An Improved TransE Algorithm

Abstract

This paper proposes the TransE-CBA (TransE Based On Improved Bernoulli And Adam) model, an improved TransE model. We used an improved Bernoulli distribution sampling method to improve the accuracy of negative samples, and used the Adam algorithm to update the gradient to improve the performance of the algorithm. Experiment result shows that, on the FB13 and WN18 datasets, the TransE-CBA model has a lower average ranking score (MeanRank) than TransE as a whole, and the proportion of the top ten triples (HITS@10) Higher, better sorting effect.