句法特征融合增强的方面级情感分析

Aspect Level Sentiment Analysis of Syntactic Feature Fusion Enhancement

方面级情感分析是情感分析领域的一项关键任务,其目的是分析目标文本中各个方面的情感极性.句法依存树曾被广泛用于方面级情感分析任务中,目前的特征提取和交互方式仅限单一特征级,未能充分利用句法依存树上的有效信息.针对该问题,提出了一种基于句法依存树的多级特征提取算法来建立方面级情感分析模型.首先利用深度优先搜索得出句子的浅层特征表示,然后通过划分子图改进传统图卷积神经网络的建模方式来提取句子的深层特征表示,最终融合多级特征的句子表示并进行情感分类.在4个开放数据集上分类准确率都取得1.64%~2.12%的提升,F1值取得2.24%~4.97%的提升.实验结果表明基于该方法建模能获取更充分的多层句法特征信息、有效提高分类效果.

Aspect level sentiment analysis is a key task in the field of sentiment analysis,whose purpose is to analyze the sentiment polarity of all aspects in the target text.Syntactic dependency tree(SDT)has been widely used in aspect level sentiment analysis tasks.In the past.feature extraction and interaction methods were limited to a single feature level,which failed to make full use of the effective information on the SDT.To solve this problem,a multi-level feature extraction algorithm based on syntactic dependency tree is proposed to build aspect level sentiment analysis model.Firstly,the shallow feature representation of sentences is obtained by depth-first search,then the traditional graph convolution neural network modeling method is improved by dividing subgraphs to extract the deep feature representation of sentences,and finally the sentence representation of multi-level features is fused and emotion classification is carried out.On the four open datasets,the classification accuracy is improved by 1.64%to 2.12%,and the Fl score is improved by 2.24%to 4.97%.The experimental results show that this method can obtain more sufficient multi-level syntactic feature information and improve the accuracy of classification.