基于双向LSTM的维吾尔语事件因果关系抽取

摘要

针对传统方法不能有效抽取维吾尔语事件因果关系的问题,该文提出一种基于双向LSTM(Bidirectional Long Short-Term Memory,BiLSTM)的维吾尔语事件因果关系抽取方法.通过对维吾尔语语言以及事件因果关系特点的研究,提取出10项基于事件内部结构信息的特征;同时为充分利用事件语义信息,引入词嵌入作为BiLSTM的输入,提取事件句隐含的深层语义特征并利用批样规范化(Batch Normalization,BN)算法加速BiLSTM的收敛;最后融合这两类特征作为softmax分类器的输入进而完成维吾尔语事件因果关系抽取.实验结果表明,该方法用于维吾尔语事件因果关系的抽取准确率为89.19％,召回率为83.19％,F值为86.09％,证明了该文提出的方法在维吾尔语事件因果关系抽取上的有效性.%: Since the traditional events causal relation has the disadvantages of small recognition coverage, a method for causal relation extraction of Uyghur events is presented based on Bidirectional Long Short-Term Memory (BiLSTM) model. In order to make full use of the event structure information, 10 characteristics of the Uyghur events structure information are extracted based on the study of the events causal relationship and Uyghur language features;At the same time, the word embedding is introduced as the input of BiLSTM to extract the deep semantic features of the Uyghur events and Batch Normalization (BN) algorithm is usded to accelerate the convergence of BiLSTM. Finally, concatenating these two kinds of features as the input of the softmax classifier to extract the Uyghur events causal relations. This method is used in the causal relation extraction of Uyghur events, and the results show that the precision rate, the recall rate and F value can reach 89.19 ％, 83.19％ and 86.09 ％, indicating the effectiveness and practicability of the method of causal relation extraction of Uyghur events.