Efficient Hyper-parameter Search for Knowledge Graph Embedding

Anonymous

Abstract: While hyper-parameters (HPs) are important for knowledge graph (KG) embedding, existing methods fail to search them efficiently. To solve this problem, we first analyze the properties of different HPs and quantize the transferability from small subgraph to the large graph. Based on the analysis, we propose an efficient two-stage search algorithm, which efficiently explores HP configurations on small subgraph at the first stage and transfers the top configurations for fine-tuning on the large whole graph at the second stage. Experiments show that our method can consistently find better HPs than the baseline algorithms with the same time budget. We achieve 10.8% average relevant improvement for four embedding models on the large-scale KGs in open graph benchmark.