Using Graph Representations for Semantic Information Extraction from Chinese Patents

Abstract

This paper proposes a graph representation approach to automatically extract semantic information from Chinese patents, which can be used to provide semantic support for text-content based patent intelligent analysis. Two graph models are devised using graph representations, i.e., a keyword based text graph model and a dependency tree based text graph model. The first graph model is constructed by computing the similarities between two keywords, while the second graph model is constructed by extracting syntactic relations from text sentences. In the case study a frequent subgraph mining algorithm is utilized to discover frequent subgraph patterns based on the above two models, and such patterns were further used as features to build text classifiers for the purpose of testing the expressivity and effectiveness of the proposed graph models. The experimental results proves the validation of the proposed graph representation methods.