摘要
情感分类任务是自然语言处理中的下游任务,具有重要的研究价值。情感分类任务的性能非常依赖于语句中的上下文信息,但现有的基于深度卷积神经网络的方法无法较好地捕捉句子上下文依赖,也无法对特征内部的相互依赖性进行建模。针对该问题,提出了一种基于Bi-LSTM-Attention网络的英文文本情感分类模型。该算法同时使用Bi-LSTM和自注意力机制来对句子的上下文情感依赖关系进行建模。具体地,模型主要由词嵌入层、Bi-LSTM层、自注意力层以及分类器层组成。实验结果充分表明,相比于基准模型,所提出的方法在MR数据集和SST-2数据集上取得了较好的性能,其准确率分别达到了82.4%与88.3%,有效提升了分类模型的性能。
The sentiment classification is a downstream task in natural language processing,and has important research value. The performance of sentiment classification depends on the context information in the sentence,while the existing work based on deep convolutional neural networks cannot capture sentence contextual dependencies better and cannot model the interdependencies within features. To address this problem,an English text sentiment classification model based on Bi-LSTM-Attention network is proposed. The proposed model utilizes Bi-LSTM and self-attention mechanism to explicitly model the context dependence of sentences. Specifically,the model mainly consists of a word embedding layer,a Bi-LSTM layer,a self-attention layer,and a classifier layer. Extensive experimental results show that compared with the benchmark model,the proposed method achieves the best performance on MR and SST-2 datasets,and the accuracy reaches 82.4% and 88.3% respectively,which effectively improves the performance of the classification model.
作者
朱亚辉
ZHU Yahui(School of Foreign Languages,Changsha Normal University,Changsha 410100,China)
出处
《电子设计工程》
2022年第16期27-30,共4页
Electronic Design Engineering
基金
湖南省社科基金(17YBA034)。
