情感子句预测与原因子句提取方法

Emotion clause prediction and cause clause extraction method

为减少人工成本,提出在未给定情感标签情况下预测文本情感子句,同时提取原因子句的方法。使用CNN提取局部语义信息,使用带有注意力的Bi-LSTM提取句子上下文语义信息以及情感表达的关键部分信息,将这3类信息结合获取更好的句子特征来进行情感预测;通过注意力将预测的情感标签与句子特征结合,提取原因。实验结果表明,模型在情感子句预测和原因子句提取任务中均取得目前最好结果,在未给定文本情感标签的情况下,原因提取效果仍优于大部分传统模型。

To reduce labor costs,a method was proposed to predict the emotional clause of text and extract the cause clause at the same time(EPCEM).CNN was used to extract the local semantic information,the Bi-LSTM with attention was used to extract the sentence context semantic information and the key part of emotion expression information,and these three kinds of information were combined to obtain better sentence features for emotion prediction.The predicted emotion tag was combined with sentence feature to extract the cause.Experimental results show that the model achieves the best results in the task of emotion clause prediction and cause clause extraction,and the effect of cause extraction is still better than that of most of the traditional models without given emotional label of the text.