基于上下文感知的方面类别情感分类

由于一个评论往往会涉及多种方面类别及情感倾向,而传统注意力机制难以区分方面词和情感词的对应关系,从而影响评论同时存在多种方面类别时的情感极性分析.为了解决上述问题,提出了一种基于上下文感知的方面类别情感分类模型(MA-DSA).该模型通过重构方面向量捕获句子中更多样且有效的语义特征,并将其融入上下文向量,然后将上下文向量通过DiSA模块进一步捕捉句子内部情感特征,确定方面词与情感词的关系,进而对指定方面类别进行情感分类.在SemEval的三个数据集上的实验结果表明,MA-DSA模型在Restaurant-2014数据集上的三个指标值均优于基准模型,证明了该模型的有效性.

Context-aware learning for aspect category sentiment classification

A review often involves multiple categories and their emotional tendencies, but the traditional attention mechanism is difficult to distinguish the correspondence between aspect words and emotion words, which affects the analysis of emotional polarity when there are multiple aspect categories in a review. In order to solve the above problems, this paper proposed an context aware learning for aspect category sentiment classification model(MA-DSA). The model captured more diverse and effective semantic features in the sentence by reconstructing the aspect vector, and integrated it into the context vector. Then, it used the context vector to further capture the internal emotional characteristics of the sentence through the DiSA module to determine the aspect and emotion words. Then the sentiment classification was performed on the specified aspect category. The experimental results on SemEval''s three datasets show that the three index values of the model on the Restaurant-2014 dataset are better than the baseline model, which proves the effectiveness of the MA-DSA model.