Attention-based combination of CNN and RNN for relation classification

Abstract

Relation classification is an essential task in natural language processing (NLP) in order to extract structured data from sentences. In this paper, we propose a novel model Att-ComNN combining convolutional neural network (CNN) and bidirectional recurrent neural network (RNN) for relation classification. By combining RNN and CNN, we obtain more accurate context representations of words, which benefits classifying relations. Besides, with both shortest dependency path (SDP) attention and pooling attention added, this model captures the most informative context representation for better classification without using other handcrafted features. The results of experiments show that our model improves the relation classification performance on the SemEval-2010 Task 8 and outperforms most of previous state-of-the-art methods, including those depending on much richer forms of handcrafted features and prior knowledge.