Probability Calibration for Knowledge Graph Embedding Models
This addresses the issue of unreliable probability estimates in knowledge graphs for researchers and practitioners, representing an incremental improvement by applying existing calibration techniques to a new domain.
The paper tackles the problem of probability calibration in knowledge graph embedding models, showing that popular models are uncalibrated and presenting a novel method that uses Platt scaling and isotonic regression to achieve well-calibrated models with significantly better results than uncalibrated ones, including state-of-the-art accuracy without relation-specific thresholds.
Knowledge graph embedding research has overlooked the problem of probability calibration. We show popular embedding models are indeed uncalibrated. That means probability estimates associated to predicted triples are unreliable. We present a novel method to calibrate a model when ground truth negatives are not available, which is the usual case in knowledge graphs. We propose to use Platt scaling and isotonic regression alongside our method. Experiments on three datasets with ground truth negatives show our contribution leads to well-calibrated models when compared to the gold standard of using negatives. We get significantly better results than the uncalibrated models from all calibration methods. We show isotonic regression offers the best the performance overall, not without trade-offs. We also show that calibrated models reach state-of-the-art accuracy without the need to define relation-specific decision thresholds.