Data4TTMF_A Triple Trustworthiness Measurement Frame for Knowledge Graphs

There are no explicit labelled errors in FB15K. Considering the experience that most errors in real-world KG derive from the misunderstanding between similar entities, we consider the methods described in paper DoesWilliamShakespeareREALLYWrite Hamlet? Knowledge Representation Learning with Confidence to generate fake triples as negative examples automatically with less human annotation. Three kinds of fake triples may be constructed for each true triple: one by replacing head entity, one by replacing relationship, and one by replacing tail entity. We assign a label of 1 to positive examples and a label of 0 to negative examples. We also assure that the number of generated negative examples should be equal to that of positive examples.