基于局部上下文和GCN的方面级情感分类模型

Aspect-level sentiment classification model based on local context and GCN

对于方面级情感分析,目前的深度学习方法未能充分利用方面词的相近上下文中隐含的情感信息,基于此,提出一种基于局部上下文和门控卷积网络(gated convolutional network,GCN)的方面级情感分类模型。利用对上下文特征动态加权的方法捕捉与方面语义相关的局部上下文;采用门控卷积网络获取与方面相关的情感特征;通过多头自注意力机制捕捉句子内部的语义关联;使用Softmax识别出最终的情感极性。实验结果表明,该模型具有良好的情感分类性能,较已有的情感分类模型准确率和F_(1)值更高,能更好地掌握用户评论的情感倾向。

For aspect-based sentiment analysis,current deep learning methods fail to make full use of the sentiment information implied in the close context of aspect words.Given this problem,a new aspect-level sentiment classification model based on local context and the gated convolutional network was proposed.The dynamic weighting of context features was used to capture local context related to aspect semantics.The gated convolutional network was used to obtain the aspect-related emotional features.The semantic associations within sentences were captured by the multi-head self-attention mechanism.The final emotional polarity was identified by using Softmax.Experimental results show that this model has better classification performance,higher accuracy,and F_(1) value compared with the existing sentiment classification model,and can better grasp the emotional orientation of user comments.