A hierarchical interactive multi-channel graph neural network for technological knowledge flow forecasting

Technological advancement can provide new and more cost-effective solutions to challenges in critical areas. Therefore, as one of the important sources for technological progress, technological knowledge flow (TKF) forecasting, i.e., predicting the directional flows of knowledge from one technological field to another, has become a hot issue of widespread concern. However, existing researches either rely on labor-intensive empirical analysis or ignore the intrinsic characteristics inherent in TKF. To this end, we present a data-driven solution in this article, namely a hierarchical interactive multi-channel graph neural network (HIMTKF). Specifically, HIMTKF generates final predictions using two types of vector representations for each technology node (a diffusion vector and an absorption vector), which is realized by four components: high-order interaction module (HOI), co-occurrence module (CO), improved hierarchical delivery module (IHD) and technological knowledge flow tracing module (TFT). For one thing, HOI and CO are designed to represent high-order network relationships and co-occurrence relationships between technologies on the same hierarchy level. For another, IHD is aimed to model the hierarchical relationships between technologies while also taking their personalities into account. Then, TFT is intended for capturing the dynamic feature evolution of technologies with the above relations involved. Additionally, we develop a hybrid loss function and propose a new evaluation metric for better forecasting the unprecedented knowledge flows between technologies. Finally, we conduct extensive experiments on a large dataset of real-world patents. The results validate the effectiveness of our approach and shed light on several intriguing phenomena about technological knowledge flow trends.