Multi-Partition Embedding Interaction with Block Term Format for Knowledge Graph Completion

29 Jun 2020  ·  Hung Nghiep Tran, Atsuhiro Takasu ·

Knowledge graph completion is an important task that aims to predict the missing relational link between entities. Knowledge graph embedding methods perform this task by representing entities and relations as embedding vectors and modeling their interactions to compute the matching score of each triple. Previous work has usually treated each embedding as a whole and has modeled the interactions between these whole embeddings, potentially making the model excessively expensive or requiring specially designed interaction mechanisms. In this work, we propose the multi-partition embedding interaction (MEI) model with block term format to systematically address this problem. MEI divides each embedding into a multi-partition vector to efficiently restrict the interactions. Each local interaction is modeled with the Tucker tensor format and the full interaction is modeled with the block term tensor format, enabling MEI to control the trade-off between expressiveness and computational cost, learn the interaction mechanisms from data automatically, and achieve state-of-the-art performance on the link prediction task. In addition, we theoretically study the parameter efficiency problem and derive a simple empirically verified criterion for optimal parameter trade-off. We also apply the framework of MEI to provide a new generalized explanation for several specially designed interaction mechanisms in previous models. The source code is released at https://github.com/tranhungnghiep/MEI-KGE.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Link Prediction FB15k MEI (small) Hits@10 0.878 # 3
Hits@1 0.757 # 2
Hits@3 0.823 # 3
MRR 0.800 # 2
Link Prediction FB15k MEI-BTD MRR 0.806 # 6
Hits@10 0.893 # 7
Hits@3 0.843 # 3
Hits@1 0.754 # 3
Link Prediction FB15k-237 MEI MRR 0.365 # 14
Hits@10 0.552 # 10
Hits@3 0.402 # 11
Hits@1 0.271 # 12
Link Prediction KG20C MEI (small) MRR 0.230 # 1
Hits@1 0.157 # 1
Hits@3 0.258 # 1
Hits@10 0.368 # 1
Link Prediction WN18 MEI (small) MRR 0.951 # 5
Hits@10 0.960 # 5
Hits@3 0.953 # 9
Hits@1 0.946 # 6
Link Prediction WN18 MEI-BTD MRR 0.950 # 8
Hits@10 0.957 # 13
Hits@3 0.952 # 10
Hits@1 0.946 # 6
Link Prediction WN18RR MEI MRR 0.481 # 33
Hits@10 0.551 # 49
Hits@3 0.496 # 27
Hits@1 0.444 # 22
Link Prediction YAGO3-10 MEI MRR 0.578 # 5
Hits@10 0.709 # 5
Hits@1 0.505 # 4
Hits@3 0.622 # 2
MR 756 # 1

Methods