Chinese Agricultural Entity Relation Extraction via Deep Learning

Kai Zhang, Chuan Xia, Guoliang Liu, Wenyu Wang, Yunzhi Wu, Youhua Zhang, Yi Yue

Abstract: With the advent of Deep Learning (DL), Natural Language Processing (NLP) has progressed at a high speed in the past few decades. Some DL models have been established for Relation Extraction and outperform than the traditional Machine Learning (ML) methods. In this paper, we built four DL models: Piecewise Convolutional Neural Network (PCNN), Convolutional Neural Network (CNN), Recurrent Neural network (RNN) and Bidirectional Recurrent Neural Network (Bi-RNN) to tackle the task. Using PCNN, we outperform than other three models, and achieve Area under Curve (AUC) of 0.154 with just 24 epochs. And we use some selector mechanism to improve the model. Our experimental results show that: (1) Attention mechanism got the best compatibility with all models, but in some case, max pooling may perform better than it. (2) Using only word embeddings, the performance of the model will discount a lot.