Enhancing Cross-domain Link Prediction via Evolution Process Modeling

2024-02-03

AI Summary

This paper addresses cross-domain link prediction in dynamic graphs, proposing DyExpert—the first model to achieve zero-shot cross-domain generalization: it accurately predicts future links in unseen target-domain graphs without any target-domain training. Methodologically, DyExpert unifies graph structural evolution and link prediction via a conditional link generation mechanism, incorporates a decoder-specific Transformer architecture, and undergoes multi-domain joint pretraining on 6 million dynamic edges. Its core innovation lies in explicitly modeling historical evolutionary paths to learn domain-specific dynamic patterns and adaptively calibrate predictions. Evaluated on eight unseen test graphs, DyExpert achieves an average 11.40% improvement in Average Precision (AP) over eight fully supervised baselines. This work significantly advances cross-domain transfer learning for dynamic graphs by enabling robust, training-free adaptation across heterogeneous temporal graph domains.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Application Domains: Humanities & Computational Social Science

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

This work proposes DyExpert, a dynamic graph model for cross-domain link prediction. It can explicitly model historical evolving processes to learn the evolution pattern of a specific downstream graph and subsequently make pattern-specific link predictions. DyExpert adopts a decode-only transformer and is capable of efficiently parallel training and inference by extit{conditioned link generation} that integrates both evolution modeling and link prediction. DyExpert is trained by extensive dynamic graphs across diverse domains, comprising 6M dynamic edges. Extensive experiments on eight untrained graphs demonstrate that DyExpert achieves state-of-the-art performance in cross-domain link prediction. Compared to the advanced baseline under the same setting, DyExpert achieves an average of 11.40% improvement Average Precision across eight graphs. More impressive, it surpasses the fully supervised performance of 8 advanced baselines on 6 untrained graphs.

Problem

• Dynamic graph model for cross-domain link prediction
• Historical evolving processes modeling
• Efficient parallel training and inference

Innovation

• Dynamic graph model
• Decode-only transformer
• Conditioned link generation