Metapath-Guided Data-Augmentation For Knowledge Graphs

Knowledge graph (KG) embedding techniques use relationships between entities to learn low-dimensional representations of entities and relations. The traditional KG embedding techniques (such as TransE and DistMult) estimate these embeddings using the observed KG triplets and differ in their triplet scoring loss functions. As these models only use the observed triplets to estimate the embeddings, they are prone to suffer through data sparsity that usually occurs in the real-world knowledge graphs, i.e., the lack of enough triplets per entity. In this paper, we propose an efficient method to augment the triplets to address the problem of data sparsity. We use random walks to create additional triplets, such that the relations carried by these introduced triplets correspond to the metapath (sequence of underlying relations) induced by the random walks. We also provide approaches to accurately and efficiently choose the informative metapaths from the possible set of metapaths. The proposed augmentation approaches can be used with any KG embedding approach out of the box. Experimental results on benchmarks show the advantages of the proposed approach.