摘要
针对中文酒店评论情感分析会忽视特定方面情感极性的问题,提出一种复合双通道模型GCGAT,进行方面类别情感分析。该模型基于门控机制和双向门控循环单元,由四层网络结构组成;利用门控机制控制通道输出;引入方面类别影响因素词向量和注意力机制提高模型信息提取准确性;数据集方面,利用LDA-Jaccard模型进行聚类,确定方面类别并进行标注,建立了细粒度数据集。通过对比实验发现,GCGAT模型取得了很好的效果,在准确率、F1值、ROC曲线和auc值上效果均优于其他模型。
Aiming at the problem that the sentiment analysis of Chinese hotel reviews will ignore the polarity of specific aspectual sentiment, a composite two-channel model, GCGAT, is proposed for aspectual category sentiment analysis. The model is based on the gating mechanism and bidirectional gating loop unit, and consists of a four-layer network structure;the gating mechanism is utilized to control the channel output;the aspect category influencing factor word vector and the attention mecha-nism are introduced to improve the accuracy of the model’s information extraction;in terms of the dataset, clustering is carried out by using the LDA-Jaccard model to determine the aspect categories and annotate them, and a fine-grained dataset is established. Through comparison experiments, it was found that the GCGAT model achieved good results, and the effect was better than other models in terms of accuracy, F1 value, ROC curve and auc value.
出处
《计算机科学与应用》
2023年第11期2000-2011,共12页
Computer Science and Application
