Policy2Graph: A Structure-aware Perspective on Constructing Policy Knowledge Graph for Smart Governance

The social governance empowered by AI are energetically promoted in many countries with one emerging technology, namely policy knowledge graph. Normally, a classical knowledge graph that encodes the data into Resource Description Framework (RDF) triples could fit most cases. However, a large number of imperative sentences in policy discourses make it difficult to extract the key elements: entities. Thus, using words such as entities and concepts as nodes are impractical in this work. In this paper, we shed light on a high-level abstraction of the policy discourse that can explicitly reflect rich semantic and structural information (Policy2Graph). Specifically, the utterance-level semantic nodes and syntactic dependency methods are employed to represent the detailed contents of the policies. In addition, the implicit outline of policy is generalized to connect the semantic nodes. In this way, each policy discourse is constructed into a small-scale knowledge graph. Furthermore, to fusion them into a large-scale knowledge graph, similar nodes in different policy knowledge graphs could be associated through the semantic similarity, considering the weight of each feature in the dependency relationship of syntactic dependency policy trees. We have demonstrated Policy2Graph on the rural innovation and entrepreneurship policies released by the Chinese government over the past eight years. Results have validated the effectiveness of our approach.