-world dynamics.
• The absence of multi-modal architectures and datasets hinders advancements in temporal graph learning.

Purpose of the Study:

• To address the limitations of existing temporal graph learning methods by incorporating multi-modal information.
• To propose a novel multi-modal temporal graph learning framework and associated datasets.

Main Methods:

• Introduced ModalTGL, a framework utilizing dictionary graph networks for computational efficiency in dynamic scenarios.
• Employed embedding tuning for effective multi-modal fusion.
• Investigated the impact of various time encoding functions on dynamic information preservation.

Main Results:

• ModalTGL demonstrated significant performance improvements, achieving up to 18.48% enhancement.
• Experimental results on newly constructed multi-modal temporal graph datasets validated the proposed method's effectiveness.
• The study established the efficacy of integrating multi-modal data into temporal graph learning.

Conclusions:

• ModalTGL effectively enhances temporal graph learning by integrating multi-modal information.
• The proposed framework and datasets provide a foundation for future research in multi-modal temporal graph learning.
• This work highlights the importance of multi-modal data for a more comprehensive understanding of dynamic graph structures.