【摘要】 方面级别情感分析是对给定句子的不同方面进行情感极性预测。在长句子中有许多无关词会干扰情感预测的结果，且这些无关词与中心词存在一定的距离。对此，提出以下解决方案：设计上下文迭代学习网络。提出上下文注意力模块(context attention modules, CAM),模块采用上下文动态特征掩码(context features dynamic mask, CDM)遮掩距离中心词较远的词，上下文动态特征权重(context features dynamic weighted, CDW)减小较远词的权重。文中设计的CAM经过多层迭代，增强了方面词在上下文部分的特征提取。在公共的基准数据集上进行一系列的试验比对，试验结果证明文中提出的方法是有效的。更多还原

【Abstract】 The aspect-based sentiment analysis is the prediction of sentiment polarity from different aspects of a given sentence. There are many irrelevant words in long sentences that interfere with the sentiment prediction results, and these irrelevant words are found to be at a certain distance from the central word. A solution to the problem above is proposed: first, an iterative contextual learning network is designed. Context attention modules(Context attention modules, CAM) are proposed, which use Context Features Dynamic Mask(CDM) to mask words that are far from the central word and use CDW to reduce the weights of the more distant words. The CAM module designed in this paper enhances the feature extraction of aspectual words in the contextual part after multiple iterations. A series of experimental comparisons are conducted on common benchmark datasets, and the experimental results prove that the method proposed in this paper is effective.更多还原