Generalized Relation Learning with Semantic Correlation Awareness for Link Prediction
This work provides an incremental improvement for existing link prediction models by enhancing their ability to handle unbalanced and unseen relations, which is beneficial for researchers and practitioners working with knowledge graphs.
This paper addresses the challenges of unbalanced relation distributions and unseen relations in knowledge graph link prediction. The proposed Generalized Relation Learning (GRL) framework improves existing models by making them insensitive to unbalanced distributions and capable of learning unseen relations, as demonstrated through experiments on six benchmarks.
Developing link prediction models to automatically complete knowledge graphs has recently been the focus of significant research interest. The current methods for the link prediction taskhavetwonaturalproblems:1)the relation distributions in KGs are usually unbalanced, and 2) there are many unseen relations that occur in practical situations. These two problems limit the training effectiveness and practical applications of the existing link prediction models. We advocate a holistic understanding of KGs and we propose in this work a unified Generalized Relation Learning framework GRL to address the above two problems, which can be plugged into existing link prediction models. GRL conducts a generalized relation learning, which is aware of semantic correlations between relations that serve as a bridge to connect semantically similar relations. After training with GRL, the closeness of semantically similar relations in vector space and the discrimination of dissimilar relations are improved. We perform comprehensive experiments on six benchmarks to demonstrate the superior capability of GRL in the link prediction task. In particular, GRL is found to enhance the existing link prediction models making them insensitive to unbalanced relation distributions and capable of learning unseen relations.