Relation extraction via deep-fusion convolution neural network
Xiong Zhang, Fucai Chen, Ruiyang Huang · 2017
Aiming at the problem that the traditional single neural network method is limited in feature dimension extraction, a new deep-fusion convolutional neural network is proposed. It uses two kinds of different representations (i.e., word vector and shortest dependency path) as different inputs of convolutional neural network, therefore, it is capable to learn more dimension text features automatically and fuse them deeply in high dimensional feature space and then it improves the accuracy of relation extraction. Experiments are conducted on the SemEval-2010 Task 8 data sets, the results show that the proposed relation extraction method based on deep-fusion convolutional neural network can effectively combine the traditional text features and high dimensional features of neural network learning, and improves the effectiveness of relation extraction.