基于关系特征交互的方面级情感分类方法

An Aspect-level Sentiment Classification Method Based on Relational Feature Interaction

方面级情感分类是一项细粒度的情感分析任务,其目的是识别一句话中的方面词、观点项及其对应的情感极性。现有的方面级情感分类方法对模型的构建存在不足,难以有效利用句子中的依存关系信息,从而导致分类准确率较低。基于此,该文提出一种基于关系交互的图注意力网络模型。该模型首先利用单词之间的依存关系构建句法依存树,并使用双向门控循环单元提取句子上下文特征,然后将两者融入图注意力网络和关系感知网络中进行关系交互,以学习句子间的句法和语义信息,最后将关系的表征结果结合并输出方面词的情感类别(正面、负面、中性)。在四个公开数据集上的实验结果表明,该模型在方面级情感分类任务上充分挖掘并利用了文本的句法关系信息,进一步提升了情感分类的准确率。

Aspect-level sentiment classification is a fine-grained sentiment analysis task, which aims to identify aspect terms, opinion items and their corresponding sentiment polarity in a sentence. Existing aspect-level sentiment classification methods are inadequate for model construction, and it is difficult to effectively utilize the information of dependency relations in sentences, which leads to low classification accuracy. Based on this, we propose a relational interaction graph attention network(RIGAT). Firstly, the model builds a syntactic dependency tree based on the dependencies between words, and uses bi-directional gated recurrent unit(Bi-GRU) to extract sentence context features. Secondly, we integrate the two into the graph attention network and the relational-aware network for relational interaction to learn the syntactic and semantic information between sentence. Finally, we combine the representation results of the relationship and output the sentiment polarity(positive, negative, or neutral) of the aspect word. Experimental results on four public datasets show that the model fully excavates and utilizes the syntactic relationship information of the text in aspect-level sentiment classification tasks, and further improves the accuracy of sentiment classification.