Machine Learning-Based Patent Transferability Forecasting for Emerging Technology: A case study of graphene-based supercapacitors

Graphene-based supercapacitors, as an emerging technology, are at the vanguard of innovation and transformation in the field of energy storage. The accurate identification of the technology transfer potential of graphene supercapacitor patents can facilitate the market application of emerging technologies, thereby promoting innovation in energy technology and sustainable development. The objective of this study is to introduce and compare forecasting models for the transferability of graphene sensor patents and to address the lack of interpretability in existing patent transferability models. This study focuses on the process of knowledge creation and diffusion, selecting patent text and patent examination features to represent patent quality and citation network features to quantify the level of technological innovation of patents. It then constructs a patent technology transfer potential evaluation system consisting of sixteen indicators. In order to ascertain the technology transfer potential of graphene-based supercapacitors, patent transfer and licensing data were employed as proxy variables. Patent transfer and licensing data from universities and research institutes were used as positive samples, while data from patents held by universities and research institutes that had not been transferred until their expiration were used as negative samples. These were combined to construct the training sample set and the test sample set. This study utilises five distinct prediction models. The models employed were the Random Forest (RF), Support Vector Machine (SVM), Gradient Boosting Decision Tree (GBDT), Adaptive Boosting (AdaBoost), and Back Propagation (BP) Algorithm. The performance of these models is evaluated using a series of metrics, including precision, recall, F1 score, and accuracy. Additionally, the Shapley Additive Explanations (SHAP) model is introduced to identify key features. The findings of the research indicate that the GBDT model Algorithm model has the highest level of accuracy. The features of citation networks, including Network Constraint Index, Out-Closeness Centrality, In-Closeness Centrality and Out-Degree Centrality, exert a considerable impact on the potential for technology transfer in patents.