Combining two and three-way embedding models for link prediction in knowledge bases

摘要: This paper tackles the problem of endogenous link prediction for knowledge base completion. Knowledge bases can be represented as directed graphs whose nodes correspond to entities and edges relationships. Previous attempts either consist powerful systems with high capacity model complex connectivity patterns, which unfortunately usually end up overfitting on rare relationships, or in approaches that trade simplicity order fairly all frequent not. In this paper, we propose TATEC, a happy medium obtained by complementing high-capacity simpler one, both pre-trained separately then combined. We present several variants different kinds regularization combination strategies show approach outperforms existing methods types relationships achieving state-of-the-art results four benchmarks literature.