Association Rules Enhanced Knowledge Graph Attention Network

Embedding knowledge graphs into continuous vector spaces has recently attracted increasing interest in base completion. However, most existing embedding methods, only fact triplets are utilized, and logical rules have not been thoroughly studied for the completion task. To overcome problem, we propose an association enhanced graph attention network (AR-KGAN). In this paper, jointly modeled proposed unified framework to achieve more predictive entity relation embeddings. Association corresponding correlation degrees between them can be automatically obtained according our designed mining algorithm. The major component of AR-KGAN is encoder effective neighborhood aggregator, which addresses problems by aggregating neighbors with both based weights. decoder enables translational entities relations while keeping superior link prediction performance. Then, global loss minimized over atomic complex formulas manner, learn embeddings compatible rules, certainly acquisition inference. results show that model achieves significant consistent improvements state-of-the-art methods on three benchmark datasets.