Abstract

Science and technology innovation policy analysis often suffers from the disconnect between unstructured policy texts and underused patent data. This paper asks: how can a unified, interpretable heterogeneous graph framework structurally align policy semantic intentions with patent collaboration networks to support policy evaluation and technology evolution analysis? Policy documents contain rich information on instruments, targets, technology priorities, and timelines, yet remain difficult to operationalize as narrative texts. Patent data record inventive activity, collaboration, and classification but are frequently reduced to scalar indicators, losing network and pathway details. To bridge this gap, we propose an integrated framework combining policy knowledge graphs, patent collaboration networks, and the Heterogeneous Graph Transformer (HGT). Using Beijing science and technology innovation policy texts, we extract policy tools, technology fields, sentence-level propositions, and temporal information to build a policy knowledge graph. Patent records are organized into a heterogeneous network comprising patents, organizations, IPC classes, and collaboration links. These two layers are linked through a policy–industry–IPC–patent schema and merged for joint representation learning. The pipeline yields structured policy data, networked patent representations, cross-domain embeddings, and preliminary temporal forecasts. The main contribution is an interpretable graph-based infrastructure that allows previously separate policy and patent materials to be analyzed jointly. The framework supports policy comparison, technology trajectory analysis, collaboration monitoring, and preliminary evaluation of delayed innovation responses, demonstrating the value of graph methods for data-driven innovation governance.

Download

Share

COinS