Knowledge Graph Embedding for Triples Fact Validation.

This poster presents a methodology for designing and implementing a knowledge graph fact checking using graph embeddings models. The implementation has been tested on the dataset of task 1 of the ISWC 2019 challenge to assess the correctness of a statement. We trained 6 embedding models : DistMult, HolE, TransE, TransR, ComplEx and RDF2VEC. Several machine learning algorithms have been tested to classify the triples given their embeddings using a 4-fold cross validation scheme on the entire dataset. The results indicate that RDF2VEC gives the higher AUC score of 0.877 for the prediction of the correctness of the statements. According to the evaluation report obtained from the challenge board, our team’s score came third among nine participating teams to the fact validation task 1 challenge.