基于多层注意力机制的农业病虫害远程监督关系抽取研究

针对大多数现有关系抽取模型存在对语义特征提取不充分、速度慢且数据集匮乏的缺点，提出一种PCNN(piecewise convolutional neural network)模型和多层注意力机制相结合的远程监督关系抽取方法进行农业病虫害领域的关系抽取。模型由两个实体把句子分成三段，对卷积后的每一段进行最大池化获得特征，同时在实例和池化特征层面上分别引入注意力机制有效降低信息噪声。在F1评价指标上比传统方法提高了5.75%，在耗时上是传统方法的10.93%，且减少了手工标注数据集的成本。

Agricultural pest and disease relation extraction based on multi-attention mechanism and distant supervision

Aiming at the shortcomings of most existing relation extraction models: insufficient semantic feature extraction, slow speed, and lack of datasets, we propose a method based on distant supervision to do relation extraction in the agricultural field. The model combines PCNN (Piecewise Convolutional Neural Network) model and multilayer attention mechanisms. The model divides the sentence into three segments by two entities, and takes the max-pooling result as the feature for each segment after convolution. Then attention mechanism is introduced at the instance feature and the pooling feature to effectively reduce the noise of information. In the F1 score, our model is 5.75% higher than the traditional model. In time consuming, our model just takes 10.93% of the traditional model. The cost of manually labeling the dataset is reduced at the same time.