A Novel Embedding Model for Knowledge Graph Completion Based on Dilated Convolution

Due to the limitations of information extraction techniques and concerns over data privacy, most existing knowledge graphs suffer from incompleteness. Knowledge graph embedding models have become the mainstream solution to address this issue. However, these models are unable to achieve a satisfactory balance between model expressiveness and complexity. In this paper, we propose ConvAdr, which constructs relation convolutional filters and incorporates dilated convolution. This approach effectively enhances the interaction between entities and relations while avoiding a substantial increase in the model's parameter size, thereby improving performance. Additionally, we introduce a residual network to address the issue of information loss, enabling the model to better learn complex feature representations. By evaluating on the three baseline datasets, our model has varying degrees of superiority on most of the evaluation metrics when compared to the other baseline models.