A Novel Method for Boosting Knowledge Representation Learning in Entity Alignment through Triple Confidence

• DOI: 10.3390/math12081214
• Publication Date: 2024-04-18T10:21:12Z
• Authors: Xiaoming Zhang; Tongqing Chen; Huiyong Wang
• Keywords: Adjacency matrix; Boosting

Entity alignment is an important task in knowledge fusion, which aims to link entities that have the same real-world identity two graphs. However, process of constructing a graph, some noise may inevitably be introduced, must affect results entity tasks. The triple confidence calculation can quantify correctness triples reduce impact on alignment. Therefore, we designed method calculate and applied it representation learning phase calculates based pairing rates three angles between relations. Specifically, uses as features, are then fed into feedforward neural network for training obtain confidence. Moreover, introduced methods improve their performance For graph network-based GCN, considered when calculating adjacency matrix, translation-based TransE, proposed strategy dynamically adjust margin value loss function These were alignment, experimental demonstrate compared with without integrating confidence, confidence-based achieved excellent task.