Very happy to announce our new paper accepted in @eswc_conf
#ESWC2024: "Treat Different Negatives Differently: Enriching Loss Functions with Domain and Range Constraints for Link Prediction"!
📎 https://arxiv.org/pdf/2303.00286.pdf
w/ N. Hubert, A. Brun, and D. Monticolo
#knowledgeGraph #semanticWeb #machineLearning #linkPrediction #neurosymbolicAI #artificialIntelligence #linkedOpenData #graphEmbeddings #embeddings #graphNeuralNetworks
Contrary to negative sampling, we sample all kinds of negatives but enrich the three main loss functions for link prediction such that negatives are treated differently based on their semantic validity.
We posit that negative triples that are semantically valid w.r.t. signatures of relations (domain and range) are high-quality negatives. The proposed loss functions systematically provide satisfying results which demonstrates both the generality and superiority of our approach.
The enriched loss functions (1) lead to better MRR and Hits@10 values, and (2) drive KGEMs towards better semantic correctness as measured by the Sem@K metric. This highlights that relation signatures globally improve KGEMs, and thus should be incorporated into loss functions.
Looking forward to feedback from the community!
Pierre Monnin