A Dual-Encoding Dynamic Patent Quality Assessment Network With an Application to Green Technology Patents

Patent quality serves as a critical indicator of technological innovation and intellectual value. While existing studies predominantly frame patent quality assessment as a static classification task, this work reconceptualizes it as a dynamic multidimensional process evolving under technological and market uncertainties. We propose the Dual-channel Dynamic Patent Quality Assessment Network (DPQAN), which synergistically models intrinsic patent attributes and extrinsic influence trajectories through two novel encoders: (1) cross-sectional time-series encoding capturing annual multidimensional quality state and (2) dimensionwise evolutionary sequence encoding tracking longitudinal indicator patterns. Additionally, we investigate a dual-stage attention mechanism for performance analysis. Experiments on green technology patents demonstrate DPQAN’s superiority over baseline methods in multiyear forecasting while maintaining computational efficiency. This work bridges the gap between dynamic IP valuation and AI-driven innovation policy, offering tangible tools for sustainable R&D allocation.