Canonical Tensor Decomposition for Knowledge Base Completion

ICML 2018 Timothée LacroixNicolas UsunierGuillaume Obozinski

The problem of Knowledge Base Completion can be framed as a 3rd-order binary tensor completion problem. In this light, the Canonical Tensor Decomposition (CP) (Hitchcock, 1927) seems like a natural solution; however, current implementations of CP on standard Knowledge Base Completion benchmarks are lagging behind their competitors... (read more)

PDF Abstract

Evaluation results from the paper

Task Dataset Model Metric name Metric value Global rank Compare
Link Prediction FB15k ComplEx-N3 (reciprocal) MRR 0.86 # 1
Link Prediction FB15k ComplEx-N3 (reciprocal) [email protected] 0.91 # 1
Link Prediction FB15k-237 ComplEx-N3 (reciprocal) MRR 0.37 # 2
Link Prediction FB15k-237 ComplEx-N3 (reciprocal) [email protected] 0.56 # 2
Link Prediction WN18 ComplEx-N3 (reciprocal) MRR 0.95 # 4
Link Prediction WN18 ComplEx-N3 (reciprocal) [email protected] 0.96 # 2
Link Prediction WN18RR ComplEx-N3 (reciprocal) MRR 0.48 # 4
Link Prediction WN18RR ComplEx-N3 (reciprocal) [email protected] 0.57 # 3
Link Prediction YAGO3-10 ComplEx-N3 (reciprocal) MRR 0.58 # 1
Link Prediction YAGO3-10 ComplEx-N3 (reciprocal) [email protected] 0.71 # 1