Distributed Representations of Entities in Open-World Knowledge Graphs
This paper tackles the problem of handling emerging new entities in real-world knowledge graphs, which is a significant challenge for existing GNN-based methods that require observing all entities during training.
The paper addresses the challenge of new entities in knowledge graphs by introducing Decentralized Attention Network (DAN), which distributes entity semantics among neighbor embeddings using neighbor context as a query vector. DAN, trained with self-distillation, achieves competitive performance on conventional entity alignment and entity prediction, and significantly outperforms existing methods in open-world settings.
Graph neural network (GNN)-based methods have demonstrated remarkable performance in various knowledge graph (KG) tasks. However, most existing approaches rely on observing all entities during training, posing a challenge in real-world knowledge graphs where new entities emerge frequently. To address this limitation, we introduce Decentralized Attention Network (DAN). DAN leverages neighbor context as the query vector to score the neighbors of an entity, thereby distributing the entity semantics only among its neighbor embeddings. To effectively train a DAN, we introduce self-distillation, a technique that guides the network in generating desired representations. Theoretical analysis validates the effectiveness of our approach. We implement an end-to-end framework and conduct extensive experiments to evaluate our method, showcasing competitive performance on conventional entity alignment and entity prediction tasks. Furthermore, our method significantly outperforms existing methods in open-world settings.