Entity Alignment Algorithm Based on Dual-Attention and Incremental Learning Mechanism

• Main Authors: Yang Yang, Maojie Hao, Yonghua Huo, Liandong Chen, Zhipeng Gao
• Format: Article
• Language: English
• Published: IEEE 2019-01-01
• Series: IEEE Access
• Subjects: Entity alignment; knowledge map; attention; neural network; binary classification
• Online Access: https://ieeexplore.ieee.org/document/8892565/

With the development of artificial intelligence and big data technology, large-scale general knowledge map construction is becoming increasingly important. One of the most efficient methods is undoubtedly the integration of existing knowledge maps, and entity alignment is the key in the process of knowledge map fusion. The merits of the entity alignment algorithm directly affect the efficiency and accuracy of the knowledge map fusion. However, there are some problems with the current Chinese knowledge map entity alignment algorithm, such as its low accuracy, difficulty in generating solid vectors, and difficulty in obtaining a priori alignment data. In this paper, the entity alignment algorithm is understood to be a neural network binary classification model, and we propose an entity alignment algorithm based on the dual-attention mechanism. The algorithm improves the entity vector training process, proposes a dual-attention mechanism, and applies an incremental learning mechanism. The experiments show that the improvements proposed in this paper effectively improve the classification accuracy of the algorithm, and the overall effect of the algorithm is better than that of the existing physical alignment algorithm.