Knowledge Base Completion by Learning to Rank Model

Knowledge base (KB) completion aims to predict new facts from the existing ones in KBs. There are many KB completion approaches, one of the state-of-art approaches is Path Ranking Algorithm (PRA), which predicts new facts based on path types connecting entities. PRA treats the relation prediction as a classification problem, and logistic regression is used as the classification model. In this work, we consider the relation prediction as a ranking problem; learning to rank model is trained on path types to predict new facts. Experiments on YAGO show that our proposed approach outperforms approaches using classification models.