Introduction: Identifying the technologies that will drive technological changes over the coming years is important for the optimal allocation of firms’ R&D resources and the deployment of innovation strategies. The citation frequency of a patent is widely recognized as representative of the patent’s value. Thus, identifying potential highly cited patents is an important goal. A number of studies have attempted to distinguish highly cited patents from others based on statistical models, but a more effective and applicable method needs to be further developed.Methods: This paper treats the prediction of later patent citations as a classification problem and proposes a novel framework based on machine learning methods. First, a indices system to identify highly cited patents is constructed using multiple factors that are believed to influence citation frequency. Second, various machine learning models are utilized to identify highly cited patents. The optimized model with the best generalization capability is selected to predict the future impacts of newly applied patents, which may be representative of emerging significant technologies. Finally, we select the electric vehicle (EV) domain as a case study to empirically test the validity of this framework.Results: The optimized support vector machine (SVM) model performs well in identifying highly cited EV patents. Technological frontiers in the EV domain are identified, which are related to the topics of information systems, batteries, stability control, wireless charging, and vehicle operation.Discussion: The good performance in prediction accuracy and generalization capability of the method proposed in this paper verifies its effectiveness and feasibility.

Introduction: Detecting radical innovations in the solar energy domain could offer innovation references and support the promotion of solar energy. However, relevant studies in the solar energy domain are lacking, and the related methods need to be improved.Methods: In this paper, a new framework to identify radical innovations in the solar energy domain is proposed by combining a technological convergence study and scientific relation analysis, and the link prediction method is utilized to detect potential radical innovations in this domain.Results: 1) The distributions of both the technological classes and scientific categories are uneven in the solar energy domain. The top 15 technological classes account for nearly 75.46% of all classifications. Fifteen scientific categories are cited by all the patents, and applied physics, multidisciplinary material science, energy and fuels play important roles in this domain. 2) The relationships among technological classes have evolved over time and have mainly focused on neighbouring disciplines. 3) A total of 130 patents containing new convergence relationships and/or closely related to science are identified as radical innovations. Radical innovative topics are related to the subdomains of solar photovoltaic (solar PV), heat storage, heat exchangers, and solar collectors. 4) Five potential radical innovative topics are identified. Automatic plants for producing electric energy, solar energy ecology houses, and so on are considered to have great potential in the future.Discussion: The results are consistent with the authoritative report and previous studies, which verify the viability of our methods. And the findings have important implications for scientists, policy-makers, and investors in this domain.