Entity Position-aware RCNN for Relation Classification
Author:Qifeng Xiao, Shaochun Wu Author Unit: Department of Intelligent Information Processing, Shanghai University, Shanghai 200044,China.
Abstract:Relation Classification is a foundational task with regards to many other natural language processing(NLP)
tasks, which has caught many attentions in recent years. The difference with text classification is that relation
classification needs to give attention to entity position. In this paper, we propose an entity position-aware
RCNN architecture for this task. Our model takes full advantage of position feature and position indicator,
which makes model focus on the marked entities. In addition, we utilize the RCNN model to Relation
Classification, which takes sequential context information and word information into full consideration. The
experimental results on SemEval-2010 Task 8 benchmark dataset show that RCNN based method achieves
better performances than traditional method based on CNN and RNN, and introducing more features could
further improve the performance.
Keywords:Relation classification; RCNN; Position information
Download information [SIZE:659.18 KB Amount of downloads: second]