Incorporating Uncertainty of Entities and Relations into Few-Shot Uncertain Knowledge Graph Embedding

Jingting Wang, Tianxing Wu, Jiatao Zhang

Abstract: In this paper, we study the problem of embedding few-shot uncertain knowledge graphs. Observing the existing embedding methods may discard the uncertainty information, or require sufficient training data for each relation, we propose a novel method by incorporating the inherent uncertainty of entities and relations (i.e. element-level uncertainty) into uncertain knowledge graph embedding. We introduce different metrics to quantify the uncertainty of different entities and relations. By employing a metric-based framework, our method can effectively capture both semantic and uncertainty information of entities and relations in the few-shot scenario. Experimental results show that our proposed method can learn better embeddings in terms of the higher accuracy in both confidence score prediction and tail entity prediction.