Self-Adaptive Knowledge Embedding for Large-Scale Electronic Component Knowledge Graph

Abstract

Substitution of electronic components is an important research topic in the supply chain management of design and manufacture of electronic products. Previous studies mainly use simulation technology and case study, the system is complex and unable to comprehensively evaluate the different properties of components in each application environment. In this paper, we propose the Electronic Component Knowledge Graph (ECKG), which helps to discover knowledge from a large amount of data and assist in the substitution of electronic components. The ECKG integrates the electronic component data from different manufacturers and contains substitution relations labeled by domain expert experience. The ECKG contains two types of nodes: the central node is the representation of electronic components, and the peripheral node contains the attribute values that provides semantic support for the central node, which helps learning the structural knowledge. Moreover, we present the Self-adaptive Knowledge Embedding (SAKE) approach that integrates the semantic information of peripheral nodes into their corresponding central node. The SAKE is pre-trained on our large-scale ECKG with a knowledge-based attention mechanism to obtain the contextual representation of the central nodes. Experiment results show that SAKE outperforms other counterparts on the entity typing and link prediction tasks.