Automatic Construction of a Domain-specific Knowledge Graph for Chinese Patent Based on Information Extraction

Knowledge graph has been proved as an effective tool in diverse domains including patent service. However, despite partial structure of data, most patent information lies in unstructured data like abstract text, from which it is relatively difficult to build a knowledge graph, since traditional methods rely on predefined human-annotated resources of entities and their relationships. Furthermore, the difficulty in Chinese language processing worsens the problem. This paper proposed an unsupervised method to automatically construct a Chinese patent knowledge graph without pre-built dataset. This research first builds a basic graph frame using structured patent data. Then the proposed method mainly utilizes keyphrase extraction algorithm to find patent properties and semantic role labeling method to dig deeper relations. The experimental result proves the effectiveness of the proposed method.