An Improved Translation-Based Method for Knowledge Graph Representation
Jincheng Xu, Ge Yunsheng, Zhengxia Wu · 2020
Traditional knowledge graph representation methods based on translation models have low-quality negative samples and false negative examples. In order to solve the shortcomings of translation models, this paper proposes an improved translation model knowledge representation model which is called TransE-KC. First, introduce the K-Means algorithm for clustering, and then randomly select 20 entities in different clusters to calculate the similarity between the replaced entities, and rank them, select the highest ranked entity, and perform Replacement; secondly, for the problem of "false negatives", this article introduces a Cuckoo Filter to filter it. The experimental results on the public data set show that, compared with the TransE model, the TransE-KC model has a greater improvement in link prediction average ranking and triplet classification accuracy.