Exploiting Patent Documents for Cross-Domain Knowledge Transfer in Innovative Engineering Design: A Doc2Vec-GAT-Based Approach

The innovative design of complex engineering products is a knowledge-intensive and technology- driven task. Since technologies with the same functions could be deployed in similar situations under different domains, cross-domain knowledge transfer would prosper the success of innovative engineering product design. However, transferring proper specialized domain knowledge faces the defect of returning arbitrary results. One of the causes is the insufficient knowledge mining of the knowledge resources. Given that patents are high-quality knowledge resources in almost all technological fields, this study presents a Doc2Vec-Graph Attention Network (GAT)-based approach to exploit patent documents for cross-domain knowledge transfer. First, a three-dimensional patent knowledge model containing domain, function, and technology was built to define the key features of patent contents. Second, an approach integrating Doc2Vec and GAT was proposed to learn the patent content and patent citation relationships respectively, thereafter constructing a patent knowledge space. Third, to formalize the designer queries, a knowledge requirement space is established by matching the vectorized queries to the most similar patents. Finally, a cross-domain knowledge transfer mechanism is proposed based on a diversified searching method to map the knowledge requirement space to the patent knowledge space. A case study of the knowledge transfer on a sealing structure design of aircraft fuel tank and quantitative comparative experiments verified the feasibility of our approach.