Learning Representation For Relation Extration With Knowledge Graph

The development of knowledge search relies on the construction and improvement of large-scale knowledge graph, such as Freebase and Wikipedia. Nevertheless, they were wrote and improved cooperatively by people across the world. In the Big Data era, this approach has become inefficient. So it is very essential to focus on the information extraction technology that can extract information automatically.In this paper, we studied relation extraction, the main technique of information extraction. We proposed the method of learning representation for relation extraction with knowledge graph. The main work and contribution are listed below:1. Combined with weak labeled text corpus and ranking relation extraction framework, it's convenient to learn representation for text and introduce knowledge graph embedding model into our relation extraction system.2. Propose the methods of learning mention representation based on CNN and RNN with better performance than traditional method.3. Introduce knowledge graph embedding model into text-based relation extraction model. This improved the system performance.4. Design and realize all of the relation extraction models mentioned above and perform experiments to test these models' performance.In conclusion, in this paper, we used the method of representation learning without feature design and extraction to get better generalization and improve the relation extraction performance.