Inter-personal Relation Extraction Model based on Dependency Parsing and Bidirectional Gating Recurrent Unit

Abstract

Relationship extraction is a fundamental component of various information extraction systems. Traditional relationship extraction methods are mainly rule-based methods and machine learning methods. Rule-based methods require induction and analysis of the corpus, followed by extraction of relationship extraction rules and finally pattern matching. The machine learning approach requires a large amount of manually annotated train data and manual extraction of features. However, these methods require a lot of statics and higher time costs. Considering these issues in the traditional relationship extraction methods and the linguistic characteristics of Chinese text, this paper proposes a new deep neural network structure. Firstly, the dependency relationships between sentence components are analyzed by using dependency parsing, which reveals the syntactic structure of the sentence and enhance the potential semantic information. Secondly, the important semantic information in the sentences is captured by using the sentence-level attention mechanism. Finally, the Bidirectional Gating Recurrent Unit model is used to simultaneously capture the contextual information of the text, and to improve the performance of relation extraction. The experimental results show that the model proposed in this paper is more effective than existing methods.