基于Bi-LSTM和CRF的药品不良反应抽取模型构建

Construction of an Adverse Drug Reaction Extraction Model Based on Bi-LSTM and CRF

【目的】提出处理社会媒体上不规范文本的方法,以提高社会媒体中药品不良反应抽取效果。【方法】基于TensorFlow深度学习框架,将LSTM和CRF结合,借助LSTM能利用上下文信息的特性,通过CRF层考虑输出标记的依赖关系,构建基于Bi-LSTM和CRF的药品不良反应抽取模型。【结果】在Twitter数据集上开展系列实验,结果表明与传统CRF、前向LSTM、后向LSTM和Bi-LSTM相比,Bi-LSTM-CRF可以取得最高的F-measure值(0.7963)。【局限】实验数据源相对单一,未来可以在多个数据源上验证Bi-LSTM-CRF方法的有效性。【结论】结合LSTM和CRF能够有效处理社会媒体上不规范文本,所构建的模型可较好识别不良反应实体,辅助相关部门进行决策。

[Objective] To improve the performance of extracting adverse drug reactions from social media,a method is proposed to deal with non-standard texts in social media.[Methods] This method Bi-LSTM-CRF combined LSTM and CRF,and was implemented using TensorFlow.LSTM Could utilize context information,while CRF Could consider the dependence of output tags.An adverse drug reaction extraction model was constructed based on Bi-LSTM-CRF.[Results] A series of experiments were carried out on the Twitter dataset.The experimental results showed that the proposed Bi-LSTM-CRF method achieved the highest F-measure(0.7963)for adverse drug reaction extraction,compared with other methods,including CRF,forward LSTM,backward LSTM,and Bi-LSTM.[Limitations] The experiments were performed on only one corpus,and the validity of the proposed method need be verified on other data sources.[Conclusions] Combining Bi-LSTM and CRF can effectively deal with non-standard texts in social media.The constructed model in this paper can identify adverse drug reactions effectively and support relevant departments in decision-making.