Dual view graph transformer networks for multi-hop knowledge graph reasoning.

Area of Science:

• Artificial Intelligence
• Data Science
• Knowledge Representation

Background:

• Knowledge graphs (KGs) often suffer from incompleteness, necessitating multi-hop reasoning to infer missing information.
• Existing multi-hop reasoning methods face challenges with training inefficiency (reinforcement learning) or loss of structured knowledge (sequence-based methods).

Purpose of the Study:

• To propose a novel method, Dual View Graph Transformer Networks (DV4KGR), for effective multi-hop knowledge graph reasoning.
• To enable joint learning of structured and serialized views for improved reasoning capabilities.

Main Methods:

• DV4KGR utilizes a structured view for global relation representation and a serialized view for reasoning semantics.
• Employs supervised contrastive learning to represent one-to-many relations effectively.
• Combines structured knowledge with rule induction for action smoothing to mitigate overfitting.

Main Results:

• DV4KGR demonstrates superior performance compared to state-of-the-art baselines across four benchmark datasets.
• The joint learning approach enhances the ability to capture and represent complex relations.

Conclusions:

• DV4KGR offers a more efficient and interpretable approach to multi-hop knowledge graph reasoning.
• The dual-view strategy effectively addresses limitations of prior methods, improving overall knowledge graph completion and understanding.