Hetgeormer: Heterogeneous Graph Enhancement Transformer for Additive Manufacturing Knowledge Graph Completion

Knowledge Graph Completion (KGC) aims to predict the missing entities and relations in the Knowledge Graphs (KGs), which is the key task to improve the quality and completeness of the additive manufacturing KGs. Traditional triple-based methods mainly achieve structured embedding by means of short-range interactions while ignoring the global graph structure information. The Graph Neural Network-based methods are limited by the few number of entity neighbors in the KGs, which contain noisy and irrelevant information. To address the above issues, we propose a novel heterogeneous graph enhancement transformer (Hetgeormer), which aims to fully capture global heterogeneous graph information to enhance the representation of entities and relations. More specifically, we first design a novel context subgraph generation strategy, aiming to extract the effective information of the center entity in the current triple from the graph level. Then we use a transformer block to capture the long-range dependencies of the generated context subgraph sequence, meanwhile designing a homogeneous graph attention enhancement to reduce structural information loss. We finally use the masked language model to implement knowledge graph embedding representation learning for better KGC tasks. Experimental results demonstrate the effectiveness of our method, especially for entities that have few neighbors in additive manufacturing. In addition, ablation experiments also indicate that each submodule of our method is effective and robust.