基于联合向量和神经网络的事件因果关系抽取

Event Causality Extraction Based on the Joint Vector and Neural Network

事件关系抽取是自然语言处理领域中一项重要的语义处理任务。针对当前事件因果关系抽取研究中存在词汇特征不符合上下文语境、语义特征不足等问题,提出了基于联合词向量和Attention-BiGRU网络的事件因果关系抽取方法。该方法把因果关系抽取转化为分类问题,首先利用word2vec和ELMO模型对CEC语料进行表示,形成事件关系对的文本表示模型,以获取词向量矩阵及考虑文本语境的动态词特征,从而构成联合向量;然后利用Attention-BiGRU网络,深层提取语义特征信息,并对特征进行权重调整,突出重要词汇对因果关系抽取的贡献;最后将加权特征输入softmax分类器进而完成事件因果关系的抽取。在数据集上进行对比实验,该方法取得了88.56%的F1值,结果表明所提出方法的有效性。

Event relation extraction is an important semantic processing task in the field of natural language processing.Aiming at the problems of lexical features unable to meet the context and insufficient semantic features in current event causality extraction research,the new extraction method based on the joint word vector and Attention-BiGRU network was proposed,which transformed the problem of causality extraction into a classification problem.Firstly,the data was vectorized by using the word2vec and ELMO models to form a text representation model of event relationship pair to obtain the corresponding word vector matrix and dynamic word features generated from the consideration of the context,so as to form a joint vector.Then the semantic features information was deeply extracted with the Attention-BiGRU network and weighted to show the contributions of some important words to the causal relationship extraction.Finally,the weighted features were input into softmax classifier to complete the event causality extraction.The comparison experiments on data sets show that the proposed method achieves 88.56%F1 value indicating the effectiveness of the proposed method.